F@H Core 16: Taking AMD GPU Folding to the Next Level?
A few weeks ago, Stanford introduced the new Core 16 Project 11293 work units which are specifically tailored towards OpenCL-supporting AMD graphics cards. But do they bring the hoped-for increase in Folding@Home performance or is this yet another step towards disappointment?...
[ Read full article ]
Been working on this one since the new core came out. Hope you enjoy guys. This will be the first of many F@H articles in the coming months. :thumb:
Thanks for this Sky! Will come in handy when some of the AMD folders join up for the Chimp Challenge.
The high CPU usage bothers me a bit. That said, if I recall correctly (I was a rookie folder then), the first nVidia cores used more cpu cycles then too, but that could have been a CUDA issue as well.
Hopefully the cpu cycles come down on those WUs and the GPU folding won't rely on a 'CPU crutch' too much in the future, but at least Stanford is making some progress. The Gromacs team will have to do lots of work on this as well.
but plz explain again why the 5870 beats the 6970 at folding? :)
That's part of my theory and I won't bore you with the rest. :thumb:
basically, depending on the application using the cores, one can be faster than the other and viceversa?
looking at 3d apps you would think the 6970 would leave the ole 5870 way way back
interesting. also,whenever you got time, give the on chip gpu a folding and see how it does, unless its not supported yet :)
danke herr sky
Thanks for the write-up SKY! A thorough assessment. If the project maturation for OpenCL units turns out to be anything like that for CUDA, then later, larger projects will likely yield even greater dividends for cards with more stream processors as opposed to those with higher clock frequencies.
Given the resource requirements for AMD-based projects, it behooves Stanford to figure out a way to park the data in VRAM rather than rely on swaps with the system memory. User-added environment variables could partially do this with Brook. I really don't see any reason why Stanford can't do it completely with OpenCL. Certainly, in the current bigadv era, this advancement is too-little-too-late for dedicated folders and should be understood as an improvement for regular users. And regular users don't mess around with environment variables. As you said - a good first step. But there remains much to improve for AMD folders.
Interesting read, thanks for this article and the work behing
And wanted to add,
Congrats to HWC reviewers who did receive some loving from PCPER this week this is the second HW review Jeremy Hellstrom has highligted on the main page.
PC Perspective - The #1 Choice for PC Hardware Reviews and Information
Props to PcPer as well for continually bringing up to date news to their front page. They do one hell of a job!
|All times are GMT -7. The time now is 07:30 AM.|