Why are CPU projects worth so few points relative to GPU?

[vica153]

Anyway, the original question was about CPU vs GPU points. I was under the impression that some projects required the use of a CPU instead of GPU. Apparently that was incorrect and it is simply a distribution of work to available computing resources. That pretty much invalidates the rest of my arguments about the point system.

[BobWilliams757]

FAH is not waiting around 8 days for your GT430 to finish it, but it's certainly possible that on somebody else's machiine that may be sort of true. They're waiting around to give the GT430 credit for trying. FAH waiting one day for that WU to be completed. Once the WU reaches the 1-day timeout, the WU will be duplicated and sent to somebody else. After the timeout, FAH has given up on your GT430 and decided they'd be better off finding somebody else to depend on who might happen to have a 1050 Ti.

After the Timeout expires, FAH is waiting for 8 days to consider granting a modicum of points for the failed effort to complete the WU. After the time out expires, the QRB points are disabled. Then they're willing to wait 7 more days to award (only) baseline points or to decide to give no credit at all to that slow GPU.

Forgive the newbie question, but is this to say that essentially FAH will move on and not ever waste time on WU's if there is more capable hardware to take it over if not completed in that one day period?

I have no concern about points or the QRB, just trying to clarify to make sure the work gets done quickly when possible. I'm assuming the WU's are assigned with the intention that the target machine can do the WU within a day.

I did some folding years ago when my hardware was more "cutting edge" then got away from it when hardware became less of a priority to me. Being I'm now running a stock 2400G setup, I know there are machines that run circles around it.

[JimboPalmer]

Between when you get a WU and it's timeout, F@H will not assign it to others. If it times out, then some other folder will get it, F@H does not have a clear idea who is 'capable' or not. All they get for a CPU is the number of cores assigned.

[Neil-B]

… but just because a WU might not complete within its timeout doesn't mean it should not be completed … it might not get reissued if the folding resource are limited (true, not necessarily the case at the moment) … but if the WU is completed in say 1.5x the timeout it may well complete before the reissued WU completes and so it is better for science to keep going - the machine to which it is reissued might be slower than the one that can't make the timeout !!

[Roadpower]

After reading this thread and watching my 2070 Super perform, I'm thinking more than ever about getting a second one. First one cost me a bit more than $600 with the taxes but I'm not complaining and it is cranking out the work at a good pace. I also can't take the cash to the grave with me.

[BobWilliams757]

Thanks for the responses. The timeout limits are generous enough that just about anything that will fold should be able to handle it.

[Nuitari]

So should I turn off the GT520 so that there is an extra core available for the CPU (i5-4590S) ?

Would it hurt the science to keep the GT520 running? It takes about 4 to 5 days to complete a WU. Recently it shows that most completed WU were also done by someone else, but looking at the past it is less likely.
Is there a setting that we can use to flag that its not a powerful gpu ?

[PantherX]

So should I turn off the GT520 so that there is an extra core available for the CPU (i5-4590S) ?...

That's a choice for you to make. My recommended is to use the PPD to make that decision as it means you get the most science done out of your system and contribution.

...Would it hurt the science to keep the GT520 running? It takes about 4 to 5 days to complete a WU. Recently it shows that most completed WU were also done by someone else, but looking at the past it is less likely...

As long as you successfully finish all the assigned WUs before the Timeout date, you're not hurting Science at all. Occassionally, a WU might be assigned to few Donors when the original WU either reports an error (to verify that it is indeed a bad WU and not an unstable GPU) or it is "lost" (wasn't returned by the Timeout date).

...Is there a setting that we can use to flag that its not a powerful gpu ?

At this stage, there's no flag to identify a powerful GPU from a a low-end one. We can only allocate WUs based upon the GPU's architecture, not its model.

[JimboPalmer]

So should I turn off the GT520 so that there is an extra core available for the CPU (i5-4590S) ?

Would it hurt the science to keep the GT520 running? It takes about 4 to 5 days to complete a WU. Recently it shows that most completed WU were also done by someone else, but looking at the past it is less likely.
Is there a setting that we can use to flag that its not a powerful gpu ?

You will do a little more science with one more CPU core, but do what you like so you keep doing it.

Finishing late hurts because the second attempt won't start until you time out.

But sincerely, do what you enjoy!

