- Feb 1, 2006
- 2,034
- 1
- 19,780
Last Update: 4-14-06
4-14-06 Changes: Removed some content users found to be "Biased". Added preliminary information on "Clovertown", and added more to AMD's Future Products.
*Refer to the very bottom of this post for past changes*
Main Idea!! DO NOT Make this a Fanboy Thread or Flame Thread, Thank you.
I wrote this originally as a way to describe the difference between an AMD and Intel CPU, but over time, it changed to adding in technologies related to the CPU's and eventually, an overall computer hardware explanation. This has been geared towards newbies to enthusiasts and everybody in between. This has been kept as Technically-Negated as possible, so everybody can understand it. If you find something wrong with this, or wish to have something added that is lacking, PM me or drop a post here and I will be happy to accomodate.
Table of Contents
I. Intel Processors - Pentium 4
II. Intel Processors - Pentium 4: Recommendations
III. Intel Processors - Pentium D
IV. Intel Processors - Pentium D: Recommendations
V. Intel Processors - Xeon
VI. Intel Processors - Conroe
VII. Intel Processors - Conroe Speeds
VIII. AMD Processors - Athlon 64
IX. AMD Processors - Athlon 64: Recommendations
X. AMD Processors - Athlon 64 X2
XI. AMD Processors - Athlon 64 X2: Reommendations
XII. AMD Processors - Socket AM2
XIII. AMD Processors - Socket AM2 Speeds
XIV. AMD Processors - Opteron 64
XV. AMD Processors - Opteron 64 Speeds
XVI: AMD Processors - Opteron 64: Socket 1207 "F"
XVII: Technical Information (FSB, HyperTransport, etc.)
XVIII: Memory (Timings, Bandwidth, etc.)
XIX: Cache Comparison (Level 2 Performance)
XX: Future Technologies
XXI: What's That? (Terms)
XXII: Motherboards
XXIII: Naming Schemes: What They Mean
XXIV: Past Changes
Intel Processors
I. Pentium 4 Processors [Sockets: 423 478 775]:
The main thing to point out is that the Pentium 4 uses the traditional "Front Side Bus". This is the "data pipeline" in which information moves from the CPU to the Motherboard. This is actually composed of 3 bus's in of itself, a "Data", "Command", and "Address". Maybe you heard of the 800MHz Front Side Bus before, the truth is is that only the "Data" bus runs at 800MHz, the "Command" & "Address" run at 200MHz (I will explain system clock speeds and Front Side Bus speeds at the end).
On a Pentium 4, the Memory (RAM) is accessed via the Memory Controller that lies within the Northbridge, which means that the CPU must first access the Northbridge which in turn accesses the Memory. After retrieving the data from the Memory, it than goes back through the Northbridge, then to the Front Side Bus back into the CPU for processing. This is actually a con on the Pentium 4, because the Front Side Bus is a Half Duplex, Bidirectional Bus. This means that the CPU cannot send and receive information on the Front Side Bus (HyperTransport in Athlon 64;s) at the same time.
With the Prescott core on some Pentium 4's, another concern arises: Heat. The Prescott core runs very hot compared to the Northwood or Athlon 64 cores, which means that if you plan to purchase an Intel Pentium 4 system with Prescott core, make sure to purchase an Aftermarket Cooler with it. The Prescott though, can be overclocked very high, so if you're an Overclocking Enthusiast or want to get into overclocking, the Prescott Pentium 4 will not let you down, just make sure to get a very good Aftermarket Heatsink & Fan to cool it.
The Prescott Pentium 4 also has a smaller Level 1 Cache than the Athlon 64, avg. of 48KB Level 1 vs. 128KB Level 1 in the Athlon 64. The Pentium 4 Prescott has larger Level 2 Cache, 2MB vs. 512KB or 1MB in the Athlon 64. The larger L2 Cache in the Pentium 4 helps leverage the bottlenecked Front Side Bus, but I personally believe that larger Level 2 Cache does not mean greater performance, and in some cases, I believe it to decrease performance (Refer to "Cache Comparison" at end).
A new core was recently released on 65nm, known as "Cedar Mill". The major changes from Prescott is the Thermal Design Power (TDP) has been lowered to 86 Watts. On the 6x2 (e.g., P4 632) versions, there is also something knew, this is known as Virtualization Technology. This Vritualization Technology allows you to run multiple operatings systems w/o the cost of emulation. This technology is being implemented in 9xx Series Dual Core Pentium 4's and Xeon 7xxx Series, as well as Core Duo processors.
One renowned thing about the Pentium 4, is HyperThreading. HyperThreading fakes the Operating System into thinking that there is 2 processors inside the chip, kinda like Dual-Core but there really isn't a 2nd processor. This works by (Not getting too technical) allowing the CPU to operate on 2 threads vs. 1 on a Non-HyperThreaded CPU. One thing to note though, if I recall correctly, is that HyperThreading is Asynchronous, meaning the CPU cannot process both threads at the same time. This does help in Multi-Tasking as it allows the CPU to start processing another thread w/o wasting CPU clock cycles sittng idle waiting for a program to end. The real-word performance benefit from HyperThreading is 5-15%, with some scenarios resulting in decreased performance, especially in Server Benchmarks.
The first Pentium 4's ran on Socket 423, but due to an inadequate electrical design, a Socket 423 P4's max speed was 2.0GHz, this was replaced with Socket 478, this uses Dual Channel DDR1 Memory and allows for much higher clock speeds. The newer generation of Pentium 4's is on Socket 775 (Socket T) and uses Dual Channel DDR2 Memory. If you buy a Socket 775, make sure you purchase the right Memory, all the 6xx Series processors are s775 and use DDR2, as well as the 5xx s775, 8xx, and 9xx Series CPU's.
Another thing to note is whether or not to get a 64-bit processor. I strongly recommend getting a 64-bit CPU, so if you want that, you must get one of the 6xx Series Pentium 4's (Or one of the newer 5xx Series, but I recommend a 6xx Series). I have changed the P4 Single Core to reflect Cedar Mill, Cedar Mill Pentium 4's are 631 through 661, there is no 671 (3.8Ghz) Cedar Mill. The best Pentium 4 offerings are:
II. Pentium 4: Recommendations
Pentium 4 631 (3.0GHz)
Pentium 4 641 (3.2GHz) <--Recommended 6xx Series Processor
Pentium 4 651 (3.4GHz)
Pentium 4 661 (3.6GHz)
Pentium 4 670 (3.8GHz)
III. Pentium D 8xx & 9xx Processors [Sockets: 775]:
The Intel Pentium D 8xx & 9xx CPU's are Dual-Core, meaning there is 2 processors on 1 chip. The 2 processors in the 8xx and 9xx series communicate to each other over the standard Front Side Bus, thus again creating another bottleneck and latency issue as stated in the Pentium 4. The 8xx and 9xx Pentium D's also run on 800MHz Front Side Bus as stated, with the exception of the new Intel Pentium D 9xx Extreme Editions which have a 1066MHz Front Side Bus.
On the Intel Pentium D 8xx Series, the standard models (820 to 840) do not have HyperThreading, only the Extreme Edition has HyperThreading turned on, which makes it seem like you have 4 Processors (HyperThreading 2 x 2 CPU's). The Same goes for the Intel Pentium D 9xx Series, the 920 to 950 have 800MHz Front Side Bus w/o HyperThreading, the 955 Extreme has 1066MHz Front Side Bus and HyperThreading turned on. Another thing to note is the 8xx Series is 90nm process, whereas the 9xx Series is 65nm. 65nm will aid in overclocking and heat issues, thus making the 9xx Series superior to the 8xx Series. I would not recommend a 8xx Series, better to get the 9xx Series.
The 9xx Series also has 4MB Level 2 Cache (2MB Per Core) vs. the 8xx Series which has 2MB Level 2 Cache (1MB Per Core). The Pentium D's, both 8xx and 9xx, run on Socket 775. So remember what I said bout Socket 775 and memory in the Pentium 4 posting. I have added the Pentium D 805 as a possible buy, due to it's extremely low price and ability to be overclock very high. The best Pentium D offerings:
IV. Pentium D: Recommendations
Pentium D 805 (2.66GHz) <=Recommended For Budget/Overclockers
Pentium D 920 (2.8GHz)
Pentium D 930 (3.0GHz) <--Recommended 9xx Series Processor
Pentium D 940 (3.2GHz)
Pentium D 950 (3.4GHz)
Pentium D 955 (3.46GHz Extreme Edition/1066MHz FSB)
Pentium D 965 (3.73GHz Extreme Edition/1066MHz FSB)
V. Xeon Processors [Sockets: Slot 2 603 604]:
The Intel Xeon CPU was first released in 1998 named "Intel Pentium II Xeon" named Drake. It was a replacement for the Pentium Pro and had 512KB, 1MB or 2MB Cache w/ 100MHz FSB and was a Slot 2 processor. It also came with Dual, Quad, and Octal CPU configurations w/ additional logic and a newer chipset. In 1999, the Pentium III Xeon was released, this was almost identical to the Pentium II Xeon but it had a few differences, such as SSE and a cache controller enhancement. There was a 2nd version released, named Cascades, that was often suffixed with Cascades 2MB, to differentiate it from the Tanner. The Cascades had 1MB or 2MB Cache vs. Tanner which had 256KB.
The Tanner had a 133MHz FSB whereas the Cascades 2MB had a 100MHz FSB. After that, in Mid-2001, the standard Xeon (lacking the "Pentium" aftermark) was introduced. This was the first Xeon CPU to use the "Netburst" architecture, this CPU was named "Foster". This chip, however, was outperformed by Cascades 2MB and Athlon MP's. Foster also required Rambus Memory, making it quite expensive. And Foster only allowed up to 2 CPU's, so a 2nd version, named Foster MP, was released w/ 1MB of Level 3 Cache. In 2002, Prestonia, a 130nm Xeon was released. This was the first time "HyperThreading" was used with Xeon's, these also had 512KB of Level 2 Cache.
These Xeon CPU's had a 533MHz FSB and supported Dual Channel memory. The Xeon Prestonia would bring 1MB or 2MB of Level 3 Cache. The Xeon MP was the Gallatin, this brought 4MB of Level 3 Cache. Even later, a 64-bit version of the Xeon, named Nocona, was released. This also brought support for PCI-Express, DDR2, and SATA. After that, Irwindale was released, a 64-bit Xeon w/ Double the Level 2 Cache and the ability to reduce clock speeds in times of low processing. In April 2005, 64-bit Xeon MP's were released, named Cranford and Potomac. Cranford had 4MB Level 2 Cache and Potomac had 8MB Cache.
Xeon speeds range from 400MHz in Tanner to 3.667GHz in Xeon MP's. A Dual-Core Xeon was released on October 10, 2005. This runs at 2.8GHz w/ an 800MHz FSB. The architecture for these is the same, 2 CPU's share 1 front side bus, which gives each CPU 1600MB/s Bandwidth using DDR400 (4-Way Xeon's have 800MB/s Bandwidth per CPU). These had 2MB of Cache per core (4MB Total). A Xeon MP version was released, one with 1MB of Cache per core and one with 2MB Cache per core. Both Xeon Dual-Core CPU's are named Paxville.
The Next Generation of Xeon's is known as "Dempsey". These CPU's sport a 266MHz FSB Quad Pumped or a 333MHz FSB Quad Pumped. They also implement something known as Dual Independent Bus's, which gives each Core its own FSB to the Northbridge. Early benchmarks compared to Current Opteron 64 CPU's put Dempsey ahead in the race in some benchmarks. Preliminary benchmarks put "Clovertown", the Quad-Core version of "Dempsey", on par with a 2.0GHz AMD Opteron 64 in Single-Threaded environments Clock-For-Clock. "Clovertown" also has HyperThreading, which gives it the ability to work on 8 threads.
