Prime95 - How Come P4's are better at it?

[Glavinsolo]

How come my overclocked 3500+ @ 2.502GHZ is like seconds behind slow 3.6ghz P4's?

I just don't understand...

 

[rogue1979]

I have a mobile Barton running 2300MHz+ and a P4 running 3.3GHz. In gaming benchmarks they are about dead even, the Barton feels more responsive in desktop activities.

But when it comes to running Prime95, the P4 crushes the Barton, appearing to run it about 1 1/2 times faster.

 

[Lithan]

Prime95 has sse2 & sse3 optimizations maybe. I never really gave a damn.

 

[gururu]

why wouldn't they be better at it? maybe they do some things really well afterall...who would've thought?

 

[TheStu]

that's pretty funny...

 

[Glavinsolo]

Well when many applications are used to benchmark CPU's you don't see much of a gap. But when it comes to prime it is as if AMD is back a few years. Maybe I'll run it again when I get my winXP 64bit and see where it stands.

 

[Shenkoa]

Wow, Intel is better at something. AMD and Intel have their strong points, most AMD fanboys only see in one direction.

 

[PremiumG]

Do P4's have more cache? I believe if thats the reason if it does.

i remember an article a while back where the slower cached Katmai P3's would beat the coppermine P3's at Prime95 because of more cache (even though it was running at a slower speed)

 

[clarkey01]

depends on what model, Northwood = 512k, willy = 256k, Prescott = 1 meg.

 

[DAPUNISHER]

I can't really say why the P4 is faster, unless as suggested it is optimized for the architecture or very clockspeed influenced? P95 is OK as a a stability test, but I feel other DC projects such as F@H are more useful Subsequently I don't run P95 or the included TT at all, particularly since Gromacs WUs have proved a superior stability tester for myself.

 

[cbehnken]

Hey Dapunisher, what happens if F@H detects a problem? I've had prime95 fail at about 2520 on my machine, but F@H never complains.

 

[DAPUNISHER]

Originally posted by: cbehnken
Hey Dapunisher, what happens if F@H detects a problem? I've had prime95 fail at about 2520 on my machine, but F@H never complains.

The reason it won't complain is that it'll just terminate the WU and move to another. If you check the log though you will see the error message and it even mentions overclocking could be the reason for it. I've run P95TT for 24hrs no problem then had a Gromacs error out in under an hour@the same settings. Others will tell you stressful things such as DIVX encoding can turn up instabilities P95TT didn't detect either.

 

[Goi]

Prime95 is very heavily FPU dependent, and the AMD A64 has the best x86 FPU around, so the only reason I can think of is that Prime95 is SSE/SSE2 optimized, and even though the A64 supports those SIMD instruction sets, they may not have been implemented as well as intel. Also, the clockspeed advantage here might count for more.

 

[Childs]

I believe SIMD is more important to the P4 architecture than it is for Athlon. Athlons usually don't see the same performance increases as P4s when its enabled. When SIMD instructions are disabled, and you measure the performance deficits when on and off you'll notice that the P4's really need it.

EDIT: Now, 3DNow and 3DNow 2 supposedly do much more for Athlons than SSE/SSE2, but no one really supports those SIMD instructions.