[Crunchtimer]

Hi guys
I'm wondering if anybody has an idea of the correlation between WU distribution, number of steps per WU, GPU utilization, credits etc.
I have two servers equipped with 4 Intel(R) Xeon(R) Platinum 8259CL CPU @ 2.50GHz and 1 NVIDIA Tesla T4 GPU dedicated to F@H.
Each WU assigned to the GPUs varies in the total number of steps required, e.g. 500.000, 2.000.000, 8.000.000 etc. however, each WU %-step takes, more or less, the same amount of time ~150 seconds for them to complete, i.e. the servers require ~ the same amount to time to complete 5.000 steps as 80.000 steps for different WUs. Now WUs generatedifferent amount of credit depending on the number of steps required it seems, for example 77.500 credits for a WU of 500.000 steps and 147.000 credits for a WU of 2.000.000 steps.

My question is now:
1) Can I manually decide how big WUs I can allocate my servers as they seem underutilized taking on less than 10Mega-steps WUs.
2) Are there any correlation between steps and complexity of WUs

Not sure if my questions are in the right order, but please share your thoughts and knowledge on this one.

Thanks!

[ajm]

Not sure I understand your description. But each project has WUs with the same "Base Credit", that indicates its complexity (for the donor). The number of steps is always 100 (ie from 1 to 100%) and your "speed" is expressed as TPF (time per frame, in minutes), and that speed will vary according to the Base Credit. viewtopic.php?p=327412&f=24#p327409

No, you can't decide which WU will be allocated to you. The server will just match your hardware with the requirements defined by the scientists.

[Joe_H]

One correction to arm's post, a "step" is a specific amount of time between each iteration of the calculation. A WU may need to process for 500,000 steps or 5 million. What he is thinking of is referred to as a "frame". Now a frame is always equal to 1 percent of progress, but in the early years of F@h might not have been.

So yes there is a bit of correlation between complexity and number of steps. A smaller, less complex system sill usually be configured to run WUs for more time steps so the total time spent will be on the same order as a WU from a larger, more complex project. Consideration of upload size of the completed WU is also taken.

There is a setting in the client left over from dial up access and slower internet connections for requesting WUs that will have a results file smaller than a certain size. It can be specified in MB or "small", "normal", or "big". But this only refers to the normal upload size of the WU, and needs to have been set on the serve by a researcher. It does not have a direct link to length of time needed to process, or difficulty.

[Crunchtimer]

Not sure I understand your description. But each project has WUs with the same "Base Credit", that indicates its complexity (for the donor). The number of steps is always 100 (ie from 1 to 100%) and your "speed" is expressed as TPF (time per frame, in minutes), and that speed will vary according to the Base Credit. viewtopic.php?p=327412&f=24#p327409

No, you can't decide which WU will be allocated to you. The server will just match your hardware with the requirements defined by the scientists.

Thanks for the replies to my questions!
Well if you check the fahclient log.txt you can see that each WU has different number of steps in addition to the %, and as Joe_H explained they don't seem to have a direct correlation to credits:

Code: Select all

20:20:26:WU01:FS01:0x22:Completed 1340000 out of 2000000 steps (67%)
20:20:33:WU00:FS00:0xa7:Completed 305000 out of 500000 steps (61%)
20:21:39:WU01:FS01:0x22:Completed 1360000 out of 2000000 steps (68%)
20:22:45:WU00:FS00:0xa7:Completed 310000 out of 500000 steps (62%)
20:22:50:WU01:FS01:0x22:Completed 1380000 out of 2000000 steps (69%)
20:24:01:WU01:FS01:0x22:Completed 1400000 out of 2000000 steps (70%)

I took some samples and calculated 2 ratios Credits/second (9.2 - 12.1) and Credit/step (0.01 - 0.16) which further supports what Joe_H explained, I guess.
Then I can happily sit back and expect 10x3600x24x2 credits each day from my 2 servers

[bruce]

You've mentioned several important measurements of a WU but you've left out the biggest driver for project complexity and the associated points: The number of atoms. If the protein has 50 000 atoms and another has 100 000 the latter is not twice as complex; it's 4x as complex. Motions are governed by the forces between each PAIR of atoms,

See column 3: https://apps.foldingathome.org/psummary

[Crunchtimer]

Thanks Bruce, that's interesting!
But I guess it's still like ajm says, that you can't decide which WU will be allocated to you?

Happy folding everyone!