VI. Conroe Processors [Sockets: 775]:
The Next Generation for Desktop CPU's from Intel is the "Conroe". "Conroe" will sport DDR2, and FSB's ranging from 800MHz, 1066MHz, and 1333MHz for the Extreme Edition. Clock Speeds from 2.4GHz up to 3.33GHz in the Extreme Edition. The main difference between Conroe and past Pentium 4's, is Conroe has a 14-Stage Pipeline, SSE4 Instructions (Maybe), and a 4-Issue Core vs. 3-Issue in past CPU's. Conroe is all 65nm and 45nm in the future. The Unified L2 Cache of Conroe can cause Cache Thrashing, and this could only be alleviated by CPU Overhead controlling it, which could mean alot of extra clock cycles. Cache Thrashing could prove to be a problem for Server Environments, if it proves true, it could spell bad news for Intel's Server Market.
Early benchmarks of Conroe either put it on par with AMD's K8's, or a little ahead, depending on the test.
Current Conroe offerings (Not yet released):
VII. Conroe Speeds
Core Duo E8?00 - 3.33 GHz (4MB L2 Cache) 1,333MHz FSB
Core Duo E6800 - 2.93 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6700 - 2.67 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6600 - 2.40 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6400 - 2.13 GHz (2MB L2 Cache) 1,066MHz FSB
Core Duo E6300 - 1.86 GHz (2MB L2 Cache) 800MHz FSB
Core Duo E4200 - 1.66 GHz (2MB L2 Cache) 800MHz FSB (Single-Core)
AMD Processors
VIII. Athlon 64 Processors [Sockets: 754 939]:
The AMD Athlon 64 processors come in 2 packages, Socket 754 and Socket 939. The main difference is Socket 939 utilizes Dual Channel memory, where the Socket 754 uses Single Channel memory. The fastest Socket 754 Athlon 64 CPU you can get, is the AMD Athlon 64 3700+ 2.4GHz Clawhammer. The Fastest Athlon 64 Socket 939 is the Athlon 64 4000+ San Diego 2.4GHz.
The Athlon 64 is different from the Pentium 4 in a few ways, mainly is the Onboard Memory Controller and the use of HyperTransport instead of the traditional Front Side Bus. The Onboard Memory Controller means that the CPU can access the memory directly w/o having to go through the Northbridge to retrieve information from memory. This reduces latencies and bottlenecks and provides more efficient throughput than the Pentium 4 processor.
The use of HyperTransport is a big step for AMD, this takes the 3 bus's in the Front Side Bus (Data, Address, & Command) and groups them into 32-bit packets. Think of it like a highway, instead of having 3 people run down the road as fast as they can, they all 3 hop into a car and drive down 1 busline (16-bits wide). Also one thing to note is the speed difference, Hypertransport allows speeds in current Athlon 64's of either 800MHz or 1000MHz (1GHz) Speeds (Socket 754 uses 800MHz while Socket 939 uses 1000MHz). Also, HyperTransport is a Full Duplex, UniDirectional architecture, and each link consists of 2 16-bit lanes for the packets to travel on. This also means that information can be sent and received from the CPU to Northbridge at the same time, each at 800MHz or 1000MHz (1GHz).
It is important to know that Athlon 64 CPU's do not have a Front Side Bus, rather they use HyperTransport which very much acts like the Front Side Bus but it is a replacement to reduce latencies and bottlenecks. Athlon 64's have either 512KB of Level 2 Cache or 1MB of Level 2 Cache, as stated in the Pentium 4 part. All Athlon 64 and above CPU's have 128KB of Level 1 Cache.
The Socket 754 Athlon 64 CPU's are mostly made on 130nm, vs. 90nm for newer Socket 939 processors. There is 1 90nm Socket 754, it is the Athlon 64 3000+ Venice, and there are alot of 130nm Socket 939 processors. I recommend getting a Socket 939 90nm Athlon 64 CPU for best performance and overclocking potential. If you plan to overclock, one thing to note is that you must adjust the HyperTransport speed. HyperTransport becomes unstable after 1000MHz operation, so make sure to set it's multiplier to something lower if you overclock so as not to disrupt the stability of the system.
I only recommend getting an Athlon 64 Socket 939 Processor, mainly because Socket 939 provides Dual Channel Memory Access, Faster Processors, and newer SLI & CrossFire Motherboards. Because of this, I have listed some Socket 939 Athlon 64 processors (All are 90nm). The current Socket 939 Athlon 64 offerings:
IX. Athlon 64: Recommendations
(CPU Speed/Core/L2 Cache)
Athlon 64 3000+ (1.8GHz/Venice/512KB)
Athlon 64 3200+ (2.0GHz/Venice/512KB)<--Recommended 939 A64 CPU
Athlon 64 3500+ (2.2GHz/Venice/512KB)
Athlon 64 3700+ (2.2GHz/SanDiego/1MB)
Athlon 64 3800+ (2.4GHz/Venice/512KB)
Athlon 64 4000+ (2.4GHz/SanDiego/1MB)
X. Athlon 64 X2 Processors [Sockets: 939]:
The Athlon 64 X2 Processor is the Dual-Core Athlon 64, meaning it has 2 processors on 1 chip. The difference between this Dual-Core and the Pentium D Dual-Core, is mainly the way he 2 CPU's communicate. On the Athlon 64 X2 Processor, the 2 CPU's communicate via a "Crossbar", which is an area in the CPU that the 2 CPU's can communicate at CPU frequency, rather than using the Front Side Bus or HyperTransport to communicate, this provides better "Coherency", which is communication between multiple CPU's.
All Athlon 64 X2 processors are on Socket 939, with newer ones being released for Socket AM2 when it is released. I don't shun away from any X2 CPU, so I state them all (All Athlon 64 X2's with 1MB L2 Cache per core run on Toledo Core, X2's with 512KB per core run on Manchester Core except the new 3800+ that runs on Toledo as well but with 512KB L2 Cache). Current Athlon X2 offerings (I included the FX-60 as it runs on an X2 core):
XI. Athlon 64 X2: Recommendations
(Speed/L2Cache Per Core)
Athlon 64 X2 3800+ (2.0GHz/512KB)
Athlon 64 X2 3800+ (2.0GHz/512KB/Toledo Core)
Athlon 64 X2 4200+ (2.2GHz/512KB)
Athlon 64 X2 4400+ (2.2GHz/1MB) <--Recommended 939 A64 X2 Processor
Athlon 64 X2 4600+ (2.4GHz/512KB)
Athlon 64 X2 4800+ (2.4GHz/1MB)
Athlon 64 FX-60 (2.6GHz/1MB/Toledo Core)
XII. AM2 Processors [Sockets: 940]:
The next Athlon 64 CPU's will be on Socket AM2, which as 940 pins. One thing to note is that AM2 is not pin compatible w/ Opteron 64 940 CPU's. Socket AM2 will use DDR2 and bring 65nm CPU's. AM2 will also incorporate a new technology called Z-RAM and others, which will increase the size of the Cache to enormous amounts, keep Segregated Caches and SGoI, which, according to IBM & AMD, will bring 40% increase in performance for AM2 and offer up to 4GHz Operation. Current Socket AM2 offerings (Not available yet):
XIII. Socket AM2 Speeds
(Clock Speed/L2 Cache/# of Cores)
Athlon 64 3500+ (2.333GHz/512KB/Single)
Athlon 64 3800+ (2.500GHz/512KB/Single)
Athlon 64 4000+ (2.500GHz/1MB/Single)
Athlon 64 X2 3800+ (Unknown/Dual)
Athlon 64 X2 4200+ (2.333GHz/512KB/Dual)
Athlon 64 X2 4400+ (Unknown/Dual)
Athlon 64 X2 4600+ (2.500GHz/512KB/Dual)
Athlon 64 X2 4800+ (2.500GHz/1MB/Dual)
Athlon 64 X2 5000+ (2.667GHz/512KB/Dual)
Athlon 64 X2 5200+ (Unknown)
Athlon 64 X2 5400+ (Unknown)
Athlon 64 FX-62 (2.800GHz/1MB/Dual)
Athlon 64 FX-64 (3.000GHz/1MB/Dual)
Athlon 64 FX-66 (Unknown)
XIV. Opteron 64 Processors [Sockets: 940]:
The first AMD Opteron 64 CPU's were releasd on April 22, 2003. These were 130nm processors running on Socket 940 ranging from 1.4GHz to 2.4GHz. They had 800MHz HyperTransport and 1MB Level 2 Cache each and ran on the Sledgehammer Core. Later, on February 14, 2005, 90nm versions were released, on the Venus (1xx Opterons) Troy (2xx Opterons) and Athens (8xx Opterons) cores. Each of these as well had 1MB of Level 2 Cache and added in SSE3 Instructions and a 1000MHz HyperTransport. The speeds ranged from 1.6GHz to 2.8GHz, there is a 3GHz version being released in the near future.
The Opteron differs from the Xeon because of it's implementation of HyperTransport. Opteron 64 CPU's perform far beyond Intel Xeon CPU's in Multi-Processor systems, due to CPU Coherency and Memory Access. AMD Opteron 64 systems can have as much as 22GB/s System Bandwidth and 11GB/s or more Memory Bandwidth, whereas an Intel Xeon CPU can have as much as 6.4GB/s System/Memory Bandwidth. Running under 64-bit Mode, the AMD Opteron 64 CPU is vastly superior to that of the Intel Xeon, especially in Linux x64 where performance can be up to 75% greater than Xeon. All AMD Opteron CPU's are 64-bit.
Dual-Core Opteron 64's compared to Dual-Core Xeon's, the Opteron 64's are again, vastly superior. This is due to the fact that just like Athlon 64 X2 processors, there is a "Crossbar" inside the CPU that allows the 2 cores to communicate to each other @ CPU Frequency. The Dual-Core Xeon CPU's have to use the Front Side Bus to communicate, and creates an even larger latency and performance impact. Current AMD Opteron s940 CPU's:
XV. Opteron 64 Speeds
130nm SledgeHammer
1.4GHz - 1MB L2 Cache - 800MHz HyperTransport
1.6GHz - 1MB L2 Cache - 800MHz HyperTransport
1.8GHz - 1MB L2 Cache - 800MHz HyperTransport
2.0GHz - 1MB L2 Cache - 800MHz HyperTransport
2.2GHz - 1MB L2 Cache - 800MHz HyperTransport
2.4GHz - 1MB L2 Cache - 800MHz HyperTransport
90nm Venus/Troy/Athens (1xx/2xx/8xx)
1.6GHz - 1MB L2 Cache - 1000MHz HyperTransport
1.8GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.0GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.2GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.4GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.6GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.8GHz - 1MB L2 Cache - 1000MHz HyperTransport
90nm Denmark/Italy/Egypt (1xx/2xx/8xx) (Dual-Core)
1.6GHz - 2MB L2 Cache - 1000MHz HyperTransport
1.8GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.0GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.2GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.4GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.6GHz - 2MB L2 Cache - 1000MHz HyperTransport
XVI. Opteron 64 Processors [Sockets: 1207]:
The Next Generation Opteron 64 Processors will be on Socket 1207, which s "Socket F". These Opteron 64 CPU's will bring Quad-Core CPU's, ability to use Quad Channel Memory, use DDR2 Memory, DDR3 in the future, as well as FB-DIMM Memory. The Socket F Opterons can also take advantage of Multiple Memory Controllers, as can Socket AM2, which will increase performance of Memory to each CPU.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XVII. Technical Information
Now for some technical stuff and explanations of clock frequencies and multpliers. This needs some tuning up, so anybody willing, please help to add some (don't start talking about TDP's or extremely technical things as this is geared towards newbies).
