Bigger and better GPU sieving drive: Discussion

[vaughan]

Help please:
What does the -t# part of the batch file do? I thought it was used if you did CPU work not GPU CUDA work and it was set to the number of cores available.

I'm running this on an AMD Athlon X2 6000 CPU under Windows XP-64bit and it is crunching two instances (separate folders) of LLRnet for NPLB on Port = 3000. The GPU is an Nvidia GeForce GTX 260 (as confirmed with GPU-z version 0.4.7).

My batch file is:
tpsieve-cuda-boinc-x86-windows.exe -p20000G -P20500G -k 5 -K 9999 -n 2M -N3M -ffppr3M_20000G-20500G.txt -M2 -q

I removed the reference to -t# as it gave an error message and the application would not run.

I see a warning message in the DOS box that says:

Didn't change nstep from 30

Should I be worried by this message?

Also, it doesn't give any output progress on screen but if I Ctrl-C the application I can see the factors file has been created and it contains factors in the relevant range (20000G - 20500G) for this computer. Perhaps I'm not seeing any progress updates as I'm viewing this PC remotely using LogMeIn Ignition and last time when I was running the previous sieving effort using ppsieve only 1 of 4 computers displayed the 1 minute progress updates when viewed with LogMeIn.

[mdettweiler]

The -t# switch is for CPU versions of tpsieve only; Ken has been doing some experimental stuff with using -t to run on multiple GPUs with ppsieve, but I don't know if those changes have made it into tpsieve yet.

Don't worry about the "Didn't change nstep from 30" message; that's just an "info" message that's useful to the developers but not really to anyone else.

As for the progress, that's a side effect of the fact that you're using the "BOINC" binary (which can be run either through BOINC or in standalone mode, as you're doing here). Try running the "tpsieve-cuda-x86-windows.exe" binary instead and you should see progress.

[Ken_g6]

Originally Posted by mdettweiler

The -t# switch is for CPU versions of tpsieve only; Ken has been doing some experimental stuff with using -t to run on multiple GPUs with ppsieve, but I don't know if those changes have made it into tpsieve yet.

Yes, it has. Although I'm not sure anyone's tested it yet. (I'm looking at you, frmky.) And it wouldn't work with the BOINC binary anyway.

[PCZ]

I use -t2 with dual cards.
Spits out the same factors as a single card just twice as fast.
My dual GPU machines have identical cards.

I did see that Ken had some concerns about running multi threads on the GPU client and did some tests myself.

The factors found were the same running single or multi GPU.
Speed up is linear too which is nice.

I will test tpsieve with a small range on a single card and then a pair of cards.
Dont expect any issues.

Ran a quick test file attached.

Factors appear to be the same although the order they are listed is slightly different.
I expect this is because there are multi threads writing to the file and they are working on different numbers in parallel.

[firejuggler]

from a radeon serie 4800
windows 32 bit

ppsieve-cl-x86-windows -p42070e9 -P42070030e6 -k 1201 -K 9999 -N 2000000 -c 60
ppsieve version cl-0.2.0-beta (testing)
nstart=76, nstep=32
ppsieve initialized: 1201 <= k <= 9999, 76 <= n < 2000000
Sieve started: 42070000000000 <= p < 42070030000000
Thread 0 starting
Detected 160 multiprocessors (800 SPUs) on device 0.
42070000070587 | 9475*2^197534+1
42070000198537 | 3373*2^1046686+1
42070003101727 | 4207*2^1054290+1
42070003511309 | 6057*2^1043547+1
42070006307657 | 1513*2^1771812+1
42070006388603 | 2059*2^1816098+1
42070007177519 | 5437*2^1121592+1
42070007396759 | 7339*2^1803518+1
42070008823897 | 4639*2^952018+1
42070008858187 | 2893*2^317690+1
42070010190569 | 5625*2^1903125+1
42070011430123 | 3821*2^1406279+1
42070012301263 | 1957*2^1185814+1
42070013521999 | 1965*2^404493+1
42070013970587 | 7143*2^1462422+1
42070013989247 | 5037*2^838603+1
42070017332953 | 6237*2^1916994+1
42070018235321 | 1941*2^363948+1
42070019542387 | 8587*2^1703626+1
42070023987581 | 9811*2^318944+1
42070024339237 | 9257*2^1170495+1
42070024532551 | 4311*2^1690093+1
42070024936837 | 5679*2^1726142+1
42070024995961 | 9111*2^1707153+1
42070026021997 | 4039*2^1819590+1
42070027452199 | 1323*2^854008+1
42070029006583 | 5943*2^663870+1

Thread 0 completed
Waiting for threads to exit
Sieve complete: 42070000000000 <= p < 42070030000000
Found 27 factors
count=955289,sum=0x2dbc17167afb6a8d
Elapsed time: 31.69 sec. (0.02 init + 31.67 sieve) at 951840 p/sec.
Processor time: 2.53 sec. (0.03 init + 2.50 sieve) at 12058624 p/sec.
Average processor utilization: 2.00 (init), 0.08 (sieve)

just to see the speed

[vaughan]

Originally Posted by mdettweiler

...
As for the progress, that's a side effect of the fact that you're using the "BOINC" binary (which can be run either through BOINC or in standalone mode, as you're doing here). Try running the "tpsieve-cuda-x86-windows.exe" binary instead and you should see progress.

Should I switch to the non-BOINC version or will that upset the sieving?

[Ken_g6]

Originally Posted by vaughan

Should I switch to the non-BOINC version or will that upset the sieving?

Yes! But...

Yes, you should switch as soon as is practical. The non-BOINC version would read the same checkpoint files as the BOINC version.

But...I believe the BOINC version, running as a standalone client, will never write a checkpoint file! Except maybe if you hit Ctrl-C.

But if it doesn't you can probably start sieving with the non-BOINC version from the last factor in the file.

[AMDave]

or ... not until you have finished the current range.
then when the unknowns are no longer a factor, make the switch.

... just hoping to make the hind-sight experience a better one :)

[vaughan]

Very strange. I connected remotely from QLD tonight and found the application wasn't running and the GPU was at zero load. Checked the factors file and it was only around the 20323G mark so it still has a way to go.

I restarted it using the same application (as AMDave posted). I didn't want to risk upsetting the work already done. When I finish this range I'll switch to the non-boinc version for the next run.

[Ken_g6]

That could happen because of a Computation Error. Check stderr.txt. It won't happen for the non-BOINC client - it will print "Computation Error" but it will keep going.

How stable is that GPU?