Operation of AMD & Intel CPU's

[DCB_AU]

OK, AMD CPU's are better for gaming and Intel for video compression, etc.

Question to technical persons -

briefly, what is the reason (operation) of the AMD CPU ie Athlon 64 3200+ 512k L2, 939 Venice, being faster in gaming but slower in video compression when compared to Intel Prescott 3.2 GHZ, 2mg L2, 775.

(this is not one of those AMD vs Intel questions)

Thanks,

 

[MadModMike]

OK, AMD CPU's are better for gaming and Intel for video compression, etc.

Question to technical persons -

briefly, what is the reason (operation) of the AMD CPU ie Athlon 64 3200+ 512k L2, 939 Venice, being faster in gaming but slower in video compression when compared to Intel Prescott 3.2 GHZ, 2mg L2, 775.

(this is not one of those AMD vs Intel questions)

Thanks,

One reason is due to the longer staged pipeline in the Pentium 4. If a Pentium 4 gets fed direct streams of data and keeps doing the same thing, it can really crank out speed. But in Gaming, where things are constantly changing, the shorter pipelined Athlon 64 excels because it doesn't take as long to flush out the pipeline. Every time a program branches (changes what it does), it has to flush the pipeline, and this is where the 31-stage pipeline in P4's starts to drag for gaming and office applications. And IMO, I believe the larger Level 1 Cache to be another reason for better performance in Gaming and overall system performance.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[cisco]

The reason an AMD processor at a lower clock speed out performs an Intel is because the AMD uses an integrated memory controller and instead of a FSB on the northbridge of the motherboard, AMD uses a HTT bus which they developed which allows the communication between the ram and the processor to run in full duplex and at speeds of 2000mhz vs half duplex 800FSB on the Intel. And that is the main reason AMDs out perform Intels in gaming and other areas. As for video that gap has closed and in some case has gone to AMD as well. But the main reason Intels were so good at video compression is because it is a very clock cycle intensive process meaning the higher the mhz the better the result.

 

[MadModMike]

The reason an AMD processor at a lower clock speed out performs an Intel is because the AMD uses an integrated memory controller and instead of a FSB on the northbridge of the motherboard, AMD uses a HTT bus which they developed which allows the communication between the ram and the processor to run in full duplex and at speeds of 2000mhz vs half duplex 800FSB on the Intel. And that is the main reason AMDs out perform Intels in gaming and other areas. As for video that gap has closed and in some case has gone to AMD as well. But the main reason Intels were so good at video compression is because it is a very clock cycle intensive process meaning the higher the mhz the better the result.

Actually, that is incorrect. The Integrated Memory Controller actually creates a latency in gaming, because it has to go to the Northbridge, than to the CPU, than to RAM, vs. the Pentium 4 that goes Northbridge to RAM. So in theory, the Athlon 64 should be inferior in gaming to the Pentium 4, but due to the FSB and 31-stage pipeline, this is untrue. Another reason the P4 excels in Video Compression is due to the 2MB Level 2 cache.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[cisco]

HTT is a seperate bus that does not go to the northbridge it goes directly to the ram. Intel still goes thru the northbridge using traditional FSB

 

[MadModMike]

HTT is a seperate bus that does not go to the northbridge it goes directly to the ram. Intel still goes thru the northbridge using traditional FSB

You are severly misinformed, but luckily, I am here to enlighten you . HyperTransport goes to the Northbridge, HyperTransport is the replacement for the Front Side Bus in an Athlon 64 and Opteron 64 CPU. HyperTransport goes to the RAM as well, but a little different way. In Opteron 64 CPU's (And I submise Athlon 64 as well) the CPU goes to the RAM @ CPU Frequency, meaning if you have a 2.2GHz CPU, that is how fast you GO TO the RAM. The Athlon 64's and Opteron 64's than have a direct link back from RAM to Memory Controller @ RAM Speed. This is why AMD is better suited for DDR2's Bandwidth than Intel's FSB because the FSB @ 800MHz is 6.4GB/s and 1066MHz is 8.1GB/s, DDR2 is over 10GB/s on 667MHz. Intel Pentium 4's use the Front Side Bus to the Northbridge, and that is where the Memory Controller resides, in the Northbridge.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[cisco]

What I meant to say it runs to the northbridge just not in the traditional sense since the memory controller is integrated. Although AMD can use the bandwidth of ddr2 you will see little improvement because AMD doesn't like loose timings in ram. I noticed in another thread you wrote that they are not effected as much by slow timings when in fact they are. As you see with AM2 even though the ram tested was 667mhz because of the slack timings there was no improvement in performance. And that will be the case until timings come down on DDR2. For example, if you can run your memory at 215 MHz at CAS 2 2:2:6 and you loosen your timings to CAS 2.5 3:3:7, you may need 230-235MHz memory speed to make up for the higher ram timings. The performance hit from 400 to 333 ddr is only 3-6%.

