ABP1 CUDA applications

John Clark
John Clark
Joined: 4 May 07
Posts: 1087
Credit: 3143193
RAC: 0

RE: Dearie me, keep it

Message 95607 in response to message 95603

Quote:

Dearie me, keep it civil folks, we are all working for the good of humankind after all.

Can I just clarify something. I gather from above that the CUDA application might only use the GPU for a small part of the time - 4% was mentioned. Does this mean that it can be working on Collatz etc for the the other 96%? Or is it not that simple and the GPU needs to hang around waiting for Einstein?

If it is the case that my GPU will be in a state of enforced idleness then I think I might chose to set it to work on something that keeps it a little busier.

You are possibly already aware of this, but I will state it anyway.

I find that GPU crunching (mainly based on ATI 38xx and 48xx creatures) will take 1 project on the GPU and a single, or more, on the CPUs. But crunching Collatz and MW on the same GPU seems impossible and the GPU remains idle.

Outside of using the CPU for other projects, the way I go is to attach to say Collatz and GPU crunch. I then set the project to NNT, and attach to MM and GPU crunch here. If problems remain on the project then it;s a simple case of moving from Collatz to MW and back by suspending the one I don't want to crunch.

Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!

Grutte Pier [Wa Oars]~GP500
Grutte Pier [Wa...
Joined: 18 May 09
Posts: 39
Credit: 6098013
RAC: 0

RE: Dearie me, keep it

Message 95608 in response to message 95603

Quote:

Dearie me, keep it civil folks, we are all working for the good of humankind after all.

Can I just clarify something. I gather from above that the CUDA application might only use the GPU for a small part of the time - 4% was mentioned. Does this mean that it can be working on Collatz etc for the the other 96%? Or is it not that simple and the GPU needs to hang around waiting for Einstein?

If it is the case that my GPU will be in a state of enforced idleness then I think I might chose to set it to work on something that keeps it a little busier.

Nope, that's the problem with a GPU you can't devide 2 cuda project on 1 card.

That's why EINSTEIn should run on CPU totally or only CPU.

When you run E@H on both you would also keep a better return from the total machine. but atleast you could switch between projects on the GPU.

Now you can only get a & gain for the total of EInstein or run another project on the GPU. Wich would do more good for the machine.

Ps: EInstein team: there is a cuda 3 version coming, maybe make it so that the wu's are only GPU.
CUDA 3.0

Megacruncher
Megacruncher
Joined: 11 Feb 05
Posts: 7
Credit: 1270201748
RAC: 953119

Okay then. Nice try but

Okay then. Nice try but possibly only relevant to an Einstein mono-cruncher with a buckshee high end graphics card going to waste.
The rest of us will run Einstein on our CPUs and use our GPUs for something more generous and thought out. Thanks anyway!

 

 

MarkJ
MarkJ
Joined: 28 Feb 08
Posts: 437
Credit: 137324514
RAC: 18517

Well I got one unannounced

Well I got one unannounced but let it run through to see how it went anyway. The GPU in the machine is a GTX275. The workunit is here

It took approx 1/3rd off the "normal" run time, so its faster. As the others have pointed out at the expense of using both a cpu and a gpu for 5.5 hours. A 1.25 hour cpu time saving for 5.5 hours GPU time doesn't seem a reasonable use of the GPU to me.

Lionel SARRES
Lionel SARRES
Joined: 18 Oct 09
Posts: 21
Credit: 44218747
RAC: 38303

Hello, I have a Pentium IV

Message 95611 in response to message 95610

Hello,

I have a Pentium IV 3.06 (2 logical cores) with nVidia 9400 GT.
My first cuda WU took 15 hours to be treated with 1.00 CPU + 1.00 NVIDIA GPU.
And I got... 250 credits.

I had the same amount of credits (250) for one Pulsar Search treated with 1.00 CPU only running on my laptop - Core DUO at 2.00 GHz...

It's not worth it !

So lets resume the situation : Einstein cuda WU runs fine but the claimed credits is too low. That really doesn't help to get involved in the project. The amount of credits for Cuda WU using 2 cores (CPU and GPU) should be raised up.

Lionel

John Clark
John Clark
Joined: 4 May 07
Posts: 1087
Credit: 3143193
RAC: 0

Perhaps the best approach to

Perhaps the best approach to GPU crunching (whether CUDA - nVidia - or ATI) is to use the cards on projects which can use the GPU properly (MW, Collatz, etc) and use the rest of the CPUs on Einstein. That is the approach I am taking.

Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!

JockMacMad TSBT
JockMacMad TSBT
Joined: 5 Jun 09
Posts: 1
Credit: 211021
RAC: 0

The is also a change to the

The is also a change to the CUDA drivers coming when they support DX11 I beliee that will allow multiple CUDA processes on to a GPU the way ATI does at the moment.

Michael Goetz
Joined: 11 Feb 05
Posts: 21
Credit: 3060502
RAC: 0

RE: Einstein cuda WU runs

Message 95614 in response to message 95611

Quote:
Einstein cuda WU runs fine but the claimed credits is too low.

Actually, the credits are correct given the amount of work that's actualy being done. The problem is that the application is barely using the GPU at all. Even though the GPU is tied up for hours, most of the crunching is still being done on the CPU.

It's not that you're not getting credit for what your GPU is doing; the problem is that E@H is wasting the GPU time you're giving it and not doing much on the GPU.

Do what I (and lots of others) have done -- go to the E@H preferences and turn off GPU processing. Run E@H only on the CPU, and let some other project use the GPU.

Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG.

Mark Rush
Mark Rush
Joined: 18 Jan 05
Posts: 1
Credit: 594633302
RAC: 372962

I agree with others that

Message 95615 in response to message 95614

I agree with others that using the GPU only a fraction of the time is a waste of resources. I have been using my GPU exclusively for SETI. "Fortunately" SETI went down this weekend and I have worked through my backlog of SETI Cuda WUs, so the Einstein Cuda WU's impose no opportunity cost to me by forcing my GPU to set idel while I have uncrunched SETI Cuda WUs. When SETI is back up, I shall (sadly) select no GPU for Einstein. But I will keep an active eye out for ABP2 application to see if it more efficiently uses the GPU. And I'll be really eager for the "all Cuda" port because I would very much like to have both SETI and Einstein splitting time on my GPU.

Mark

Lionel SARRES
Lionel SARRES
Joined: 18 Oct 09
Posts: 21
Credit: 44218747
RAC: 38303

Hello, Is this the way we

Message 95616 in response to message 95614

Hello,

Is this the way we know that time is wasted :

My Cuda WU was created at 02:18:30 UTC and was crunched straight to 15:25:36. That makes 47226 sec. The result indicate 52288 sec.

Using one CPU and one GPU - two processing unit - should give
47226 x 2 = 94452, no ?

that makes a difference of 42164 sec... wasted ?

Is it correct ?

Bsts rgrds.

Lionel.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.