On every Athlon 64, X2, Pentium 4, and Pentium D processor (excluding Pentium Extreme Editions which runs at 266MHz, and some lower clocked P4's @ 133 and 166) there is a 200MHz Core Clock frequency, I like to refer to this as the "Crystal Clock". This is the speed at which the Processor truly runs at. There is something called "Clock Stepping" or a "Multiplier" as it is commonly known, where this allows the Internal Clock Frequency of the CPU to run at a much higher rate than the External Clock (Front Side Bus/HyperTransport).
Running a 200MHz Crystal Clock and a 10x Muliplier, means your CPU is running at 2000MHz (2GHz) Internal Speed, but this has no bearing on the speed of the Front Side Bus/HyperTransport. HyperTransport operates on Multiplier of it's own, up to 5x for 200MHz x5 = 1000MHz (1GHz) Operation, and HyperTransport is Full Duplex so it is 2000Mhz (2GHz) effective speed. Front Side Bus runs 200MHz Quad Pumped (All 4 sides of the Square Wave) which equals 800MHz effective speed since the Front Side Bus is Half Duplex.
With standard DDR1 Memory, it has 184 pins, and has very low latencies (2-2-2-5 and below) and provides up to 300MHz x2 Double Data Rate performance (with higher timings). On DDR2, it uses 240 pins, has higher latencies (3-3-3-8 and above) but runs on 333MHz x2 Double Data Rate and above, providing higher bandwidth than DDR1 at a price of higher latencies. DDR1 and DDR2 allow Dual Channel operation, meaning it has 2 lanes for information to pull from.
The Athlon 64 Socket 754 has a 64-bit Memory Controller, meaning it can only run Single Channel Operation, Socket 939 Athlon 64's as well as Socket 478 and Socket 775 use 128-bit Memory Controllers that allow Dual Channel Operation. Double Data Rate means it sends the pulse twice per clock cycle, or "Double Pumped", don't confuse this with Single & Dual Channel.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XVIII. Memory
I decided to add this in because I feel it is important for some people to know when purchasing a processor. The following information was obtained on Personal Experience and in no way guarantees that it will be what you experience or to be the exact truth, so please do not scrutinize unless I really messed up (I might have, heh).
Memory Latency & Speed: I have found that on an Athlon 64 system, comparing latencies such as 2-2-2-5 to 3-3-3-8, that the performance hit is not that great. I take this to be because of the Onboard Memory Controller. Being that the OMC reduces latencies tremendously already, higher latencies will not effect the A64 as much. On a Pentium 4 processor, I have found that higher latencies, will be a greater impact on performance (Not extravagent, but noticeable) and this is because the memory transmits along the Front Side Bus back to the CPU, vs. a direct connection back to the CPU @ RAM Speed (DDR400, DDR433, etc.).
When the Memory Controller (Be it OMC in A64 or NB in P4) goes to access something in memory, it first sends out a command (I like to refer to this as the RFI, Request For Information) to obtain data, this is where your latencies come from. First, a bank of memory is accessed, known as Chip Select (CS) and this is where your "Command Rate" comes into play. Your "Command Rate" is how many cycles it takes to latch onto a physical chip, 1T means after the 1st cycle, 2T meaning after the 2nd cycle. First, the row (RAS) is found to where the data is in, then there is a wait for the RAM to actually access the row (RCD). Then, it defines the column the data is in (CAS), and then it waits for the data to be sent to the pins, and out back to the CPU. After that is done, it precharges the row to be used again (RP). These are defined by your 2-2-2-5 (CAS-RCD-RP-RAS, sometimes a "t" previews the letters, this "t" stands for "time").
A few things to note: latencies of .5 (e.g., 2.5) can only be obtained on Double Data Rate memory, where 2 parts of each clock cycle are used. There is a built-in SPD (Serial Presence Detect) that you may notice with new RAM, it may set your timings and volts to something other than advertised. This is to ensure stability on 1st boot. When you attempt to overclock memory, it is best practice to relax the timings (e.g., go from 2-2-2-5 to 3-3-3-8 @ higher frequencies) this is to ensure stability, also note you may have to increase voltage to the DIMM to remain stable at higher frequencies.
In memory, comparing Single Channel to Dual Channel, the CPU is a limiting factor. What I mean is that for each pin on the RAM, e.g., 184-Pin DDR1, that requires 1 pin from the CPU as well. In addition to that, for each channel, it requres double that of it's Single Channel counterpart. This means to run Dual Channel DDR1 w/ 184 Pins, it would require 368 Pins (184x2) from the CPU and this is why the Socket 754 processors do not support Dual Channel, as well as because the Memory Controller is 64-bit whereas a 128-bit Memory Controller (such as on the s939 and s478/s775) is required for Dual Channel.
Memory - Overclocking
The following is based on Personal Experience and what I have seen and been told of by other friends, colleagues, and abroad. The results I present may not be what you experience.
Undoubtably, the best overclocking RAM you can buy is OCZ...pure and simple. OCZ for AMD & Intel has proven to be the best overclockers for RAM, period. OCZ even took it to a new level, with their VX Series, known as eXtended Voltage, you can run the tightest timings (2-2-2-5) @ DDR500 speeds and above, w/ lots 'o' volts! The following is a list of the companies for which the RAM is the best and their best, as well as a overall best overclocking RAM at the bottom:
OCZ
Crucial
Corsair
G.Skill
Mushkin
Patriot
GigaRAM
Kingston
PNY
Viking
OCZ - Platinum Series
Crucial - Ballistix & Tracer Series
Corsair - XMS Series
G.Skill - Extreme Series
Mushkin - Redline Series
The overall best overclocking memory is the OCZ Platinum series, overclocking to DDR600 and beyond is amazing feat's for any RAM!
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XIX. Cache Comparison
This is a comparison of 512KB, 1MB, and 2MB of Level 2 Cache. This is based off of Personal Experience and may not be what you experience.
Some people like to say more cache, more performance, but I don't think that is always true. In most desktop environments, such as Gaming, I can see little to no increase in performance w/ 1MB of Cache over 512KB, and 2MB seems to even slow down compared to 512KB in games. It is true that a CPU can access 512KB of Cache in less time, but there is less total cache. Whereas 1MB or more, it takes a little longer to sort through, but there is alot more of it present. I find that in alot of applications that don't branch much (change what they're doing), more cache seems to be of benefit there.
If you're a solid gamer, I recommend simply an A64 CPU w/ 512KB Cache, such as the 3200+, or P4 530 w/ 1MB Cache. The less cache in games will give you better performance, in my experience. AMD says that 512KB Cache provides 200MHz Increase in performance, which is why you have a 3500+ w/ 512KB and a 3700+ w/ 1MB and they're both 2.2GHz. Intel seems to think there is an even greater increase in performance, but this is just due to the bottlenecked Front Side Bus in my opinion.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XX. Future Technologies
This section is about the future technologies AMD and Intel are planning to implement into their Desktop and Server lines of CPU's. All of it may be of speculative nature from various websites and myself, I will not say "This tech will pwn the other!" or say any bias statement, but rather what it is targeted to do, nothing more.
Intel Future Technologies
Intel plans to increase the FSB on their future "Conroe" and "Woodcrest" CPU's up to 1066MHz (266MHz Quad Pumped) and 1333MHz (333MHz Quad Pumped). This will alliviate the need to place the Memory Controller On-Die for a period of time. The 1066MHz for "Conroe" will be in the mainstream CPU's with 1 version, the "Extreme Edition" being with a 1333MHz Front Side Bus & 3.33GHz Internal Clock. Another technology they are developing is called DIB (Dual Independent Bus). This will give each CPU a Front Side Bus to the Northbridge in a Dual-Core system. This will not, however, increase Memory Bandwidth significantly due to there still being only a single Memory Controller on the Northbridge.
Another technology they are developing, though it might be facing some delays, is known as CSI (Common System Interface). This is the challenger to AMD's HyperTransport Technology. There isn't a whole lot known about CSI right now, it's a Serial Interconnect like HyperTransport, but initially, according to some sources, CSI was designed to beat HyperTransport, but it appears that that plan has failed. CSI seems to be the equivilent to a 1600MHz FSB. The first implementation of CSI will be in the core "Tukwila" that is Quad-Core. CSI is also planned for the Intel Xeon processor, it will be in the Quad-Core "Whitefield" core. The Xeon version w/ CSI will come out in 2009 while the Itantium version should be out in 2008.
There is another architecture Intel is developing, known as Nehalem, there is absolutely nothing known about this as far as I can tell, other than it will be in the Xeon core "Gainestown" that will be Quad-Core, come post-2010, and start at 3GHz.
AMD Future Technologies
AMD is quite the busy-bee, they are working on several new technologies, most notably, HyperTransport 3.0 and core revisions. HyperTransport 3.0 should scale beyond 1600MHz (3200MHz effective) and use 32-bit or 64-bit data widths. There is also HyperTransport 4.0 being developed, but nothing is official as of yet. Another technology is Direct Connect Architecture 2.0. This tech, set to be released in 2008, will be inplemented to have greater than 8-CPU servers. A revison, only codename is "K8L", but that term was coined by TheInquirer.net, is set to be released in 2006, and contains mainly double number of FPU's which should bring 1.5x the performance of K8.
AMD is also working on a tool that will allow multiple CPU's to work on a Single Thread, it's kind of like "Reverse HyperThreading". It's to be implemented in the next-generation of AMD CPU's.
Others such as the recently licensed Z-RAM technology, should bring 20% to 50% improvements to the current K8 architecture, according to some sources. AMD is also working on putting Level 3 Cache into the Opteron 64 Socket 1207 "F" platform, the L3 may be Z-RAM or another form of Static RAM. DDR2 is to be brought into Socket AM2 and Socket 1207 Opteron 64 servers, as well as DDR3 in 2007. AMD plans to bring 65nm CPU's at the end of 2006 and 45nm sometime in 2008. There are also plans to bring a new architecture, K10, in 2008. This will, according to AMD, bring 10GHz Operation, better Branch Prediction, better FPU performance, and allow for Multi-Multi-CPU systems.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXI. What's That?
This section explains some of those things you see such as "ECC & Registered RAM", "Level 2 Cache", "RAM Timings", etc. This does not get technical so I can maintain the original idea of being everybody-friendly. If something sounds so non-technical that it sounds wrong, please make sure to tell me so I can fix it. I will continue to update this as necessary to stay up to date
32-bit & 64-bit computing - 32-bit and 64-bit refer to the size of the registers inside the CPU and the data-size of the information processed.
Bandwidth - Bandwidth (in RAM, Video Cards, Front Side Bus, etc.) are found in a mathematical equation, it is: Frequency x Data Width / 8 = Bandwidth. This means, a Frequency of 1000MHz (on say, a Video Card) and a 256-bit Memory Interface (the physical size of how wide the bus is) will give you 32GB/s Memory Bandwidth, 1000x256/8=32000MB/s.
Cache - Cache is used to store often-used information or commands. In the case of Level 2 Cache in a CPU, it holds frequently used data and instructions for where in RAM the data can be found.
