Athlons not scaling as well as P4.

I'm a big fan of AMD, own shares etc, and of course I'm happy how well XP is doing. But I've been doing some spreadsheet work, looking how benchmark scores (mainly Sisoft Sandra) increase as the frequency / performance rating increases.

My results seem to indicate:

AMD's PR are increasingly becoming fiction. Consider the XP1500. Average the sisoft scores (from the last comprehensive CPU test at this site), of course first taking them as a percentage increase over a base processor. I use a 1GHz athlon tbird as a base. I also threw in Q3arena 640x480 rendering benchmark into the average, should be CPU bound.

Now, a XP1500 is only 92% of what you'd expect if you overclocked an Athlon 1Ghz to 1.5GHz. Not surprising, cos the XP is only running at 1.33GHz really. Still a bit fishy, why is it a 1500 not say a 1400? Going onwards, the numbers gradually reduce, until the XP2000 (the XP1900 OC'd that they reviewed here recently) is only 85% of its expected score. There is a steady drop. Assuming this trend continues, I get the XP2800 will be about 78% of the base.

Compare this with P4. A P4 1.4GHz gets a poor 80% of the expected score for an Athlon Tbird. But the P4 2.0GHz is only down to 78%. Much better scaling. The P4 2.8GHz therefore gets a 77% scaling rating, assuming linear increases in all the benchmarks. So by 2.8GHz, the P4 will be equal to a XP2800, rather than somewhat poorer.

So, I thought, maybe it's just the PR rating that is cheating. Substituting the actual clock frequencies for Athlon points to that - but with a very slight falloff. The XP1500@1.33GHz gets a rating of 104% of the benchmark, the XP2000@1.66GHz gets 102%. So, still very slight falloff, probably due to memory bandwidth and other constants gradually pulling down the top end processors.

A lot of this is probably because each increment in Athlon clock frequencies is 67MHz, but the PR rating implies 100MHz. Taking the first jump as an example, 1400MHz actual speed compared to 1333MHz actual speed is a 5.0% jump. But PR1500 to PR1600 implies a 6.6% speedup. That is the essense of the lie - AMD are implying speedup of each increment processor using the PR rating that more than you should get by an increase in the actual frequency, i.e. super-speedup. And we all know that's not likely without architectural changes.


Any comments? I can supply the spreadsheet if required.

Will Smith
will@willsmith.org

actually amd said the pr rating is actually the equivalent speed to that of the t-bird.

SO a axp 1600+ is the same as a t-bird 1600mhz.

check it out at amd's site.. i'll check now to confirm it ...

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

Huh, you deleted the post I was going to correct you on. Oh well.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

ya thats because it was complete BS.. just some idea i had floaten around. After i read AMD's little blurb i realized that ya this post is entirely stupid.

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

bad bad he's really really bad..
" The P4 2.8GHz therefore gets a 77% scaling rating, assuming linear increases in all the benchmarks. "

big oopsie... NEVER assume that anything will scale like this when benchmarking...

I'm mad I'm mad, I'm really really mad.

Athlon XP's are going to lose in some benchmarks to higher clocked T'Birds
if you just go on memory bandwidth even the processor can't help much..

But Sisoft FPu and ALU Bench's in particular have all to do with Clock speed and nothing to do with architecture enhancements to teh TBird.. SO, new conclusion -> T-bird vs. Palamino will perform the same under certain synthetic benchmarks.

balzi P.S sorry 'bout the Shrek quotes.. I'm just adding some flavour to this possibly boring tech. stuff.. hehe


I spilled coffee all over my wife's nighty... ...serves me right for wearing it?!?

Actually, if you read further down, they explain how the pr is compared to the tbird, that link has been used dozens of times to say the pr is compared to the p4, however on the very same page at the bottom the truth is revealed.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

Good thing the xp1900 is beating the 2ghz p4 by around 10-15% which is about, wow, 200 or 300 mhz on the p4 side, perfectly fitting into the model you proposed.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

LoL, but you see, whatever their motive, and what you think of it, the only thing we can proove is that it is what they say. The rest is circumstancial.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!