Folding Forum

Hi, I have a Win7 32bit Core2 Quad Core Q6600 2.4GHz 4GB RAM.
Sapphire HD6570 GPU.
Client V7

Project:8013
My CPU gets between 400 & 650 credits and can do a WU in 10-12 hours. SMP option set to 2 not -1 (I see no change in credits or speed when I did this, my cores are now at 70-95%)

Project:11293
My GPU gets 1838 credits all the time and takes about 10-12 hours to do a WU.

Is this about right for my rig?

p24601 wrote:Project:8013
My CPU gets between 400 & 650 credits and can do a WU in 10-12 hours. SMP option set to 2 not -1 (I see no change in credits or speed when I did this, my cores are now at 70-95%)

SMP set to 2 so only 2 cores are used? If only 2 cores are used then yes it sounds about right.

Thanks.

I see no reason why you can't set your system or smp 3 and get a bit more folding done by SMP. That's what you would be doing if your system had an AMD X3 rather than an Intel Quad.

codysluder wrote:I see no reason why you can't set your system or smp 3 and get a bit more folding done by SMP. That's what you would be doing if your system had an AMD X3 rather than an Intel Quad.

I thought the client didn't like odd numbers?

iceman1992 wrote:I thought the client didn't like odd numbers?

More accurately, the SMP folding cores don't work well with "large primes". How large? That is not real clear. Definite problems show up with settings of 11, 13 or larger prime numbers or their multiples. WU's from some projects show issues with a SMP setting of 7. There are a few reports connected with the use of 5, but no documented problems with a setting of 3.

Yet.

Joe_H wrote:More accurately, the SMP folding cores don't work well with "large primes".

So it will work well with 9 since it's not a prime?

Define "working well" ? Why 9? You have a 10 core box running a GPU client? (possible, but not likely)

Non primes work better than primes, even numbers work better than odd numbers. YMMV.

7im wrote:Define "working well" ? Why 9? You have a 10 core box running a GPU client? (possible, but not likely)

Non primes work better than primes, even numbers work better than odd numbers. YMMV.

Working well = no problems. No I don't, I'm just curious why it behaves that way.

Forum search "domain decomposition"

7im wrote:Forum search "domain decomposition"

Okay

7im wrote:Yet.

If a documented issue with a setting of 3 can be shown, there will be a lot of pressure on the code maintainers (folding cores and Gromacs) to identify and fix the problem. That is because mathematically and computationally, domain decomposition that has a problem with 3 will also have problems with any multiple of 3. Given the number of hex-core, and multiple CPU servers with them, that would be significant effect on WU processing. It also would show up with quad-core w/HT or equivalent setups that reserve a core for GPU or personal use. So far that has not been the case, issue reports with settings of 6, 12 or 24 have not been coming in.

iceman1992 wrote:So it will work well with 9 since it's not a prime?

In theory it should work. How well is another issue, and waits on reports of actual experience of persons using that setting. In practice 9 happens to not be a very useful number given the normally available multi-core processors and how they get used in single or multi-processor machines.

No more pressure than fixing that recent project that started bombing with -smp 7 when most other projects work fine. That caused a lot of trouble with SMP users also folding on a GPU. But nothing changed. Like you, I don't forsee a problem, but that doesn't preclude one...

Hi, my CPU is now doing project 7809 and ETA was (I think it was)2.75 days. I had 1.05 days left when I turned SMP back from 2 to -1.
I've checked this morning and the ETA was 3.75 days what?! I've now put the SMP to 2 again, shut down the client and restarted waited a few minutes and the ETA is now 20 hours 46 minutes

So what happened?