Coherency - Coherency refers to the communication between 2 devices or components that are to act in symmetry. In CPU's, it's the communication beween CPU's to know what the other is doing and what it is in it's RAM/Cache. AMD Opteron 64 based-computers have extremely more efficient coherency than Xeon's, as well as the Athlon X2 has much better coherency than a Pentium D due to it's Crossbar.
CSI - CSI (Common System Interface) is a Serial Point-To-Point protocol designed by Intel to counter Hypertransport. Current sources put CSI comparable to a 1600MHz Front Side Bus.
Crossbar - The Crossbar in a Dual-Core AMD64 CPU is the component that allows the 2 cores to communicate to each other @ CPU speed, rather than having to use the Front Side Bus as in the Xeon or Pentium D.
ECC - ECC stands for Error Correcting Code. ECC is used for mainly in servers, although some Socket 939 AMD Athlon 64 motherboards support ECC because the Memory Controller has ECC as well. ECC corrects some errors in memory where 111% integrity is needed. ECC works by adding 1 bit of parity per 8 bits, that means it can fix some 1-bit errors, it will only detect 2-bit errors but won't fix them.
HyperTransport - A Serial Point-To-Point protocol designed by AMD & Founders for High Bandwidth & Low Latency communication between devices. This is currently implemented in many networking devices (such as switches, routers, and such) and also in certain computer Microprocessors, namely the Athlon 64 and Opteron 64. The fastest HyperTransport is 32-bit x 32-bit 1400MHz (2800MHz Effective) and provides 22.4GB/s bandwidth.
NUMA - NUMA (Non-Uniform Memory Access) allows multiple memory banks to be shared among multiple CPU's in a Multi-Processor system.
Register - A CPU register is something that stores currently used information for the CPU to process and send out.
Registered - Registered RAM is a chip on the module that acts kinda like another Memory Controller. The Memory Controller only sees the chip, and the Memory Controller sends commands to the Chip and that chip disperses and controls the module.
TDP - TDP (Standing for Thermal Design Power or Point) is the maximum heat the chip will dissipate (via any method) without reaching the maximum temp for the CPU. TDP does not say how much power the CPU will draw, but just the maximum power it will draw running @ a "real" load, which can be far from real-world use.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXII. Motherboards
This section is about the overall best motherboards for stability and overclocking for the End-User, it is denoted by brand, as well as a specific board. This also depicts some of the best features and the best overall motherboard based on Features, Price, Overclocking, and Stability. Note: The following is obtained by either Personal Experience, or by me speaking with others about the boards. You may experience different results than what I present.
Best Overclocking Motherboard (AMD): DFI - LanParty nF4 SLI-DR Expert Edition
The Expert Edition motherboard has been one of the best overclocking boards, I have personally taken this board past 300MHz with an Opteron CPU using minimal volts. This board has been known to overclock up to 350MHz and beyond with ease, which can clearly make it the best board. DFI itself is known to be the best for overclockers, they always give the gamer and enthusiast what they want and this board doesn't dissapoint. The price tags for some DFI boards can leave you breathless, but once you take the baby home and feed it, you're going to wish you had 2!
Best Overclocking Motherboard (Intel): Gigabyte - Gigabyte GA-G1975X
This board is a really nice overclocker, nice layout, and has a nice ergonomic design to it. This board has been overclocked to over 4.5GHz with a P4 and I wouldn't doubt 5GHz or above is achievable. The price of this board, and the fact it lacks SLI, is something to note, but if you're a hardcore enthusiast, you go the road I go, Single Card Solutions and OC To The Max!
Best Stable Motherboard (AMD & Intel): Tyan - Server & Workstation Boards
Tyan may not be the first choice for gamers or enthusiasts, but it is certaintly the talk of the town when it comes to Server and Workstation motherboards. Tyan has a proven record of reliability in the Workstation and Server area, and I would personally say through experience that one of their best boards is the Tyan K8WE S2985. If you're looking for a stable solution and don't plan to overclock, look no further than Tyan for your next build!
Best Priced Motherboard (AMD): ASRock & EpoX - Variants
ASRock & EpoX are showing their contendership with the likes of ASUS, DFI, and other big-name builders, and they're accomplishing this with prices that show they are a force to be reckoned with. The boards from these 2 makers are exceptionally well in their price/performance and even can contend with high-end boards such as LanParty's if given the right builder.
Best Priced Motherboard (Intel): ASUS - Variants
ASUS is pretty much the best motherboard builder for Intel that offers the best prices. ASUS may not be so hot with AMD's performance motherboards, but on the Intel Camp, ASUS is a fair trader in performance/price battle against the top Gigabyte and Intel offerings.
Overall Best Motherboard - Features, Etc. (AMD): ASUS A8N-SLI Premium
This board has it all....SLI, Sata 2 3Gb/s, PATA RAID, 8 SATA Connectors, 3 PCI Slots, Dual Gigabit...this board is one of the best boards for all around performance, price, and features for the AMD motherboard options. This board may not be touted as the best overclocker or best stability, but it's feature richness means it can contend anyday with all the other boards, and it's at a fairly reasonable price considering its features.
Overall Best Motherboard - Features, Etc. (Intel): ASUS P5N32-SLI
This board has alot of features, 6 SATA that are 3Gb/s, SLI, support for 16GB of RAM, Dual Gigabit, 2 PCI Slots, 5 PCI-E Slots of various speeds, this board is the most feature-rich board for the Intel platform, this does come at a hefty price, but it's got tons of features to make it worth while.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXIII. Naming Scheme: What They Mean:
This section is to familiarize users with the naming schemes that the b@!stards @ AMD & Intel keep coming up with to cunfubble us!
AMD
Opteron 64 Series (All have 1MB Cache):
Opteron 64 x44 - 1800MHz - Single Core
Opteron 64 x46 - 2000MHz - Single Core
Opteron 64 x48 - 2200MHz - Single Core
Opteron 64 x50 - 2400MHz - Single Core
Opteron 64 x52 - 2600MHz - Single Core
Opteron 64 x54 - 2800MHz - Single Core
Opteron 64 x56 - 3000MHz - Single Core
Opteron 64 x65 - 1800MHz - Dual Core
Opteron 64 x70 - 2000MHz - Dual Core
Opteron 64 x75 - 2200MHz - Dual Core
Opteron 64 x80 - 2400MHz - Dual Core
Opteron 64 x85 - 2600MHz - Dual Core
Opteron 64 x90 - 2800MHz - Dual Core
Athlon 64 Series:
Athlon 64 2800+ - 1800MHz (512KB Cache) - Socket 754
Athlon 64 3000+ - 2000MHz (512KB Cache) - Socket 754 & 1800MHz (512KB Cache) Socket 939
Athlon 64 3200+ - 2200MHz (512KB Cache)/2000MHz (1MB Cache) Socket 754 & 2000MHz (512KB Cache) Socket 939
Athlon 64 3400+ - 2400MHz (512KB Cache) - Socket 754
Athlon 64 3500+ - 2200MHz (512KB Cache) - Socket 939
Athlon 64 3700+ - 2400MHz (1MB Cache) - Socket 754 & 2200MHz (1MB Cache) - Socket 939
Athlon 64 3800+ - 2400MHz (512KB Cache) - Socket 939
Athlon 64 4000+ - 2400MHz (1MB Cache) - Socket 939
Athlon 64 X2 Series:
Athlon 64 X2 3800+ - (512KB Cache Per Core) - 2000MHz - Socket 939
Athlon 64 X2 4200+ - (512KB Cache Per Core) - 2200MHz - Socket 939
Athlon 64 X2 4400+ - (1MB Cache Per Core) - 2200MHz - Socket 939
Athlon 64 X2 4600+ - (512KB Cache Per Core) - 2400MHz - Socket 939
Athlon 64 X2 4800+ - (1MB Cache Per Core) - 2400MHz - Socket 939
Intel
Pentium 4 Series:
Pentium 4 x3x - (512KB, 1MB or 2MB Cache) - 3000MHz - Socket 478/T
Pentium 4 x4x - (512KB, 1MB or 2MB Cache) - 3200MHz - Socket 478/T
Pentium 4 x5x - (512KB, 1MB or 2MB Cache) - 3400MHz - Socket 478/T
Pentium 4 x6x - (512KB, 1MB or 2MB Cache) - 3600MHz - Socket 478/T
Pentium 4 x7x - (512KB, 1MB or 2MB Cache) - 3800MHz - Socket 478/T
Pentium D Series:
Pentium D x2x - (1MB or 2MB Cache Per Core) - 2800MHz - Socket T
Pentium D x3x - (1MB or 2MB Cache Per Core) - 3000MHz - Socket T
Pentium D x4x - (1MB or 2MB Cache Per Core) - 3200MHz - Socket T
Pentium D x5x - (1MB or 2MB Cache Per Core) - 3400MHz - Socket T
Pentium D x55 - (2MB Cache Per Core) - 3460MHz - Socket T
Pentium D x65 - (2MB Cache Per Core) - 3733MHz - Socket T
I have lacked Xeon as there are simply WAY too many Xeon changes to note on.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXIIV. Past Changes:
4-12-06 Changes: Added a "Motherboard" section below to denote the best overclockers, stability, and features. Added specifics to the "Memory" section as to the best overclocking memory and best stability, etc. Added "Naming Schemes: What They Mean" section, displays all the namings of Opteron 64's (e.g., Opteron 170), Pentium 4's (e.g., Pentium 4 530) so one can deduce what they are without reading the jumble in advertising.
4-10-06 Changes: Added more information to "Conroe" and "AM2"
3-30-06 Changes: Added the "Pentium D 965 EE" Processor. Also fixed spellings and added more information. Also fixed some things that didn't make sense.... Added "HyperTransport", "CSI", and alphabetized the "What's That?" section.
3-27-06 Changes: Added another term to the "What's That?" section. Added a Table of Contents and bulleted sections for better viewing. Changed name to something drastic...hope it works ;
3-26-06 Changes: Fixed "Conroe" speeds and added "Opteron 64" speeds. Added more info into the thread and cleaned up some parts. Changed name again....
3-23-06 Changes: Added a "What's That?" section below "Future Technologies" that explains some Terms and Technologies that are inside current CPU's from AMD & Intel and others. Touched-up some information and cleaned-up a little. Changed the view of this from the 2 CPU's to overall computer hardware.
3-19-06 Changes: Added "Conroe" and Socket AM2 clock speeds. Made minor changes to the information and cleaned up some parts. Removed pricings on CPU recommendations. Also changed the title to a more passive presentation.
3-16-06 Changes: Added a "Future Technologies" section below "Cache Comparison".
3-12-06 Changes: Added preliminary "Conroe", "Socket 1207", "Socket AM2", "Dempsey" information. Also changed the titles of each section to a more ergonomic view.
3-5-06 Changes: Added the Intel Pentium D 805.
3-2-06 Changes: Added Opteron 64 and Xeon CPU information.
"Cache Comparison" added below "Memory". Changed some information and added more based on user posts.
3-1-06 Changes: Added more information to "Memory" section below and edited information in main thread.
2-25-06 Changes: Changed pricings, they reflect multiple vendors now. As well as some information gathered from fellow posters. Also changed some wordings, information and pretext to the thread, as well as a "Memory" section added below. Also noted on "Cedar Mill" desktop Pentium 4 processor core and edited more information.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Thank you again for reading this, and make sure to post if anything is wrong or with suggestions, peace.
~~Mad Mod Mike, pimpin' the world 1 rig at a time
4-14-06 Changes: Removed some content users found to be "Biased". Added preliminary information on "Clovertown", and added more to AMD's Future Products.
