And the first shall be last.
9 May 2006 21:02:56 UTC
Topic 191199
(moderation:
Has anyone reported or noticed a major change in relative performance of Intel vs AMD cpus using akofs' executables?
I have 5 computers running E@H at nearly 24x7. I will ignore the oldest & slowest here.
Then I have two computers that each run one Pentium socket 778 with hyperthreading.
And one dual cpu AthlonMP.
and one single cpu dual core Athlon64.
I have held all four computers fixed at akosf's
version S40.03 for long enough to get a fair performance comparison, I hope. All four computers run Linux and WINE.
Before aksof's accelerated version (starting in Feb 2005), the two AMD computers had always been clearly faster than the Intels. That did not surprise me because both of the AMD computers have two complete CPUs including two FPUs & two SSEs per computer. That is not true of the Intels.
But now running S40.03 the Intel HTs are clearly faster in terms of average points awarded.
Am I just running these computers wrong in some way?
I know akosf also has some 3D-NOW versions, but I think the Athlons I am using are SSE compatible & I think SSE is faster than 3DNOW.
Is it in fact the case that at some comparable level of technology Intels have now become clearly faster than AMDs when running E@H?
ADDMP.
And the first shall be last.
)
Take a look at the top hosts for E@H, the list is dominated by AMD chips.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
I'm not sure about the S40
)
I'm not sure about the S40 series, and it could also be an arifact of WINE, but with s41.xx AMD chips in real windows are significantly faster than their Intel rivals. IIRC the same's true for U41.xx.
DXX is generally a few percent slower than the equivilant SXX code, so there's no reason to test on that branch.
Hello ADDMP...I noticed you
)
Hello ADDMP...I noticed you said you were using S40.03 by Akos...
That app has a "zero" credit problem which I had got when I first tried it. Akos suggests to go up to the next app at least. He has his own thread for the current downloads for your review.
Yes I have noticed the zero
)
Yes I have noticed the zero credit problem, but I didn't regard it as very important because the awarded credit is nonzero. But I will try some later version as you suggest.
Thanks,
ADDMP.
I'll look around at the s41's
)
I'll look around at the s41's & try one.
Thanks,
ADDMP.
It is interesting that the
)
It is interesting that the problem is solely mine, but still puzzling since the timing seemed to coincide with my installation of s40.03.
I will try something newer, but I am flying blind here.
Thanks,
ADDMP.
RE: It is interesting that
)
If your system supports SSE then my suggestion is to use S41.07. If in doubt as to what instruction sets that your cpu uses get and use CPU-Z from here .
98SE XP2500+ @ 2.1 GHz Boinc v5.8.8
My Athlons (running Windows
)
My Athlons (running Windows 2000 and XP 64 bit) are much faster than any of my Pentiums (running Windows 2000/XP/2003). Even my Athlon XP 3000+ (at stock 2167MHz) blazes through (long) work units at 40-45 minutes each, using S41.07. With SETI Optimised apps, the Athlon XP was much slower than any of my Pentiums.
I'm no expert by any means, but I get the impression it's something to do with WINE using particular registers (to emulate Windows) which the optimised app needs. Could be the Pentiums have more of those registers. But I may be wrong.
Join the #1 Aussie Alliance on Einstein
P4's running win directly are
)
P4's running win directly are slower than athlons in einstien. the problem is that the p4's 8k L1 cache can't hold the entire working set of data, but the 32k L1 cache of an athlon can. Since L2 cache is slower than L1 it generates a noticable performance hit.
RE: P4's running win
)
I'm aware of that :-) I was just pointing out that on Windows (at least with my PC's) the story seems to be the opposite of what the original poster showed, running Linux with WINE. Seems to match your own reply below.
Join the #1 Aussie Alliance on Einstein