version 23.6 - don't bother to download it

[SalemTheCat100]

Originally Posted by trif

George has had to buy the hardware necessary for him to do testing on various processors. Neither Intel nor AMD has provided him with processors, so improvements happen as George is able to acquire hardware. The P4 improvements didn't happen until quite some time after the P4 came out, because George didn't have a processor to test on. The Athlon improvements in the newer versions are because George purchased an AMD system to work on.

If someone at AMD wants an optimized version for the just released processors, they'll have to provide George with hardware, or they will have to wait for them to become affordable enough for George to buy one (or for some kind souls to provide him with one, my personal favorite).

Prime95 works just fine on my AMD processors, it is certainly not an Intel only product.

I know that Prime95 works on the Athlon. It just seems that the AMD product line support is an afterthought. Now AMD has a processor that has doubled the XMM registers to 16. That was three months ago and all I see is more P4 improvements.

I also posted a message around that time to contact AMD for a system to use for optimising Prime95 for the AMD64 platform. It seems that either AMD or nVidia (pro 150 system) would jump at a chance to get their processors and system platforms to be seen as high performance.

Was AMD contacted? If not why? Is so, what was their reply?

Because of the lack of support Prime95 is giving to the AMD Athlon line of processors I have turned off all seven of my Athlon systems and they no longer are doing any Prime95 work.

The Athlon64 set for release on September 23, 2003 (two months from now).

Will there be an optimised version of Prime95 available for it? :(

SALEM

[Prime Monster]

I also posted a message around that time to contact AMD for a system to use for optimising Prime95 for the AMD64 platform. It seems that either AMD or nVidia (pro 150 system) would jump at a chance to get their processors and system platforms to be seen as high performance.

I did write to AMD and asked for a system like that. I got a nice but negative reply. I do not believe it was / is high on their list of priorities. Anyway, feel free try try again. I can't hurt, and I am sure George would be happy to receive a new system ;) - at least I would.

heretic

[trif]

Originally Posted by SalemTheCat100

I know that Prime95 works on the Athlon. It just seems that the AMD product line support is an afterthought. Now AMD has a processor that has doubled the XMM registers to 16. That was three months ago and all I see is more P4 improvements.

It's not an afterthought. Prior to the SSE2 work that George did, Athlons spanked Intel chips at Prime95. Until Athlon comes up with a chip that has something like SSE2 that can be used to speed up the FFT code, the P4's will continue to win. George has also noted that working on the SSE2 code is much easier than working on the other code because the SSE2 registers are named, not stack based.

I also posted a message around that time to contact AMD for a system to use for optimising Prime95 for the AMD64 platform. It seems that either AMD or nVidia (pro 150 system) would jump at a chance to get their processors and system platforms to be seen as high performance.

Was AMD contacted? If not why? Is so, what was their reply?

Yes, AMD has been contacted to request hardware for George. They declined.

Because of the lack of support Prime95 is giving to the AMD Athlon line of processors I have turned off all seven of my Athlon systems and they no longer are doing any Prime95 work.

What kind of support do you want? There have been recent improvements for the Athlon family of processors since George got a system to work on, mostly in cache prefetches and taking advantage of the Athlon's additive cache. Until such time as the instruction set or timings on existing instructions change (such as getting SSE2 with a fast implementation like the P4's), there really isn't much that can be done with the Athlon code. The 3DNow instructions aren't useful in speeding up the FFT code.

The Athlon64 set for release on September 23, 2003 (two months from now).

Will there be an optimised version of Prime95 available for it? :(

SALEM

If you want an optimised version before the chips come down in price to reasonable levels, feel free to purchase the hardware for George. Opterons are already available, but it's not easy to even find a system to run benches on, much less do development on. It's not reasonable to expect George to pop for the price premium on newly released chips. He doesn't get paid for any of this.

[clowns789]

Didn't Prime95 say he fixed the Illegal Sumout error? And what about PayPal?

[ColdFury]

Because of the lack of support Prime95 is giving to the AMD Athlon line of processors I have turned off all seven of my Athlon systems and they no longer are doing any Prime95 work.

Most DC projects don't use processor specific optimizations at all. Besides, there's not much more optimization that can be done with Athlons.

[Prime95]

Originally Posted by clowns789

Didn't Prime95 say he fixed the Illegal Sumout error? And what about PayPal?

Yes, the torture test now stops after 100 ILLEGAL SUMOUT errors.

PayPal - I'm not a big fan of the idea. Raising money leads to ongoing questions about how much money is raised and how it is spent. I buy a new computer every 2 or 3 years anyway and Scott is generously running the primenet server using a business ISP account that he needs for his other ventures.

If we ever win the $100,000 award for a 10,000,000 digit prime then we can revisit the issue. We'll have the money suspicions anyway and we'll be incorporated as a non-profit.

[Prime95]

Originally Posted by ColdFury

Most DC projects don't use processor specific optimizations at all. Besides, there's not much more optimization that can be done with Athlons.

Actually, I'm pretty sure I can get another 10% for Athlons and Pentium 3 computers by employing one of the tricks I learned in the P4 optimizations. Unfortunately, it requires a major rewrite of the x86 FFT code. This is not something I relish doing!

Sadly, I also know of a way to improve the proth primes search (seventeen-or-bust) by a substantial amount, but again it requires a major coding effort.

All feature requests (be they from end users, Intel, or AMD) and optimization ideas are weighed based on how hard they are to implement and how much benefit is gained. The Intel request of a dialog box for the torture test was a piece of cake to implement and may help your average overclocker figure out whether a problem originates in the CPU or memory.

As for Athlon64 and Opteron there are three optimizations I can think of.
One: use the extra xmm registers. This will gain maybe a percent. Not too hard to implement. Two: prefetch 64 byte cache lines instead of the P4's 128 byte cache lines. I built a special prime95 that did this and surprisingly it did not make any difference. Three: use 64-bit instructions for factoring. This is likely to be tedious to code and while making factoring faster does not speedup GIMPS thoughput greatly because the bulk of the CPU time goes to LL testing.

I don't need to own an Athlon64 or Opteron to some performance work. Someone on this forum could buy an Athlon64, install Linux, and give me SSH access to an account (high speed Internet access required too)

[eepiccolo]

Originally Posted by Prime95

I don't need to own an Athlon64 or Opteron to some performance work. Someone on this forum could buy an Athlon64, install Linux, and give me SSH access to an account (high speed Internet access required too)

Doh! That count's me out :( ;)

Out out curiosity, would ping time be important, also?

[Xyzzy]

Hey, you all can pitch in and buy an Opteron and I'll host it here! I'll take real good care of it too! Yeah, that's the ticket!

[trif]

I could host it here, but the heat from the other computers is already problem enough. Although I suppose it could just be swapped in for the K6-2/400 that is the Linux box now.

Bare minimum:
CPU
Motherboard (with built-in NIC)
Memory
PSU
Hard disk if it can't be gotten to boot remotely

Pricing it out. Er, cheapest Opteron/mobo combo right now is $600. Ouch. I vote we wait until the Athlon64 comes out and comes down in price a bit.

Would the P4 optimizations made in version 23 make a difference for Athlon 64 and Opteron since they also have SSE2? Or are the optimizations more architectural specific than just having SSE2?