AMD gpu projects

[nyanthiss]

Hi,

I own a 7950 gpu. I tried folding on it some months ago, and i got (IIRC) about 6k PPD and always only one project assigned (11293). So a few questions:

• Are there any other projects planned that can use amd gpus ?
• Are there any plans to make the fah code more efficient on amd gpus ?

So far i've seen no sign of progress on either, that's why i'm asking - i'd love to be proven wrong.

[mdk777]

Have had a 7970 since the first day of launch.

I feel your pain.

I "presume" there are plans, although there have been no announcements.

There continues to be tools issued by AMD; SDK, compilers, analyzers...etc to support open CL on their new architecture.

Again, I presume that it is only a matter of time...but I can't tell what the time is.

[nyanthiss]

There continues to be tools issued by AMD; SDK, compilers, analyzers...etc to support open CL on their new architecture.

This has nothing to do with AMD software. There are tons of OpenCL benchmarks & real world programs (e.g. boinc projects) that show, at the very least, performance comparable to NVidia's.

[Evil Penguin]

Hi,

I own a 7950 gpu. I tried folding on it some months ago, and i got (IIRC) about 6k PPD and always only one project assigned (11293). So a few questions:

• Are there any other projects planned that can use amd gpus ?
• Are there any plans to make the fah code more efficient on amd gpus ?

So far i've seen no sign of progress on either, that's why i'm asking - i'd love to be proven wrong.

Last I heard (July of this year) the problem was bugs with AMD's OpenCL compiler.
I don't know what has been done since then to fix that for a future release.

[mdk777]

I'm well aware of the potential.
(I've been following this for a long, long time...)

However, not all calculations are the same.

Luxmark V2 score might be great for rendering applications(that use Open CL) but not those that use open GL (still used by many high end CAD programs)
Porting software is not instantaneous....some functions scale on new architecture, others don't....

It does indeed have everything to do with the entire software/hardware ecosystem. There is no sense in denying the obvious critical mass/ the self perpetuating feedback of more users resulting in more and diversified support drawing in more diverse and sustaining customer base...which in turn draws in more software suppliers.

It may not be fair or right, but it is certainly true.

I happen to think AMD Hardware is very adequate and very competitively positioned. However, recent cuts in engineering staff does put them in jeopardy of eating their seed corn.

So, do I think that FOLDING "can" utilize the AMD Next Gen CORE?, certainly.
Whether they do or not? is ultimately dependent on the health of the software/hardware ecosystem.

I think it will happen...but I am no longer making market bets on the outcome at this time.

[k1wi]

All we can do is look at the hints that are being made. There is certainly progress being made, the only problem is that everything needs to line up before something meaningful can be released. By everything I mean literally everything - basically they need OpenCL, firmware/drivers/compilers, hardware (the shift from VLIW5 to VLIW4 to GCN has been tough) all stable and squared away, then they need to push GROMACS to compliance (hint: 4.6). Until they all line up you either have to chose to throw energy at a moving target or wait until there is a sense of stability.

Only then can they start work on cross platform standardisation and optimisation. Given the complexity and the highly optimised nature of GROMACS that takes A LOT of time. A length of time that is compounded both by what is not an infinite amount of resources and an absolute requirement for scientific robustness. (Think how long continuously progressive NVidia development there has been over many years with a stable ecosystem and compare it with an AMD/OpenCL equivalent that is relatively still in its infancy).

I actually think PG needs to be given quite a bit of credit tbh. Everything takes longer than people wish.

[nyanthiss]

All we can do is look at the hints that are being made. There is certainly progress being made,

Can you elaborate on what hints and what progress you mean ? Ideally something very specific, on the two questions from my original post... because, while i thank all you for your answers, they are very general... e.g. you mention gromacs 4.6... can anyone confirm it aims to solve issues i'm asking about ?

Cheers

[mdk777]

No.

We can only look at what is being done regarding drivers and support.

I don't know of any direct answer to your question. I predicted that there would be support for Keepler before Tahiti. NVIDIA has within the last month launched CUDA 5 Support for Keepler.
Hence, I would suspect optimized folding cores for Keepler to follow shortly...next 1 to 3 months.(maybe sooner)
Gromacs will continue to expand on OPEN CL.

If that will work with Tahiti SDK and OPEN CL drivers at launch of 4.6 remains to be seen.

If I had to put a best guess...3 to six months until we see an efficient core optimized for AMD Tahiti.

But as you mention, that is just my WAG based on my experience. I doubt you will find more concrete information regarding timelines however.

[nyanthiss]

OK, thanks everyone. I'll check back in some months again i guess.

Cheers