Athlon 1.2G 266 vs. 1.3G 200

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

slvr_phoenix

Splendid
Dec 31, 2007
6,223
0
25,780
0
Good rebuttal, but I do of course have to point out the flaws in it. :)

The P4 <i>is</i> running on a 100MHz FSB. FSB refers entirely to frequency. Just because a system is capable of performing operations both at the rise and the fall of the wave (or in the P4's case, the peak, the trough, and the two middle points, I'd assume) does not mean that the frequency of the wave is in any way different. The FSB itself is still 100MHz or 133MHz because it's measured in frequency, not in bandwidth. Therefore there is <i>no</i> such thing as a 266MHz FSB. (At least not yet.)

The P4 also is entirely different from the Athlon in one major respect, the P4 performs I/O operations with the chipset on each pumping. So it is <i>externally</i> quad-pumped.

The Athlon on the other hand only performs internal calculations on each pumping and I/O only once per FSB cycle. So it is only <i>internally</i> double-pumped. It's I/O with the chipset is still only at single-pumped 133MHz speed.

No, I don't have any engineering blueprints or anything to prove this, but it is very evident in the performance of the systems. If you read my explanations below, it becomes very evident that this is indeed the case.

I/O between the CPU and memory occurs very often and so effectively doubling the bandwidth should give a considerably noticably performance boost. It shouldn't be double, but it should be more than just 10% in <i>some</i> applications. (And far less in others.) Yet so far we have not seen any significant performance boost. In fact, there are SDR motherboards outperforming DDR motherboards. How can you possibly explain that if we are to believe that the DDR motherboard is faster because it has twice the bandwidth?

It is my belief (which has pretty strong benchmark evidence right on THG) that the performance gain that we are seeing from DDR systems has almost nothing to do with the memory bandwidth, but simply in a more efficient motherboard design.

Look at the performance of VIA's Intel chipsets compared to Intel's chipsets. Why do Intel's perform better with the same memory and CPU? It's because of the efficiency of the chipset and mobo itself.

So why did <i>some</i> DDR systems (but certainly not all) perform slightly better than SDR systems when they first came out? Again, motherboard and chipset efficiency.

AMD itself made a DDR chipset. Who else in the world could possible have known how to better design a chipset that supports a double-rated FSB to match the I/O of the Athlon to the I/O of the memory? Yet did we see anything even closely suggesting that the I/O bandwidth between the CPU and memory was doubled? Not unless the memory was slower than mud. If anything, the AMD chipset motherboards performed better simply because VIA sucks badly at tweaking the performance of chipsets.

So if no one, including AMD, could make a DDR chipset bridge the gap between the I/O of the Athlon and the I/O of the DDR SDRAM, then the flaw is hardly likely to be in the chipset. I mean honestly, what would it have served AMD to have made a single-rate chipset for a DDR CPU and DDR memory? No one gains money by cutting the performance of their technology in half.

And we know that the flaw isn't in the DDR SDRAM either. The flaw is quite simply that the benefits of the Athlon's double-pumped bus are internal only and that the Athlon's I/O speed is still only 133MHz.

And yes, DDR motherboards and memory will still have a small performance gain over SDR even when used with a CPU that doesn't support DDR I/O simply because of <i>synchonization</i>. The CPU asks for data from the memory. The memory responds. In a SDR system the memory's response might not be until two cycles later because it could just barely miss responding in one cycle because it's performing at the exact same speed as the CPU, but has to perform paging and send the data across the bus. Yet with DDR memory, now it's giving it's responses in half the time, so that the response is always only one cycle away from when the CPU asked for it because it's response is only half a cycle away from the CPU's I/O so that the slight time accumulated from paging and transfer still won't put the response time more than one cycle away from when the CPU asked for it.

It's the same reason why putting 133MHz SDRAM (clocked at 133MHz) into a system with a 100MHz FSB gives a performance gain. In theory if everything were synced perfectly, there would be no performance gain at all because the CPU would still be accessing the memory only at 100MHz. In practice though, there is a minor performance gain simply because the memory responds to the chipset faster now so that the CPU doesn't have to wait when the memory misses it's synchonization slightly.

And this is why the P4 gains a quite noticable performance from faster memory. The P4's I/O is quad-pumped so that each clock cycle in the chip internally is able to send and recieve data from the chipset. And the chipset able to send and recieve data that fast as well. So the faster the memory, the sooner the CPU gets what it wants. There are less internal clock cycles of the CPU wasted waiting for it's next I/O cycle like there is with a single-pumped I/O cycle.

Hopefully AMD will fix this oversight with their CPUs and soon we'll see DDR systems that actually show the benefits of DDR memory.

From my deductions though, it's quite clear that it's the Athlon CPU itself that is still operating externally only single-pumped, just as the P3 does, making the 266MHz FSB not only a scientifical misnomer, but also a complete myth.

Don't get me wrong. I'm a fan of AMD and the Athlon. I'm no Intel zealot spewing nonsense just to badmouth them. I am simply giving every shred of evidence I have why the 266MHz FSB is just a load of marketting BS. Technically so is Intel's 400MHz FSB, but at least Intel's is only scientifically misnamed and not completely a false advertisement.

If the opposite of pro is con, what is the opposite of productivity? Ground first.
 

