a way for fah to get alot more done?

science man · Jun 28, 2009

ok you know how if you set up a smp and gpu client the gpu client does 5 WUs in time it takes the smp to do one. I just wondered why didn't they ever think of making it equal by adding more work to the WU's of the gpu clients?

MARCI · Jun 28, 2009

The gpu is just very efficient at doing the calculations that F@H requires. There's really no way to make it "Bigger" it's all determined by the complexity of the protein. Some WU really stress the GPU alot more than others as it is.

You'll see that folders who have alot of GPU clients will indicate they have alot of "511s" or other Work units. Those work units are Either

A> Particularly FAST WU that increase overall PPD production
B> Particularly SLOW WU that dramaticly lower PPD production.

ChasR · Jun 28, 2009

As the client matures, we'll surely see larger protiens being simulated on the GPU. There are some obstacles. One being low end cards inability to fold stably when warm. A 10%+ failure rate on large protiens would be unacceptable. Another being the fairly large client base with 256 and 512 MB of memory. Current a2 SMP WUs use more. Swapping to system ram would likely lead to a very noticeable slowdown and possibly instability. Ram on the card is likely to be a limiting factor.

science man · Jun 28, 2009

oh well I got the ram to handle what I do. I got 4GB and while folding I still have 3.2GB free but I did realize that the WU is the actual protein and it isn't like they just created the WU.

MARCI · Jun 28, 2009

As ChasR said, it's not SYSTEM ram, it's the Ram on the Video Card that would be significant for GPU folding

science man · Jun 29, 2009

well how would you check that ram?

ChasR · Jun 29, 2009

GPU-Z or RivaTuner

science man · Jun 29, 2009

ok I have 512MB. does the gpu client suck that all up?

JackNSally · Jun 29, 2009

I don't think it would suck it all up. Some WU's use more ram than others but I don't think any WU's could use all 512mb. I think ChasR or someone else could probably give more specifics though.

MARCI · Jun 29, 2009

The problem you run into is that there are gpus that can run the client which only have 256mb, or some have 384.

In order to be able to run on as many processors as possible, which is where the REAL power in F@H comes from, they have to balance between speed, and sheer numbers available.

a way for fah to get alot more done?

science man

Disabled

MARCI

Member

ChasR

Senior Member

science man

Disabled

MARCI

Member

science man

Disabled

ChasR

Senior Member

science man

Disabled

JackNSally

Member

MARCI

Member

Similar threads