*Refer to the very bottom of this post for past changes*
Main Idea!! DO NOT Make this a Fanboy Thread or Flame Thread, Thank you.
I wrote this originally as a way to describe the difference between an AMD and Intel CPU, but over time, it changed to adding in technologies related to the CPU's and eventually, an overall computer hardware explanation. This has been geared towards newbies to enthusiasts and everybody in between. This has been kept as Technically-Negated as possible, so everybody can understand it. If you find something wrong with this, or wish to have something added that is lacking, PM me or drop a post here and I will be happy to accomodate.
Table of Contents
I. Intel Processors - Pentium 4
II. Intel Processors - Pentium 4: Recommendations
III. Intel Processors - Pentium D
IV. Intel Processors - Pentium D: Recommendations
V. Intel Processors - Xeon
VI. Intel Processors - Conroe
VII. Intel Processors - Conroe Speeds
VIII. AMD Processors - Athlon 64
IX. AMD Processors - Athlon 64: Recommendations
X. AMD Processors - Athlon 64 X2
XI. AMD Processors - Athlon 64 X2: Reommendations
XII. AMD Processors - Socket AM2
XIII. AMD Processors - Socket AM2 Speeds
XIV. AMD Processors - Opteron 64
XV. AMD Processors - Opteron 64 Speeds
XVI: AMD Processors - Opteron 64: Socket 1207 "F"
XVII: Technical Information (FSB, HyperTransport, etc.)
XVIII: Memory (Timings, Bandwidth, etc.)
XIX: Cache Comparison (Level 2 Performance)
XX: Future Technologies
XXI: What's That? (Terms)
XXII: Motherboards
XXIII: Naming Schemes: What They Mean
XXIV: Past Changes
Intel Processors
I. Pentium 4 Processors [Sockets: 423 478 775]:
The main thing to point out is that the Pentium 4 uses the traditional "Front Side Bus". This is the "data pipeline" in which information moves from the CPU to the Motherboard. This is actually composed of 3 bus's in of itself, a "Data", "Command", and "Address". Maybe you heard of the 800MHz Front Side Bus before, the truth is is that only the "Data" bus runs at 800MHz, the "Command" & "Address" run at 200MHz (I will explain system clock speeds and Front Side Bus speeds at the end).
On a Pentium 4, the Memory (RAM) is accessed via the Memory Controller that lies within the Northbridge, which means that the CPU must first access the Northbridge which in turn accesses the Memory. After retrieving the data from the Memory, it than goes back through the Northbridge, then to the Front Side Bus back into the CPU for processing. This is actually a con on the Pentium 4, because the Front Side Bus is a Half Duplex, Bidirectional Bus. This means that the CPU cannot send and receive information on the Front Side Bus (HyperTransport in Athlon 64;s) at the same time.
With the Prescott core on some Pentium 4's, another concern arises: Heat. The Prescott core runs very hot compared to the Northwood or Athlon 64 cores, which means that if you plan to purchase an Intel Pentium 4 system with Prescott core, make sure to purchase an Aftermarket Cooler with it. The Prescott though, can be overclocked very high, so if you're an Overclocking Enthusiast or want to get into overclocking, the Prescott Pentium 4 will not let you down, just make sure to get a very good Aftermarket Heatsink & Fan to cool it.
The Prescott Pentium 4 also has a smaller Level 1 Cache than the Athlon 64, avg. of 48KB Level 1 vs. 128KB Level 1 in the Athlon 64. The Pentium 4 Prescott has larger Level 2 Cache, 2MB vs. 512KB or 1MB in the Athlon 64. The larger L2 Cache in the Pentium 4 helps leverage the bottlenecked Front Side Bus, but I personally believe that larger Level 2 Cache does not mean greater performance, and in some cases, I believe it to decrease performance (Refer to "Cache Comparison" at end).
A new core was recently released on 65nm, known as "Cedar Mill". The major changes from Prescott is the Thermal Design Power (TDP) has been lowered to 86 Watts. On the 6x2 (e.g., P4 632) versions, there is also something knew, this is known as Virtualization Technology. This Vritualization Technology allows you to run multiple operatings systems w/o the cost of emulation. This technology is being implemented in 9xx Series Dual Core Pentium 4's and Xeon 7xxx Series, as well as Core Duo processors.
One renowned thing about the Pentium 4, is HyperThreading. HyperThreading fakes the Operating System into thinking that there is 2 processors inside the chip, kinda like Dual-Core but there really isn't a 2nd processor. This works by (Not getting too technical) allowing the CPU to operate on 2 threads vs. 1 on a Non-HyperThreaded CPU. One thing to note though, if I recall correctly, is that HyperThreading is Asynchronous, meaning the CPU cannot process both threads at the same time. This does help in Multi-Tasking as it allows the CPU to start processing another thread w/o wasting CPU clock cycles sittng idle waiting for a program to end. The real-word performance benefit from HyperThreading is 5-15%, with some scenarios resulting in decreased performance, especially in Server Benchmarks.
The first Pentium 4's ran on Socket 423, but due to an inadequate electrical design, a Socket 423 P4's max speed was 2.0GHz, this was replaced with Socket 478, this uses Dual Channel DDR1 Memory and allows for much higher clock speeds. The newer generation of Pentium 4's is on Socket 775 (Socket T) and uses Dual Channel DDR2 Memory. If you buy a Socket 775, make sure you purchase the right Memory, all the 6xx Series processors are s775 and use DDR2, as well as the 5xx s775, 8xx, and 9xx Series CPU's.
Another thing to note is whether or not to get a 64-bit processor. I strongly recommend getting a 64-bit CPU, so if you want that, you must get one of the 6xx Series Pentium 4's (Or one of the newer 5xx Series, but I recommend a 6xx Series). I have changed the P4 Single Core to reflect Cedar Mill, Cedar Mill Pentium 4's are 631 through 661, there is no 671 (3.8Ghz) Cedar Mill. The best Pentium 4 offerings are:
II. Pentium 4: Recommendations
Pentium 4 631 (3.0GHz)
Pentium 4 641 (3.2GHz) <--Recommended 6xx Series Processor
Pentium 4 651 (3.4GHz)
Pentium 4 661 (3.6GHz)
Pentium 4 670 (3.8GHz)
III. Pentium D 8xx & 9xx Processors [Sockets: 775]:
The Intel Pentium D 8xx & 9xx CPU's are Dual-Core, meaning there is 2 processors on 1 chip. The 2 processors in the 8xx and 9xx series communicate to each other over the standard Front Side Bus, thus again creating another bottleneck and latency issue as stated in the Pentium 4. The 8xx and 9xx Pentium D's also run on 800MHz Front Side Bus as stated, with the exception of the new Intel Pentium D 9xx Extreme Editions which have a 1066MHz Front Side Bus.
On the Intel Pentium D 8xx Series, the standard models (820 to 840) do not have HyperThreading, only the Extreme Edition has HyperThreading turned on, which makes it seem like you have 4 Processors (HyperThreading 2 x 2 CPU's). The Same goes for the Intel Pentium D 9xx Series, the 920 to 950 have 800MHz Front Side Bus w/o HyperThreading, the 955 Extreme has 1066MHz Front Side Bus and HyperThreading turned on. Another thing to note is the 8xx Series is 90nm process, whereas the 9xx Series is 65nm. 65nm will aid in overclocking and heat issues, thus making the 9xx Series superior to the 8xx Series. I would not recommend a 8xx Series, better to get the 9xx Series.
The 9xx Series also has 4MB Level 2 Cache (2MB Per Core) vs. the 8xx Series which has 2MB Level 2 Cache (1MB Per Core). The Pentium D's, both 8xx and 9xx, run on Socket 775. So remember what I said bout Socket 775 and memory in the Pentium 4 posting. I have added the Pentium D 805 as a possible buy, due to it's extremely low price and ability to be overclock very high. The best Pentium D offerings:
IV. Pentium D: Recommendations
Pentium D 805 (2.66GHz) <=Recommended For Budget/Overclockers
Pentium D 920 (2.8GHz)
Pentium D 930 (3.0GHz) <--Recommended 9xx Series Processor
Pentium D 940 (3.2GHz)
Pentium D 950 (3.4GHz)
Pentium D 955 (3.46GHz Extreme Edition/1066MHz FSB)
Pentium D 965 (3.73GHz Extreme Edition/1066MHz FSB)
V. Xeon Processors [Sockets: Slot 2 603 604]:
The Intel Xeon CPU was first released in 1998 named "Intel Pentium II Xeon" named Drake. It was a replacement for the Pentium Pro and had 512KB, 1MB or 2MB Cache w/ 100MHz FSB and was a Slot 2 processor. It also came with Dual, Quad, and Octal CPU configurations w/ additional logic and a newer chipset. In 1999, the Pentium III Xeon was released, this was almost identical to the Pentium II Xeon but it had a few differences, such as SSE and a cache controller enhancement. There was a 2nd version released, named Cascades, that was often suffixed with Cascades 2MB, to differentiate it from the Tanner. The Cascades had 1MB or 2MB Cache vs. Tanner which had 256KB.
The Tanner had a 133MHz FSB whereas the Cascades 2MB had a 100MHz FSB. After that, in Mid-2001, the standard Xeon (lacking the "Pentium" aftermark) was introduced. This was the first Xeon CPU to use the "Netburst" architecture, this CPU was named "Foster". This chip, however, was outperformed by Cascades 2MB and Athlon MP's. Foster also required Rambus Memory, making it quite expensive. And Foster only allowed up to 2 CPU's, so a 2nd version, named Foster MP, was released w/ 1MB of Level 3 Cache. In 2002, Prestonia, a 130nm Xeon was released. This was the first time "HyperThreading" was used with Xeon's, these also had 512KB of Level 2 Cache.
These Xeon CPU's had a 533MHz FSB and supported Dual Channel memory. The Xeon Prestonia would bring 1MB or 2MB of Level 3 Cache. The Xeon MP was the Gallatin, this brought 4MB of Level 3 Cache. Even later, a 64-bit version of the Xeon, named Nocona, was released. This also brought support for PCI-Express, DDR2, and SATA. After that, Irwindale was released, a 64-bit Xeon w/ Double the Level 2 Cache and the ability to reduce clock speeds in times of low processing. In April 2005, 64-bit Xeon MP's were released, named Cranford and Potomac. Cranford had 4MB Level 2 Cache and Potomac had 8MB Cache.
Xeon speeds range from 400MHz in Tanner to 3.667GHz in Xeon MP's. A Dual-Core Xeon was released on October 10, 2005. This runs at 2.8GHz w/ an 800MHz FSB. The architecture for these is the same, 2 CPU's share 1 front side bus, which gives each CPU 1600MB/s Bandwidth using DDR400 (4-Way Xeon's have 800MB/s Bandwidth per CPU). These had 2MB of Cache per core (4MB Total). A Xeon MP version was released, one with 1MB of Cache per core and one with 2MB Cache per core. Both Xeon Dual-Core CPU's are named Paxville.
The Next Generation of Xeon's is known as "Dempsey". These CPU's sport a 266MHz FSB Quad Pumped or a 333MHz FSB Quad Pumped. They also implement something known as Dual Independent Bus's, which gives each Core its own FSB to the Northbridge. Early benchmarks compared to Current Opteron 64 CPU's put Dempsey ahead in the race in some benchmarks. Preliminary benchmarks put "Clovertown", the Quad-Core version of "Dempsey", on par with a 2.0GHz AMD Opteron 64 in Single-Threaded environments Clock-For-Clock. "Clovertown" also has HyperThreading, which gives it the ability to work on 8 threads.