Intel system bus- CPU to bus interface=6,400mb/sec (64x800÷8) for the 800mhz FSB version.

AMD Athlon64-cpu to HT interconnect(for 939 chips)32bit bus(16bit up+16bit down bi-directional full duplex) x 2000mhz effective/8(bit to byte conversion=8,000mb/sec.

Although AMD utilizes ram bandwidth better they suffer more from slack timings than ddr speed.

 

[AntiHax0r]

The reason an AMD processor at a lower clock speed out performs an Intel is because the AMD uses an integrated memory controller and instead of a FSB on the northbridge of the motherboard, AMD uses a HTT bus which they developed which allows the communication between the ram and the processor to run in full duplex and at speeds of 2000mhz vs half duplex 800FSB on the Intel. And that is the main reason AMDs out perform Intels in gaming and other areas. As for video that gap has closed and in some case has gone to AMD as well. But the main reason Intels were so good at video compression is because it is a very clock cycle intensive process meaning the higher the mhz the better the result.

AMD left the MHZ race long ago before they had intergrated memory controllers, If you remember AthlonXP 1500+(about 5-6 years ago?) and up, they at the time were still based on northbridge FSB. It was just a matter of cpu architecture, instruction sets, etc...

 

[cisco]

Yes they left a race they couldn't win and addressed the bottle neck of the motherboard. And even though the older AMDs performed well they still lost in several key benchmarks to Intel. Since they changed how the processor communicates with the ram they have started taking over almost all benchmarks. This is not meant to be a knock on AMD just the opposite. They were smart enough to address the real problem in computer performance. Intel is not going to the integrated controller until the end of the decade. To quote a line from Seinfeld, "the Chinese know about the fork yet they stick with the chop sticks".

 

[KBM]

IMO, Mad Mod Mike's first reply was the perfect answer for the question.

There are lot of people here who gets confused between, HT speed, Memory speed and system speed and LDT and what ever..

AMD CPUs basically run off of a base clock of 200 MHz generated by
the clock generator. (This is what we tweak for overclocking)

There is a CPU multiplier (PLL) that multiplies this base clock to derive
the CPU's operating frequency. For example, 10x multiplier in 3200+
gives the clock speed of 2 GHz. Only the CPU core operates at this frequency.

Memory bus normally operates at this same base frequency of 200 MHz.
DDR (Double Data Rate) makes use of both rising and falling edges of
the clock for data transmission. That's how we get DDR400 (200 MHz x2).
Memory bus doesn't operate at CPU's frequency. That is a wrong concept.
However, there is a divider that can be set in the memory controller ,
so that lower speed memories like DDR333 can be used.

HT Bus operates with a multiplier of 5 over the base clock. so, HT Bus
speed is 5 x 200 = 1GHz.

While accessing external data, CPUs perform 2 basic types of I/O operations. Memory and I/O to ports. in case of AMD, these are separated
into Memory Bus and HT Bus (People who know about microcontrolers
will appreciate this difference)

In case of Intel, all the fundamentals remain the same for Base clock
and CPU multiplier. There are no separate buses for memory and I/O
and every thing must go in and out via FSB, which is quad pumped
from the base clock (200 MHz x 4 = 800 MHz).

I think the reason every one gets confused about HT is because CPUz calls 200 MHz as the HT frequency. This is wrong. This is the system frequency or the Base frequency.

I hope you enjoyed my first post in TH.

Thank you

Mohan

 

[MadModMike]

What I meant to say it runs to the northbridge just not in the traditional sense since the memory controller is integrated. Although AMD can use the bandwidth of ddr2 you will see little improvement because AMD doesn't like loose timings in ram. I noticed in another thread you wrote that they are not effected as much by slow timings when in fact they are. As you see with AM2 even though the ram tested was 667mhz because of the slack timings there was no improvement in performance. And that will be the case until timings come down on DDR2. For example, if you can run your memory at 215 MHz at CAS 2 2:2:6 and you loosen your timings to CAS 2.5 3:3:7, you may need 230-235MHz memory speed to make up for the higher ram timings. The performance hit from 400 to 333 ddr is only 3-6%.

