• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Most PPD GPU setup question

S

SpeedyVV

Supreme [H]ardness
Joined
Sep 14, 2007
Messages
4,210
Assuming a 4 card setup can someone correct the following?

Radeon R9 295x2 >
GeForce GTX TITAN Z >
Quadro K6000 >
GeForce GTX TITAN Black >
GeForce GTX 980 >
Radeon R9 290X >
GeForce GTX 970 >
Radeon R9 290 >
Quadro K4200 >
Radeon R9 285

I am also curious, do 4 GPUS get pooled together like a 2p or 4p CPU rig? meaning one WU being folded by all GPU simultaneously, or does it fold 4 separate WUs?
 
Last edited:
"Most ppd" is project dependent. I think the biggest point granters would be (in no specific order): Bitcoin Utopia, GPUGrid, PrimeGrid, Moo!, Milkyway, Poem, and Collatz Conjecture. All of these except GPUGrid can use either Nvidia or AMD GPU's. AMD cards seem to do better at Collatz, BU, Moo!, and Milkyway (if using a 280 or 280X), while Nvidia does better at PrimeGrid (PPS Sieve) and is the only card you can use at GPUGrid. I don't run Poem enough to know for sure which GPU is faster there.

In my experience, a 280X at Collatz is worth ~1.2 million PPD, but I do not know how the newer cards you have listed there do, as I don't own any of those cards. I have been buying up 280's and 280X's, since they are cheap now and have the best Double Precision performance for Milkyway.

As for your final question, no, GPU's do not usually get pooled together. Usually, each GPU crunches its own WU. The only project I know that can use multiple GPU's is MOO! and I never got that to work correctly. I believe it only work properly there on older hardware (HD6900 era), but I don't know that for sure. I just know it never worked for me.

All this being said, if money was no object, today I would buy 4 Titan Z's. Great Double Precision to go along with great Single Precision. I would guess 4 Titan Z's would be in the 8 million ppd range at Collatz and probably about the same at GPUGrid and PrimeGrid. I am just making a rough estimate, though.

Honestly, Eagle07/Patriot is probably the expert on this subject. Maybe he will chime in.
 
Last edited:
I think he is referring to FAH since he said "WU being folded". However, I could be wrong. As far as I understand, FAH only does one card per work unit like most other projects. RFGuy pretty much explained the rest though.

For a little more clarity, it is pretty safe to say if the app is written in OpenCL, it is probably going to run better on AMD/ATI cards.
 
Ah, good point, Gilthanis. I missed the "being folded" part. Oops!

Your point about OpenCL apps is very true. What is interesting to me is that in my experience CUDA apps on Nvidia GPU's do not usually run faster than OpenCL apps on AMD GPU's. One exception is PPS Sieve tasks at PrimeGrid, where the Nvidia CUDA app blows away the AMD OpenCL app. However, as a general rule, the order of speed is as follows: 1) OpenCL on AMD, 2) CUDA on Nvidia (of course), 3) OpenCL on Nvidia.
 
Last edited:
You forgot 4) OpenCL on Intel

Basically, OpenCL is better supported by AMD which is pretty much not getting much more support from either camp. CUDA is much more mature and supported by nVidia quite well and also adopted heavily by most academics. From the extensive bickerings of many in the DC world about OpenCL vs. CUDA it seems to come down to what the original app was designed for first. So when comparing OpenCL on nVidia vs. CUDA on nVidia, CUDA should win but if the coder knows OpenCL better or the original code was OpenCL, then the results will favor OpenCL. Basically, OpenCL is less efficient because of its broader hardware it is typically coded to run on. But, I can't give you a lot of the ins and outs to prove any of it. I think if apps were designed for the specific platform by a dedicated coder with expertise in that language, the results would be a lot different. You also got to consider whether the apps require double precision. If it does, then AMD will have the leg up simply by design.
 
Thanks, Gilthanis. How could I forget Intel? Maybe it is because I've been laughed at here for mentioning my Intel GPU's. That's why I don't bother to list them in my signature.

I run Einstein, Albert, Seti, and Seti Beta on my Intel GPU's and they (all my Intel GPU's are 4600's) work really well on those projects. Not bad ppd for what I consider a "freebie," since they come with the CPU and would otherwise go unused. For the extra ~30 watts used, they are well worth running on these projects.

Collatz also has an Intel app, but I only run that on my AMD and Nvidia GPU's, for obvious reasons. I'm not sure what the Intel's ppd potential is on Collatz.
 
You must log in or register to reply here.
Back
Top