VI. Conroe Processors [Sockets: 775]:
The Next Generation for Desktop CPU's from Intel is the "Conroe". "Conroe" will sport DDR2, and FSB's ranging from 800MHz, 1066MHz, and 1333MHz for the Extreme Edition. Clock Speeds from 2.4GHz up to 3.33GHz in the Extreme Edition. The main difference between Conroe and past Pentium 4's, is Conroe has a 14-Stage Pipeline, SSE4 Instructions (Maybe), and a 4-Issue Core vs. 3-Issue in past CPU's. Conroe is all 65nm and 45nm in the future. The Unified L2 Cache of Conroe can cause Cache Thrashing, and this could only be alleviated by CPU Overhead controlling it, which could mean alot of extra clock cycles. Cache Thrashing could prove to be a problem for Server Environments, if it proves true, it could spell bad news for Intel's Server Market.
Early benchmarks of Conroe either put it on par with AMD's K8's, or a little ahead, depending on the test.
Current Conroe offerings (Not yet released):
VII. Conroe Speeds
Core Duo E8?00 - 3.33 GHz (4MB L2 Cache) 1,333MHz FSB
Core Duo E6800 - 2.93 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6700 - 2.67 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6600 - 2.40 GHz (4MB L2 Cache) 1,066MHz FSB
Core Duo E6400 - 2.13 GHz (2MB L2 Cache) 1,066MHz FSB
Core Duo E6300 - 1.86 GHz (2MB L2 Cache) 800MHz FSB
Core Duo E4200 - 1.66 GHz (2MB L2 Cache) 800MHz FSB (Single-Core)
AMD Processors
VIII. Athlon 64 Processors [Sockets: 754 939]:
The AMD Athlon 64 processors come in 2 packages, Socket 754 and Socket 939. The main difference is Socket 939 utilizes Dual Channel memory, where the Socket 754 uses Single Channel memory. The fastest Socket 754 Athlon 64 CPU you can get, is the AMD Athlon 64 3700+ 2.4GHz Clawhammer. The Fastest Athlon 64 Socket 939 is the Athlon 64 4000+ San Diego 2.4GHz.
The Athlon 64 is different from the Pentium 4 in a few ways, mainly is the Onboard Memory Controller and the use of HyperTransport instead of the traditional Front Side Bus. The Onboard Memory Controller means that the CPU can access the memory directly w/o having to go through the Northbridge to retrieve information from memory. This reduces latencies and bottlenecks and provides more efficient throughput than the Pentium 4 processor.
The use of HyperTransport is a big step for AMD, this takes the 3 bus's in the Front Side Bus (Data, Address, & Command) and groups them into 32-bit packets. Think of it like a highway, instead of having 3 people run down the road as fast as they can, they all 3 hop into a car and drive down 1 busline (16-bits wide). Also one thing to note is the speed difference, Hypertransport allows speeds in current Athlon 64's of either 800MHz or 1000MHz (1GHz) Speeds (Socket 754 uses 800MHz while Socket 939 uses 1000MHz). Also, HyperTransport is a Full Duplex, UniDirectional architecture, and each link consists of 2 16-bit lanes for the packets to travel on. This also means that information can be sent and received from the CPU to Northbridge at the same time, each at 800MHz or 1000MHz (1GHz).
It is important to know that Athlon 64 CPU's do not have a Front Side Bus, rather they use HyperTransport which very much acts like the Front Side Bus but it is a replacement to reduce latencies and bottlenecks. Athlon 64's have either 512KB of Level 2 Cache or 1MB of Level 2 Cache, as stated in the Pentium 4 part. All Athlon 64 and above CPU's have 128KB of Level 1 Cache.
The Socket 754 Athlon 64 CPU's are mostly made on 130nm, vs. 90nm for newer Socket 939 processors. There is 1 90nm Socket 754, it is the Athlon 64 3000+ Venice, and there are alot of 130nm Socket 939 processors. I recommend getting a Socket 939 90nm Athlon 64 CPU for best performance and overclocking potential. If you plan to overclock, one thing to note is that you must adjust the HyperTransport speed. HyperTransport becomes unstable after 1000MHz operation, so make sure to set it's multiplier to something lower if you overclock so as not to disrupt the stability of the system.
I only recommend getting an Athlon 64 Socket 939 Processor, mainly because Socket 939 provides Dual Channel Memory Access, Faster Processors, and newer SLI & CrossFire Motherboards. Because of this, I have listed some Socket 939 Athlon 64 processors (All are 90nm). The current Socket 939 Athlon 64 offerings:
IX. Athlon 64: Recommendations
(CPU Speed/Core/L2 Cache)
Athlon 64 3000+ (1.8GHz/Venice/512KB)
Athlon 64 3200+ (2.0GHz/Venice/512KB)<--Recommended 939 A64 CPU
Athlon 64 3500+ (2.2GHz/Venice/512KB)
Athlon 64 3700+ (2.2GHz/SanDiego/1MB)
Athlon 64 3800+ (2.4GHz/Venice/512KB)
Athlon 64 4000+ (2.4GHz/SanDiego/1MB)
X. Athlon 64 X2 Processors [Sockets: 939]:
The Athlon 64 X2 Processor is the Dual-Core Athlon 64, meaning it has 2 processors on 1 chip. The difference between this Dual-Core and the Pentium D Dual-Core, is mainly the way he 2 CPU's communicate. On the Athlon 64 X2 Processor, the 2 CPU's communicate via a "Crossbar", which is an area in the CPU that the 2 CPU's can communicate at CPU frequency, rather than using the Front Side Bus or HyperTransport to communicate, this provides better "Coherency", which is communication between multiple CPU's.
All Athlon 64 X2 processors are on Socket 939, with newer ones being released for Socket AM2 when it is released. I don't shun away from any X2 CPU, so I state them all (All Athlon 64 X2's with 1MB L2 Cache per core run on Toledo Core, X2's with 512KB per core run on Manchester Core except the new 3800+ that runs on Toledo as well but with 512KB L2 Cache). Current Athlon X2 offerings (I included the FX-60 as it runs on an X2 core):
XI. Athlon 64 X2: Recommendations
(Speed/L2Cache Per Core)
Athlon 64 X2 3800+ (2.0GHz/512KB)
Athlon 64 X2 3800+ (2.0GHz/512KB/Toledo Core)
Athlon 64 X2 4200+ (2.2GHz/512KB)
Athlon 64 X2 4400+ (2.2GHz/1MB) <--Recommended 939 A64 X2 Processor
Athlon 64 X2 4600+ (2.4GHz/512KB)
Athlon 64 X2 4800+ (2.4GHz/1MB)
Athlon 64 FX-60 (2.6GHz/1MB/Toledo Core)
XII. AM2 Processors [Sockets: 940]:
The next Athlon 64 CPU's will be on Socket AM2, which as 940 pins. One thing to note is that AM2 is not pin compatible w/ Opteron 64 940 CPU's. Socket AM2 will use DDR2 and bring 65nm CPU's. AM2 will also incorporate a new technology called Z-RAM and others, which will increase the size of the Cache to enormous amounts, keep Segregated Caches and SGoI, which, according to IBM & AMD, will bring 40% increase in performance for AM2 and offer up to 4GHz Operation. Current Socket AM2 offerings (Not available yet):
XIII. Socket AM2 Speeds
(Clock Speed/L2 Cache/# of Cores)
Athlon 64 3500+ (2.333GHz/512KB/Single)
Athlon 64 3800+ (2.500GHz/512KB/Single)
Athlon 64 4000+ (2.500GHz/1MB/Single)
Athlon 64 X2 3800+ (Unknown/Dual)
Athlon 64 X2 4200+ (2.333GHz/512KB/Dual)
Athlon 64 X2 4400+ (Unknown/Dual)
Athlon 64 X2 4600+ (2.500GHz/512KB/Dual)
Athlon 64 X2 4800+ (2.500GHz/1MB/Dual)
Athlon 64 X2 5000+ (2.667GHz/512KB/Dual)
Athlon 64 X2 5200+ (Unknown)
Athlon 64 X2 5400+ (Unknown)
Athlon 64 FX-62 (2.800GHz/1MB/Dual)
Athlon 64 FX-64 (3.000GHz/1MB/Dual)
Athlon 64 FX-66 (Unknown)
XIV. Opteron 64 Processors [Sockets: 940]:
The first AMD Opteron 64 CPU's were releasd on April 22, 2003. These were 130nm processors running on Socket 940 ranging from 1.4GHz to 2.4GHz. They had 800MHz HyperTransport and 1MB Level 2 Cache each and ran on the Sledgehammer Core. Later, on February 14, 2005, 90nm versions were released, on the Venus (1xx Opterons) Troy (2xx Opterons) and Athens (8xx Opterons) cores. Each of these as well had 1MB of Level 2 Cache and added in SSE3 Instructions and a 1000MHz HyperTransport. The speeds ranged from 1.6GHz to 2.8GHz, there is a 3GHz version being released in the near future.
The Opteron differs from the Xeon because of it's implementation of HyperTransport. Opteron 64 CPU's perform far beyond Intel Xeon CPU's in Multi-Processor systems, due to CPU Coherency and Memory Access. AMD Opteron 64 systems can have as much as 22GB/s System Bandwidth and 11GB/s or more Memory Bandwidth, whereas an Intel Xeon CPU can have as much as 6.4GB/s System/Memory Bandwidth. Running under 64-bit Mode, the AMD Opteron 64 CPU is vastly superior to that of the Intel Xeon, especially in Linux x64 where performance can be up to 75% greater than Xeon. All AMD Opteron CPU's are 64-bit.
Dual-Core Opteron 64's compared to Dual-Core Xeon's, the Opteron 64's are again, vastly superior. This is due to the fact that just like Athlon 64 X2 processors, there is a "Crossbar" inside the CPU that allows the 2 cores to communicate to each other @ CPU Frequency. The Dual-Core Xeon CPU's have to use the Front Side Bus to communicate, and creates an even larger latency and performance impact. Current AMD Opteron s940 CPU's:
XV. Opteron 64 Speeds
130nm SledgeHammer
1.4GHz - 1MB L2 Cache - 800MHz HyperTransport
1.6GHz - 1MB L2 Cache - 800MHz HyperTransport
1.8GHz - 1MB L2 Cache - 800MHz HyperTransport
2.0GHz - 1MB L2 Cache - 800MHz HyperTransport
2.2GHz - 1MB L2 Cache - 800MHz HyperTransport
2.4GHz - 1MB L2 Cache - 800MHz HyperTransport
90nm Venus/Troy/Athens (1xx/2xx/8xx)
1.6GHz - 1MB L2 Cache - 1000MHz HyperTransport
1.8GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.0GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.2GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.4GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.6GHz - 1MB L2 Cache - 1000MHz HyperTransport
2.8GHz - 1MB L2 Cache - 1000MHz HyperTransport
90nm Denmark/Italy/Egypt (1xx/2xx/8xx) (Dual-Core)
1.6GHz - 2MB L2 Cache - 1000MHz HyperTransport
1.8GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.0GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.2GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.4GHz - 2MB L2 Cache - 1000MHz HyperTransport
2.6GHz - 2MB L2 Cache - 1000MHz HyperTransport
XVI. Opteron 64 Processors [Sockets: 1207]:
The Next Generation Opteron 64 Processors will be on Socket 1207, which s "Socket F". These Opteron 64 CPU's will bring Quad-Core CPU's, ability to use Quad Channel Memory, use DDR2 Memory, DDR3 in the future, as well as FB-DIMM Memory. The Socket F Opterons can also take advantage of Multiple Memory Controllers, as can Socket AM2, which will increase performance of Memory to each CPU.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XVII. Technical Information
Now for some technical stuff and explanations of clock frequencies and multpliers. This needs some tuning up, so anybody willing, please help to add some (don't start talking about TDP's or extremely technical things as this is geared towards newbies).