Intel system bus- CPU to bus interface=6,400mb/sec (64x800÷8) for the 800mhz FSB version.

AMD Athlon64-cpu to HT interconnect(for 939 chips)32bit bus(16bit up+16bit down bi-directional full duplex) x 2000mhz effective/8(bit to byte conversion=8,000mb/sec.

Although AMD utilizes ram bandwidth better they suffer more from slack timings than ddr speed.

Cisco, I don't mean to insult you, but you are severly mis-informed alot.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[K8MAN]

Any latency benchmark will contradict everything you've said about the onboard mem controller slowing things down. There is a reason why the first gen 754 CH was 30% faster clock/clock than AXP while retaining somewhere in the neighborhood of 90% of the AXP core.

 

[MadModMike]

Any latency benchmark will contradict everything you've said about the onboard mem controller slowing things down. There is a reason why the first gen 754 CH was 30% faster clock/clock than AXP while retaining somewhere in the neighborhood of 90% of the AXP core.

I've already said that in real world performance, the onboard memory controller doesn't impact gaming performance, which is all I am talking about. But if you knew anyhing about the physical architecture, you'd know it takes 2 hops to get to RAM (Northbridge to CPU and CPU to RAM) on an A64 where a P4 is 1 (Northbridge to RAM). In theory, the P4 should be faster, but of course, it is not.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[Scout]

Are you talkin' about the video card being one step further separated from the memory - as in through the CPU - versus the P4? I guess that's true, but that's the first time I've ever heard anyone assert that as a negative of the architecture and I don't think it's right...

 

[MadModMike]

Are you talkin' about the video card being one step further separated from the memory - as in through the CPU - versus the P4? I guess that's true, but that's the first time I've ever heard anyone assert that as a negative of the architecture and I don't think it's right...

That is what I am referring to, and on paper, technically, that is a negative. But throw in HyperTransport, and that negates the latency.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[K8MAN]

Any latency benchmark will contradict everything you've said about the onboard mem controller slowing things down. There is a reason why the first gen 754 CH was 30% faster clock/clock than AXP while retaining somewhere in the neighborhood of 90% of the AXP core.

I've already said that in real world performance, the onboard memory controller doesn't impact gaming performance, which is all I am talking about. But if you knew anyhing about the physical architecture, you'd know it takes 2 hops to get to RAM (Northbridge to CPU and CPU to RAM) on an A64 where a P4 is 1 (Northbridge to RAM). In theory, the P4 should be faster, but of course, it is not.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
Forgive me if I'm wrong but isn't the entire northbridge on die with the A64? A64's mobo's usually only have a southbridge so where would the RAM have to to go other than through the memory controller?

 

[psyno]

Actually, that is incorrect. The Integrated Memory Controller actually creates a latency in gaming, because it has to go to the Northbridge, than to the CPU, than to RAM, vs. the Pentium 4 that goes Northbridge to RAM. So in theory, the Athlon 64 should be inferior in gaming to the Pentium 4, but due to the FSB and 31-stage pipeline, this is untrue. Another reason the P4 excels in Video Compression is due to the 2MB Level 2 cache.

That wouldn't have been my theory, considering how often the graphics card goes to main memory compared to how often the CPU does...

I don't dispute that there are more hops and thus there is more potential latency from the graphics card to main memory for AMD than Intel, but the design principle here is to make the common case fast, and the CPU going to main memory is by far more common.

 

[psyno]

Forgive me if I'm wrong but isn't the entire northbridge on die with the A64? A64's mobo's usually only have a southbridge so where would the RAM have to to go other than through the memory controller?

You are somewhate (but not totally) wrong, and you are forgiven. 😀

Only the memory controller has been moved from the northbridge to the die. The northbridge still exists... but it has little purpose, expcept to serve graphics cards. Some chipsets have combined the north and south bridges into one chip.

Since not all the old functionality of the northbridge moved to the die, IMHO opinion it would be incorrect to say that the northbridge moved to the die. However, it may or may not be correct to say that after the removal of the memory controller, the northbridge no longer exists, or is vestigial at best.