Matisaro

Splendid
Mar 23, 2001
6,737
0
25,780
0
I/O between the CPU and memory occurs very often and so effectively doubling the bandwidth should give a considerably noticably performance boost. It shouldn't be double, but it should be more than just 10% in some applications. (And far less in others.) Yet so far we have not seen any significant performance boost. In fact, there are SDR motherboards outperforming DDR motherboards. How can you possibly explain that if we are to believe that the DDR motherboard is faster because it has twice the bandwidth?

How much data does the cpu need to run a program considering cache etc, 3d games use the videocard, however, ddr platforms show signifigant improvement in video apps(mpeg encoding etc) because those apps call for large streams of data from system ram using the higher bandwidth. When you run programs only rarely will the cpu require large amounts of data quickly(due to the design, thats just how tbirds work) when apps are being used which do need large amounts of data quickly(video etc) the ddr boards generally score higher. If your conveyor belt runs at 266mhz but you only need 100mhz worth of data, you will see no benifit above 100mhz system ram.

Hope i made sense, just got off work and am very tired. NIGHT ALL.

~Matisaro~
"Friends don't let friends buy Pentiums"
~Tbird1.3@1.55~
 

Ncogneto

Distinguished
Dec 31, 2007
2,245
0
19,780
0
Dude you are so wrong you are starting to look silly. You may actualy want to check your facts before making assumptions such as this:

<font color=red> It's I/O with the chipset is still only at single-pumped 133MHz speed.</font color=red>

AMD's implementation of the Alpha EV6 bus is 64bit thus at 133mhz and 8Bytes (8bits*8Bits=64bits) and being that it is a Double Data Rate bus (DDR, transfer data on rising and Falling edges of Xfer) it is simple math (133X8X2=2.184 or 2.1GB/s).
your wrong...got it? WRONG! This conclusively proves your I/O theory incorrect.

<font color=red>No, I don't have any engineering blueprints or anything to prove this, but it is very evident in the performance of the systems.</font color=red>

Perhpas you should actually find evidence before posting such bogus information.

<font color=red>I/O between the CPU and memory occurs very often and so effectively doubling the bandwidth should give a considerably noticably performance boost. It shouldn't be double, but it should be more than just 10% in some applications. (And far less in others.) Yet so far we have not seen any significant performance boost. In fact, there are SDR motherboards outperforming DDR motherboards. How can you possibly explain that if we are to believe that the DDR motherboard is faster because it has twice the bandwidth?</font color=red>

I have already addressed this with you in another post, one in which you choose to ignore and post your FUD yet in another thread.

The modest increase in performance ( your words) is explanable do to the fact that only the bandwith has been addressed while the latency remains the same. Thus we only see increases in performance in apps that were bandwith limited to begin with. Latency continues to be the bottleneck much more so than bandwith. This very issue was addressed in the p4 by going to a dual channel memory controller and will be addressed in the nforce as well along with a hardware prefetch.
Your observations about the inherent worth of DDR vs Sdram and the performance increase may be debateable, however your conclusions about the I/O operations of the cpu to northbridge of the athlon are totally incorrect.



A little bit of knowledge is a dangerous thing!
 

Ncogneto

Distinguished
Dec 31, 2007
2,245
0
19,780
0
Theoretically DDR-SDRAM provides double the bandwidth of normal SDRAM. At 100 MHz memory clock it's able to supply a data bandwidth of 1600 MB/s, at 133 MHz it goes up to 2100 MB/s. That is why 'Team DDR' decided to call the specs of those two DDR-memory types 'PC1600' and 'PC2100'. However, I have to disappoint you, if you should think that DDR-SDRAM is simply twice as fast as SDRAM at single data rate. The basic reason is rather simple. While DDR-SDRAM is indeed able to provide double the peak bandwidth of normal SDRAM, it has still the same (CL2 DDR-SDRAM) or even a slightly worse latency (CL 2.5 DDR-SDRAM) than its older sibling. The result is that in worst-case situations DDR-SDRAM can be as slow or even slightly slower than normal SDRAM. At the same time DDR-SDRAM can easily outperform single data rate SDRAM under ideal conditions. In average, systems with DDR-SDRAM are most certainly faster than systems with normal SDRAM, but the gain depends on the application and can hardly ever reach more than 20%.
<A HREF="http://www6.tomshardware.com/mainboard/01q1/010104/amd-133fsb-01.html" target="_new">http://www6.tomshardware.com/mainboard/01q1/010104/amd-133fsb-01.html</A>

One of the most important new features of AMD760 is the support of 133 MHz front side bus. This increases the data transfer bandwidth of Athlon's bus from 1.6 GB/s to 2.1 GB/s, which makes perfect sense, because PC266/PC2100 DDR-SDRAM can supply data at 2.1 GB/s as well. The only other noteworthy new feature of AMD760 is the ATA100-support. All the other specs are listed in the table above.
<A HREF="http://www6.tomshardware.com/mainboard/00q4/001030/athlon-05.html" target="_new">http://www6.tomshardware.com/mainboard/00q4/001030/athlon-05.html</A>

If you really want to see what is happening I suggest you take a good look at the following article:

<A HREF="http://www.hardocp.com/articles/memory/ddrovr/" target="_new">http://www.hardocp.com/articles/memory/ddrovr/</A>

A little bit of knowledge is a dangerous thing!
 

Similar threads


ASK THE COMMUNITY

TRENDING THREADS