On every Athlon 64, X2, Pentium 4, and Pentium D processor (excluding Pentium Extreme Editions which runs at 266MHz, and some lower clocked P4's @ 133 and 166) there is a 200MHz Core Clock frequency, I like to refer to this as the "Crystal Clock". This is the speed at which the Processor truly runs at. There is something called "Clock Stepping" or a "Multiplier" as it is commonly known, where this allows the Internal Clock Frequency of the CPU to run at a much higher rate than the External Clock (Front Side Bus/HyperTransport).
Running a 200MHz Crystal Clock and a 10x Muliplier, means your CPU is running at 2000MHz (2GHz) Internal Speed, but this has no bearing on the speed of the Front Side Bus/HyperTransport. HyperTransport operates on Multiplier of it's own, up to 5x for 200MHz x5 = 1000MHz (1GHz) Operation, and HyperTransport is Full Duplex so it is 2000Mhz (2GHz) effective speed. Front Side Bus runs 200MHz Quad Pumped (All 4 sides of the Square Wave) which equals 800MHz effective speed since the Front Side Bus is Half Duplex.
With standard DDR1 Memory, it has 184 pins, and has very low latencies (2-2-2-5 and below) and provides up to 300MHz x2 Double Data Rate performance (with higher timings). On DDR2, it uses 240 pins, has higher latencies (3-3-3-8 and above) but runs on 333MHz x2 Double Data Rate and above, providing higher bandwidth than DDR1 at a price of higher latencies. DDR1 and DDR2 allow Dual Channel operation, meaning it has 2 lanes for information to pull from.
The Athlon 64 Socket 754 has a 64-bit Memory Controller, meaning it can only run Single Channel Operation, Socket 939 Athlon 64's as well as Socket 478 and Socket 775 use 128-bit Memory Controllers that allow Dual Channel Operation. Double Data Rate means it sends the pulse twice per clock cycle, or "Double Pumped", don't confuse this with Single & Dual Channel.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XVIII. Memory
I decided to add this in because I feel it is important for some people to know when purchasing a processor. The following information was obtained on Personal Experience and in no way guarantees that it will be what you experience or to be the exact truth, so please do not scrutinize unless I really messed up (I might have, heh).
Memory Latency & Speed: I have found that on an Athlon 64 system, comparing latencies such as 2-2-2-5 to 3-3-3-8, that the performance hit is not that great. I take this to be because of the Onboard Memory Controller. Being that the OMC reduces latencies tremendously already, higher latencies will not effect the A64 as much. On a Pentium 4 processor, I have found that higher latencies, will be a greater impact on performance (Not extravagent, but noticeable) and this is because the memory transmits along the Front Side Bus back to the CPU, vs. a direct connection back to the CPU @ RAM Speed (DDR400, DDR433, etc.).
When the Memory Controller (Be it OMC in A64 or NB in P4) goes to access something in memory, it first sends out a command (I like to refer to this as the RFI, Request For Information) to obtain data, this is where your latencies come from. First, a bank of memory is accessed, known as Chip Select (CS) and this is where your "Command Rate" comes into play. Your "Command Rate" is how many cycles it takes to latch onto a physical chip, 1T means after the 1st cycle, 2T meaning after the 2nd cycle. First, the row (RAS) is found to where the data is in, then there is a wait for the RAM to actually access the row (RCD). Then, it defines the column the data is in (CAS), and then it waits for the data to be sent to the pins, and out back to the CPU. After that is done, it precharges the row to be used again (RP). These are defined by your 2-2-2-5 (CAS-RCD-RP-RAS, sometimes a "t" previews the letters, this "t" stands for "time").
A few things to note: latencies of .5 (e.g., 2.5) can only be obtained on Double Data Rate memory, where 2 parts of each clock cycle are used. There is a built-in SPD (Serial Presence Detect) that you may notice with new RAM, it may set your timings and volts to something other than advertised. This is to ensure stability on 1st boot. When you attempt to overclock memory, it is best practice to relax the timings (e.g., go from 2-2-2-5 to 3-3-3-8 @ higher frequencies) this is to ensure stability, also note you may have to increase voltage to the DIMM to remain stable at higher frequencies.
In memory, comparing Single Channel to Dual Channel, the CPU is a limiting factor. What I mean is that for each pin on the RAM, e.g., 184-Pin DDR1, that requires 1 pin from the CPU as well. In addition to that, for each channel, it requres double that of it's Single Channel counterpart. This means to run Dual Channel DDR1 w/ 184 Pins, it would require 368 Pins (184x2) from the CPU and this is why the Socket 754 processors do not support Dual Channel, as well as because the Memory Controller is 64-bit whereas a 128-bit Memory Controller (such as on the s939 and s478/s775) is required for Dual Channel.
Memory - Overclocking
The following is based on Personal Experience and what I have seen and been told of by other friends, colleagues, and abroad. The results I present may not be what you experience.
Undoubtably, the best overclocking RAM you can buy is OCZ...pure and simple. OCZ for AMD & Intel has proven to be the best overclockers for RAM, period. OCZ even took it to a new level, with their VX Series, known as eXtended Voltage, you can run the tightest timings (2-2-2-5) @ DDR500 speeds and above, w/ lots 'o' volts! The following is a list of the companies for which the RAM is the best and their best, as well as a overall best overclocking RAM at the bottom:
OCZ
Crucial
Corsair
G.Skill
Mushkin
Patriot
GigaRAM
Kingston
PNY
Viking
OCZ - Platinum Series
Crucial - Ballistix & Tracer Series
Corsair - XMS Series
G.Skill - Extreme Series
Mushkin - Redline Series
The overall best overclocking memory is the OCZ Platinum series, overclocking to DDR600 and beyond is amazing feat's for any RAM!
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XIX. Cache Comparison
This is a comparison of 512KB, 1MB, and 2MB of Level 2 Cache. This is based off of Personal Experience and may not be what you experience.
Some people like to say more cache, more performance, but I don't think that is always true. In most desktop environments, such as Gaming, I can see little to no increase in performance w/ 1MB of Cache over 512KB, and 2MB seems to even slow down compared to 512KB in games. It is true that a CPU can access 512KB of Cache in less time, but there is less total cache. Whereas 1MB or more, it takes a little longer to sort through, but there is alot more of it present. I find that in alot of applications that don't branch much (change what they're doing), more cache seems to be of benefit there.
If you're a solid gamer, I recommend simply an A64 CPU w/ 512KB Cache, such as the 3200+, or P4 530 w/ 1MB Cache. The less cache in games will give you better performance, in my experience. AMD says that 512KB Cache provides 200MHz Increase in performance, which is why you have a 3500+ w/ 512KB and a 3700+ w/ 1MB and they're both 2.2GHz. Intel seems to think there is an even greater increase in performance, but this is just due to the bottlenecked Front Side Bus in my opinion.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XX. Future Technologies
This section is about the future technologies AMD and Intel are planning to implement into their Desktop and Server lines of CPU's. All of it may be of speculative nature from various websites and myself, I will not say "This tech will pwn the other!" or say any bias statement, but rather what it is targeted to do, nothing more.
Intel Future Technologies
Intel plans to increase the FSB on their future "Conroe" and "Woodcrest" CPU's up to 1066MHz (266MHz Quad Pumped) and 1333MHz (333MHz Quad Pumped). This will alliviate the need to place the Memory Controller On-Die for a period of time. The 1066MHz for "Conroe" will be in the mainstream CPU's with 1 version, the "Extreme Edition" being with a 1333MHz Front Side Bus & 3.33GHz Internal Clock. Another technology they are developing is called DIB (Dual Independent Bus). This will give each CPU a Front Side Bus to the Northbridge in a Dual-Core system. This will not, however, increase Memory Bandwidth significantly due to there still being only a single Memory Controller on the Northbridge.
Another technology they are developing, though it might be facing some delays, is known as CSI (Common System Interface). This is the challenger to AMD's HyperTransport Technology. There isn't a whole lot known about CSI right now, it's a Serial Interconnect like HyperTransport, but initially, according to some sources, CSI was designed to beat HyperTransport, but it appears that that plan has failed. CSI seems to be the equivilent to a 1600MHz FSB. The first implementation of CSI will be in the core "Tukwila" that is Quad-Core. CSI is also planned for the Intel Xeon processor, it will be in the Quad-Core "Whitefield" core. The Xeon version w/ CSI will come out in 2009 while the Itantium version should be out in 2008.
There is another architecture Intel is developing, known as Nehalem, there is absolutely nothing known about this as far as I can tell, other than it will be in the Xeon core "Gainestown" that will be Quad-Core, come post-2010, and start at 3GHz.
AMD Future Technologies
AMD is quite the busy-bee, they are working on several new technologies, most notably, HyperTransport 3.0 and core revisions. HyperTransport 3.0 should scale beyond 1600MHz (3200MHz effective) and use 32-bit or 64-bit data widths. There is also HyperTransport 4.0 being developed, but nothing is official as of yet. Another technology is Direct Connect Architecture 2.0. This tech, set to be released in 2008, will be inplemented to have greater than 8-CPU servers. A revison, only codename is "K8L", but that term was coined by TheInquirer.net, is set to be released in 2006, and contains mainly double number of FPU's which should bring 1.5x the performance of K8.
AMD is also working on a tool that will allow multiple CPU's to work on a Single Thread, it's kind of like "Reverse HyperThreading". It's to be implemented in the next-generation of AMD CPU's.
Others such as the recently licensed Z-RAM technology, should bring 20% to 50% improvements to the current K8 architecture, according to some sources. AMD is also working on putting Level 3 Cache into the Opteron 64 Socket 1207 "F" platform, the L3 may be Z-RAM or another form of Static RAM. DDR2 is to be brought into Socket AM2 and Socket 1207 Opteron 64 servers, as well as DDR3 in 2007. AMD plans to bring 65nm CPU's at the end of 2006 and 45nm sometime in 2008. There are also plans to bring a new architecture, K10, in 2008. This will, according to AMD, bring 10GHz Operation, better Branch Prediction, better FPU performance, and allow for Multi-Multi-CPU systems.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXI. What's That?
This section explains some of those things you see such as "ECC & Registered RAM", "Level 2 Cache", "RAM Timings", etc. This does not get technical so I can maintain the original idea of being everybody-friendly. If something sounds so non-technical that it sounds wrong, please make sure to tell me so I can fix it. I will continue to update this as necessary to stay up to date
32-bit & 64-bit computing - 32-bit and 64-bit refer to the size of the registers inside the CPU and the data-size of the information processed.
Bandwidth - Bandwidth (in RAM, Video Cards, Front Side Bus, etc.) are found in a mathematical equation, it is: Frequency x Data Width / 8 = Bandwidth. This means, a Frequency of 1000MHz (on say, a Video Card) and a 256-bit Memory Interface (the physical size of how wide the bus is) will give you 32GB/s Memory Bandwidth, 1000x256/8=32000MB/s.
Cache - Cache is used to store often-used information or commands. In the case of Level 2 Cache in a CPU, it holds frequently used data and instructions for where in RAM the data can be found.
