Good news for AMD folders?

Mohonri

Supreme [H]ardness
Joined
Jul 29, 2005
Messages
5,772
AMD has just released their Performance Library as open source. This means that developers now have a free library of optimized code for....well, all sorts of stuff. If Stanford are able to use this, it might prove to be a boon for people folding on AMD processors.
 
If Stanford can work it into the current clients, I would think they would do it. However, if it requires having a different download for AMD clients and Intel clients they won't do it. They've stated before that it's not worth their time or effort to do both and there is also something about the licensing deal they have with Intel an the Intel compiler.

As much as I would love to see some AMD based optimizations included into the client, I don't think it's going to happen just yet.

I don't mean to rain on anyone's parade, but this has been Stanford's policy for a while now. I hope they can get something done with AMD optimizations. I know for a few years AMDs were the processors to have for folding until the Core2 line came out. Even with Intel optimized compilers the AMDs were blowing away Intel processors. I would love for my X2 to be able to put out some more points without having to replace the CPU. It would be very nice.

 
I have to agree with SmokeRngs. Stanford would need to be convinced of a really big return on investment for their coders to undergo the effort. As it stands right now, most of the hardware currently folding is Intel-based, and AMD's future appears increasingly hazy. With delayed launches and lack-luster releases compounded with a plethora of rumors concerning a possible acquisition, I can't see a compelling reason for Stanford to change their long-held policy. Then there's the licensing issue that has been mentioned a few times in the past couple of months. I fear only a paradigm shift would change things around.

 
I don't mean to rain on the parade, but... other than to *maybe* allow QMD workunits on AMD chips (if they ever return), I don't think this will make much of a difference. Most work units are Gromacs, which use SSE or SSE2 optimizations, and ALL AMD chips in the last few years run that at full speed, without the need for a customized performance library.

Although this does open up the possibilities for using this in new types of work units not based on Gromacs. ;)
 
Although this does open up the possibilities for using this in new types of work units not based on Gromacs. ;)
Hi 7im,

Are there new types of WUs going to be released this year not based on Gromacs?
 
there are three active WUs based on the Amber core floating around (in fact i got one last week), but Stanford seems to really be loving the diversity of the Gromacs core (using Gromacs, Gromacs 3.3, Double Gromacs/B, Generalized Born Gromacs, Serial Replica Gromacs, Simulated Temperting Gromacs, and the specialized Gromacs for SMP, PS3, GPU)

 
Hi 7im,

Are there new types of WUs going to be released this year not based on Gromacs?

There are the amber core based proteins. I still see those every now and then.

While Gromacs have been the basis for most work units in the last few years, just remember, the tinker core was the basis for a long time as well. I see no reason why there won't be another type out in the future.

I'll take whatever Stanford can give me for my AMD boxen. If it's not much, then I will welcome the extra speed. If it's nothing, then I still have what I'm doing now.

 
OK, I only run the SMP and GPU clients and wasn't seeing any of the Amber units. I remember them from the second time I was folding several years back. I didn't know they were still around, though. As they say, variety is the spice of life.
 
Back
Top