Sorry for the double post.

 

[K8MAN]

Forgive me if I'm wrong but isn't the entire northbridge on die with the A64? A64's mobo's usually only have a southbridge so where would the RAM have to to go other than through the memory controller?

You are somewhate (but not totally) wrong, and you are forgiven. 😀

Only the memory controller has been moved from the northbridge to the die. The northbridge still exists... but it has little purpose, expcept to serve graphics cards. Some chipsets have combined the north and south bridges into one chip.

Since not all the old functionality of the northbridge moved to the die, IMHO opinion it would be incorrect to say that the northbridge moved to the die. However, it may or may not be correct to say that after the removal of the memory controller, the northbridge no longer exists, or is vestigial at best.

Sorry for the double post.
Thanx for clearing that up Triple M wasnt being very clear

 

[MadModMike]

Actually, that is incorrect. The Integrated Memory Controller actually creates a latency in gaming, because it has to go to the Northbridge, than to the CPU, than to RAM, vs. the Pentium 4 that goes Northbridge to RAM. So in theory, the Athlon 64 should be inferior in gaming to the Pentium 4, but due to the FSB and 31-stage pipeline, this is untrue. Another reason the P4 excels in Video Compression is due to the 2MB Level 2 cache.

That wouldn't have been my theory, considering how often the graphics card goes to main memory compared to how often the CPU does...

I don't dispute that there are more hops and thus there is more potential latency from the graphics card to main memory for AMD than Intel, but the design principle here is to make the common case fast, and the CPU going to main memory is by far more common.

It's not technically a theory, but rather a fact. I wasn't trying to imply anything with the Latency statement I made, but simply informing Cisco about the Athlon 64.

Thanx for clearing that up Triple M wasnt being very clear

I was in bed when you asked your question, I obviously couldn't respond to it.

The northbridge still exists... but it has little purpose

It has little purpose currently on Athlon 64 systems, but if you move into Opteron 64 Dual CPU Workstation/Server boards, alot of them have, as you said, integrated the Southbridge components into the Northbridge. But there are alot of other uses on the NB in Opteron 64 servers, such as extra 1GbE/10GbE, PCI-X Bridge, SCSI Controllers, etc. And alot of 4-Way Opteron 64 boards do away with traditional Northbridge's all together in favor or AMD I/O Hub+PCI-X Tunnel. Also remember that the BIOS is connected to the NB. There are also some Opteron 64 boards that go with 2 Northbridges, 1 per CPU, and Multiple AMD I/O Hubs.

~~Mad Mod Mike, pimpin' the world 1 rig at a tme

 

[ATIisGhostlyblue]

pretty apparent "mad mike" does not know what he is talking about on this issue

 

[MadModMike]

pretty apparent "mad mike" does not know what he is talking about on this issue

And how do you figure this?

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[K8MAN]

So when AMD stick the PCI-E controller onboard next year we'll get the equivilant that bottleneck will be removed. Wonder how much of an improvement it will bring us...

 

[MarkG]

I suspect the real reason that AMD ran games faster than Intel well before they had an on-chip memory controller is the sucky FPU on the Pentium-4. P4 floating point performance was significantly better than Athlon if you used SSE instructions, but well behind if you didn't: and I'd imagine that most games used the old-style FPU rather than write different code for Intel and AMD chips.

 

[MadModMike]

So when AMD stick the PCI-E controller onboard next year we'll get the equivilant that bottleneck will be removed. Wonder how much of an improvement it will bring us...

I'd say not much at all.
As mentioned ealier, HyperTransport solves bascially everything and newer faster revised HTT is not too far away.
That's true.

Like Mike said earlier,
Intel P4s have longer steppings (or staging pipelines) and they're able to achieve higher clock speeds. AMDs have shorter steppings (or staging pipelines) and they're not able to achieve such high clock speeds. But of course you can overclock them.
Steppings (or staging pipelines) are the path of the process, to put it in easy terms.
Also, AMDs have bigger IPC (instructions per clock) so they can do more work per cycle.

The pipeline is a big reason like I said why Intel Pentium 4's don't perform as well, because NOP's (Non OPeration, blank data basically) have to be put into the Pipeline to prevent Hazzards (Errors in the Pipeline) and in games, the Pentium 4 pipeline is basically an NOP pipeline, lol.

~~Mad Mod Mike, pimpin' the world 1 rig at a time