Coherency - Coherency refers to the communication between 2 devices or components that are to act in symmetry. In CPU's, it's the communication beween CPU's to know what the other is doing and what it is in it's RAM/Cache. AMD Opteron 64 based-computers have extremely more efficient coherency than Xeon's, as well as the Athlon X2 has much better coherency than a Pentium D due to it's Crossbar.
CSI - CSI (Common System Interface) is a Serial Point-To-Point protocol designed by Intel to counter Hypertransport. Current sources put CSI comparable to a 1600MHz Front Side Bus.
Crossbar - The Crossbar in a Dual-Core AMD64 CPU is the component that allows the 2 cores to communicate to each other @ CPU speed, rather than having to use the Front Side Bus as in the Xeon or Pentium D.
ECC - ECC stands for Error Correcting Code. ECC is used for mainly in servers, although some Socket 939 AMD Athlon 64 motherboards support ECC because the Memory Controller has ECC as well. ECC corrects some errors in memory where 111% integrity is needed. ECC works by adding 1 bit of parity per 8 bits, that means it can fix some 1-bit errors, it will only detect 2-bit errors but won't fix them.
HyperTransport - A Serial Point-To-Point protocol designed by AMD & Founders for High Bandwidth & Low Latency communication between devices. This is currently implemented in many networking devices (such as switches, routers, and such) and also in certain computer Microprocessors, namely the Athlon 64 and Opteron 64. The fastest HyperTransport is 32-bit x 32-bit 1400MHz (2800MHz Effective) and provides 22.4GB/s bandwidth.
NUMA - NUMA (Non-Uniform Memory Access) allows multiple memory banks to be shared among multiple CPU's in a Multi-Processor system.
Register - A CPU register is something that stores currently used information for the CPU to process and send out.
Registered - Registered RAM is a chip on the module that acts kinda like another Memory Controller. The Memory Controller only sees the chip, and the Memory Controller sends commands to the Chip and that chip disperses and controls the module.
TDP - TDP (Standing for Thermal Design Power or Point) is the maximum heat the chip will dissipate (via any method) without reaching the maximum temp for the CPU. TDP does not say how much power the CPU will draw, but just the maximum power it will draw running @ a "real" load, which can be far from real-world use.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXII. Motherboards
This section is about the overall best motherboards for stability and overclocking for the End-User, it is denoted by brand, as well as a specific board. This also depicts some of the best features and the best overall motherboard based on Features, Price, Overclocking, and Stability. Note: The following is obtained by either Personal Experience, or by me speaking with others about the boards. You may experience different results than what I present.
Best Overclocking Motherboard (AMD): DFI - LanParty nF4 SLI-DR Expert Edition
The Expert Edition motherboard has been one of the best overclocking boards, I have personally taken this board past 300MHz with an Opteron CPU using minimal volts. This board has been known to overclock up to 350MHz and beyond with ease, which can clearly make it the best board. DFI itself is known to be the best for overclockers, they always give the gamer and enthusiast what they want and this board doesn't dissapoint. The price tags for some DFI boards can leave you breathless, but once you take the baby home and feed it, you're going to wish you had 2!
Best Overclocking Motherboard (Intel): Gigabyte - Gigabyte GA-G1975X
This board is a really nice overclocker, nice layout, and has a nice ergonomic design to it. This board has been overclocked to over 4.5GHz with a P4 and I wouldn't doubt 5GHz or above is achievable. The price of this board, and the fact it lacks SLI, is something to note, but if you're a hardcore enthusiast, you go the road I go, Single Card Solutions and OC To The Max!
Best Stable Motherboard (AMD & Intel): Tyan - Server & Workstation Boards
Tyan may not be the first choice for gamers or enthusiasts, but it is certaintly the talk of the town when it comes to Server and Workstation motherboards. Tyan has a proven record of reliability in the Workstation and Server area, and I would personally say through experience that one of their best boards is the Tyan K8WE S2985. If you're looking for a stable solution and don't plan to overclock, look no further than Tyan for your next build!
Best Priced Motherboard (AMD): ASRock & EpoX - Variants
ASRock & EpoX are showing their contendership with the likes of ASUS, DFI, and other big-name builders, and they're accomplishing this with prices that show they are a force to be reckoned with. The boards from these 2 makers are exceptionally well in their price/performance and even can contend with high-end boards such as LanParty's if given the right builder.
Best Priced Motherboard (Intel): ASUS - Variants
ASUS is pretty much the best motherboard builder for Intel that offers the best prices. ASUS may not be so hot with AMD's performance motherboards, but on the Intel Camp, ASUS is a fair trader in performance/price battle against the top Gigabyte and Intel offerings.
Overall Best Motherboard - Features, Etc. (AMD): ASUS A8N-SLI Premium
This board has it all....SLI, Sata 2 3Gb/s, PATA RAID, 8 SATA Connectors, 3 PCI Slots, Dual Gigabit...this board is one of the best boards for all around performance, price, and features for the AMD motherboard options. This board may not be touted as the best overclocker or best stability, but it's feature richness means it can contend anyday with all the other boards, and it's at a fairly reasonable price considering its features.
Overall Best Motherboard - Features, Etc. (Intel): ASUS P5N32-SLI
This board has alot of features, 6 SATA that are 3Gb/s, SLI, support for 16GB of RAM, Dual Gigabit, 2 PCI Slots, 5 PCI-E Slots of various speeds, this board is the most feature-rich board for the Intel platform, this does come at a hefty price, but it's got tons of features to make it worth while.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXIII. Naming Scheme: What They Mean:
This section is to familiarize users with the naming schemes that the b@!stards @ AMD & Intel keep coming up with to cunfubble us!
AMD
Opteron 64 Series (All have 1MB Cache):
Opteron 64 x44 - 1800MHz - Single Core
Opteron 64 x46 - 2000MHz - Single Core
Opteron 64 x48 - 2200MHz - Single Core
Opteron 64 x50 - 2400MHz - Single Core
Opteron 64 x52 - 2600MHz - Single Core
Opteron 64 x54 - 2800MHz - Single Core
Opteron 64 x56 - 3000MHz - Single Core
Opteron 64 x65 - 1800MHz - Dual Core
Opteron 64 x70 - 2000MHz - Dual Core
Opteron 64 x75 - 2200MHz - Dual Core
Opteron 64 x80 - 2400MHz - Dual Core
Opteron 64 x85 - 2600MHz - Dual Core
Opteron 64 x90 - 2800MHz - Dual Core
Athlon 64 Series:
Athlon 64 2800+ - 1800MHz (512KB Cache) - Socket 754
Athlon 64 3000+ - 2000MHz (512KB Cache) - Socket 754 & 1800MHz (512KB Cache) Socket 939
Athlon 64 3200+ - 2200MHz (512KB Cache)/2000MHz (1MB Cache) Socket 754 & 2000MHz (512KB Cache) Socket 939
Athlon 64 3400+ - 2400MHz (512KB Cache) - Socket 754
Athlon 64 3500+ - 2200MHz (512KB Cache) - Socket 939
Athlon 64 3700+ - 2400MHz (1MB Cache) - Socket 754 & 2200MHz (1MB Cache) - Socket 939
Athlon 64 3800+ - 2400MHz (512KB Cache) - Socket 939
Athlon 64 4000+ - 2400MHz (1MB Cache) - Socket 939
Athlon 64 X2 Series:
Athlon 64 X2 3800+ - (512KB Cache Per Core) - 2000MHz - Socket 939
Athlon 64 X2 4200+ - (512KB Cache Per Core) - 2200MHz - Socket 939
Athlon 64 X2 4400+ - (1MB Cache Per Core) - 2200MHz - Socket 939
Athlon 64 X2 4600+ - (512KB Cache Per Core) - 2400MHz - Socket 939
Athlon 64 X2 4800+ - (1MB Cache Per Core) - 2400MHz - Socket 939
Intel
Pentium 4 Series:
Pentium 4 x3x - (512KB, 1MB or 2MB Cache) - 3000MHz - Socket 478/T
Pentium 4 x4x - (512KB, 1MB or 2MB Cache) - 3200MHz - Socket 478/T
Pentium 4 x5x - (512KB, 1MB or 2MB Cache) - 3400MHz - Socket 478/T
Pentium 4 x6x - (512KB, 1MB or 2MB Cache) - 3600MHz - Socket 478/T
Pentium 4 x7x - (512KB, 1MB or 2MB Cache) - 3800MHz - Socket 478/T
Pentium D Series:
Pentium D x2x - (1MB or 2MB Cache Per Core) - 2800MHz - Socket T
Pentium D x3x - (1MB or 2MB Cache Per Core) - 3000MHz - Socket T
Pentium D x4x - (1MB or 2MB Cache Per Core) - 3200MHz - Socket T
Pentium D x5x - (1MB or 2MB Cache Per Core) - 3400MHz - Socket T
Pentium D x55 - (2MB Cache Per Core) - 3460MHz - Socket T
Pentium D x65 - (2MB Cache Per Core) - 3733MHz - Socket T
I have lacked Xeon as there are simply WAY too many Xeon changes to note on.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
XXIIV. Past Changes:
4-12-06 Changes: Added a "Motherboard" section below to denote the best overclockers, stability, and features. Added specifics to the "Memory" section as to the best overclocking memory and best stability, etc. Added "Naming Schemes: What They Mean" section, displays all the namings of Opteron 64's (e.g., Opteron 170), Pentium 4's (e.g., Pentium 4 530) so one can deduce what they are without reading the jumble in advertising.
4-10-06 Changes: Added more information to "Conroe" and "AM2"
3-30-06 Changes: Added the "Pentium D 965 EE" Processor. Also fixed spellings and added more information. Also fixed some things that didn't make sense.... Added "HyperTransport", "CSI", and alphabetized the "What's That?" section.
3-27-06 Changes: Added another term to the "What's That?" section. Added a Table of Contents and bulleted sections for better viewing. Changed name to something drastic...hope it works ;
3-26-06 Changes: Fixed "Conroe" speeds and added "Opteron 64" speeds. Added more info into the thread and cleaned up some parts. Changed name again....
3-23-06 Changes: Added a "What's That?" section below "Future Technologies" that explains some Terms and Technologies that are inside current CPU's from AMD & Intel and others. Touched-up some information and cleaned-up a little. Changed the view of this from the 2 CPU's to overall computer hardware.
3-19-06 Changes: Added "Conroe" and Socket AM2 clock speeds. Made minor changes to the information and cleaned up some parts. Removed pricings on CPU recommendations. Also changed the title to a more passive presentation.
3-16-06 Changes: Added a "Future Technologies" section below "Cache Comparison".
3-12-06 Changes: Added preliminary "Conroe", "Socket 1207", "Socket AM2", "Dempsey" information. Also changed the titles of each section to a more ergonomic view.
3-5-06 Changes: Added the Intel Pentium D 805.
3-2-06 Changes: Added Opteron 64 and Xeon CPU information.
"Cache Comparison" added below "Memory". Changed some information and added more based on user posts.
3-1-06 Changes: Added more information to "Memory" section below and edited information in main thread.
2-25-06 Changes: Changed pricings, they reflect multiple vendors now. As well as some information gathered from fellow posters. Also changed some wordings, information and pretext to the thread, as well as a "Memory" section added below. Also noted on "Cedar Mill" desktop Pentium 4 processor core and edited more information.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Thank you again for reading this, and make sure to post if anything is wrong or with suggestions, peace.
~~Mad Mod Mike, pimpin' the world 1 rig at a time
).
Facebook
Twitter
Instagram