Our team got passed by HPCS

The only thing it seems to do is drop CPU usag down by 5% using the updated core. It doesn't seem to improve the WU/PPD directly unless you count the extra time the CPU now has for the SMP.
I am surprised that the WU for the GPU gives so few points and for me is incredibly consistent in doing that. The CPU does more PPD by approx 3000. Either the i7 is one incredible CPU or the stuff I read previously about the GPU being better than the CPU at this is really a bucket load of crap!
Further add to this I could improve the CPU performance by clocking it up too.

I think, even with the ineffiency of the AMD GPU, I am still a bit surprised that the GPU does not output more PPD.
 
It appears that each process running requires 12.5% of the CPU regardless of task.

SMP 6 = 75%
SMP 4 = 50%

GPU = 13% (cannot report halves).
Uniprocessor = 13%

FahCore_16 for the GPU takes up 375Mb per process.

According to the client I can get more PPD using SMP4, GPU, GPU, Uniprocessor with 0-10% idle CPU.
SMP 6, GPU, GPU leaves the system completely maxed out but no room for the GPU to aquire more CPU time if they require which does jump from 13% to as high as 25% quite frequently.

SMP6 was offering 11k PPD
SMP4 is offering 9.5k PPD
Uniprocessor is offering 1k PPD
GPU combined was offering 15k PPD
GPU combined is offering 19k PPD

PPD of 29.5k now vs 26k before but actually using slightly less CPU time overall!

Based on a 12hour test.
 
I think, even with the ineffiency of the AMD GPU, I am still a bit surprised that the GPU does not output more PPD.

that's one of the main reasons why i don't use gpu folding. the systems i have that have worthwhile gpus have cpus fast enough that i would see no benefit at all. until they start offering bonuses for gpu work units, my only clients will be smp.
 
Indeed. Did they used too because I am sure in the past I have done WU of 15-20k as a single task on the GPU. Guess in the same vain of bigadv. They took slightly longer which is obvious but the rewards were great and at the time the same thing on a CPU took over a day.

The CPU currently is deffo topping the GPU for PPD by some 3k and that was based on SMP6 so SMP8 widens the gap further 5-7k and if you put overclocking in to the bargain you can at least remove the need of 1 GPU to make the same points.
Unfortunately the stuff at work is all virtualised so no go there and the old servers are on the floor ready to be sold. The workstations are not spectacular (Core2Quad low end) at best or laptops. My quad core pc at work is sitting doing its part but its peanuts really. Every little helps mind.
 
Back
Top