mfaktc: a CUDA program for Mersenne prefactoring - Page 309

kriesel · 2020-10-21, 16:56

Quote:

Originally Posted by Viliam Furik

Either way, I am not satisfied with the result.

I wonder what gpu-z, nvidia-smi or similar utility would show for the gpu utilization/load, and whether mfaktc.ini was tuned for the new fast gpus (>>128M GpuSieveSize, etc)
Also the classes output are numerous per second in the posted benchmark, which if on a rotating disk would slow things down; try an SSD or ramdisk or higher bit level or less_classes. Even on a well tuned RTX2080 I see throughput advantage to multiple mfaktc instances. These effects are slight but measurable even at GTX1050Ti. The faster the gpu, the stronger the effects is the trend I've observed on Windows over a gpu speed ratio of 10:1 or more.

petrw1 · 2020-10-21, 17:25

Quote:

Originally Posted by Viliam Furik

Why so poor result? I get about 4900 GHzD/D with the same work. It should be at least 10000 GHzD/D for the 3090, no? It has more than double the FP32 throughput.

It is most probably one of these two reasons:
1. The shared INT32 and FP32 cores don't play nicely with mfaktc - either incompatible code or the cores not fulfilling their promise
2. Memory bottleneck

Either way, I am not satisfied with the result.

Did you try the version that supports 2047 classes?
My 2080Ti went from 4,000 to 4,500 Ghz/Day with that version.

Viliam Furik · 2020-10-21, 17:30

Quote:

Originally Posted by petrw1

Did you try the version that supports 2047 classes?
My 2080Ti went from 4,000 to 4,500 Ghz/Day with that version.

Who, me? I am already using the 2047 version, as recommended by you. The mentioned 4900 GHz-D/D is from RTX 2080Ti, for the same workload as the 3090 was tested with.

James Heinrich · 2020-10-22, 02:28

Quote:

Originally Posted by Neutron3529

I bought a RTX 3090

Also, I'm still anxiously waiting for your gpuowl and cudalucas benchmarks, please...

moebius · 2020-10-22, 04:06

Quote:

Originally Posted by James Heinrich

Also, I'm still anxiously waiting for your gpuowl and cudalucas benchmarks, please...

And yet, the expected performance values for LL/PRP on the https://www.mersenne.ca/cudalucas.php page are still suspect to me. You can't seriously tell me a Tesla K-80 would be almost at the same Level as a Tesla P100 (PCIe 16GB).They are miles apart! Likewise,as much as I know, a Radeon VII isn't faster than a Tesla V100. In my opinion, this confuses the users of your site. I hope there are no bad purchases in the end?

James Heinrich · 2020-10-22, 04:15

Quote:

Originally Posted by moebius

And yet, the expected performance values on the https://www.mersenne.ca/cudalucas.php page are still suspect to me.

They might be. Hence my perpetual request for benchmarks.

Neutron3529 · 2020-10-22, 07:04

Quote:

Originally Posted by James Heinrich

I've looked at the code again and clearly I'm missing something because I think it should be working as I intended (but clearly it isn't). It's also difficult to test because that section of code will only get processed when a new factor is submitted (my logic works fine in my test environment, but something different is happening on the server). I have added a couple of debug lines that might help me track down the problem, if you see them next time you (collective "you", anyone reading this) submit a factor please email me either a copy-paste or screenshot of the output.

Do you want this?

Click image for larger version

Name: 2020-10-22_15-02-35.png
Views: 37
Size: 114.0 KB
ID: 23592

the server seems not to work with a TF result.

these two lines could trigger this BUG more than once and stop the upcoming results.

Code:

M104186261 has a factor: 21599873573633423090833 [TF:74:75:mfaktc 0.21 barrett76_mul32_gs]
found 1 factor for M104186261 from 2^74 to 2^75 [mfaktc 0.21 barrett76_mul32_gs]

You could test it until the BUG is removed.

---

Found 8 lines to process.
processing: TF factor 21599873573633423090833 for M104186261 (274-275) [range fully factored]
DEBUG: TF.range-complete credit(1) = 7.1037506824902
DEBUG: TF.range-complete credit(2) = 36.723102738304

aheeffer · 2020-10-22, 08:10

Quote:

Originally Posted by James Heinrich

. I have added a couple of debug lines that might help me track down the problem, if you see them next time you (collective "you", anyone reading this) submit a factor please email me either a copy-paste or screenshot of the output.

Code:

processing: TF factor 410022995157224015562287 for M333946237 (278-279) [range fully factored]
DEBUG: TF.range-complete credit(1) = 80.66706346138
DEBUG: TF.range-complete credit(2) = 183.31299318089
DEBUG: TF.range-complete credit(3) = 183.31299318089
DEBUG: TF.range-complete credit(4) = 183.31299318089
Please report these debug lines to james@mersenne.ca or post at https://www.mersenneforum.org/showthread.php?p=560500
CPU credit is 80.6671 GHz-days.

2M215856352p1 · 2020-10-22, 08:36

Quote:

Originally Posted by James Heinrich

I've looked at the code again and clearly I'm missing something because I think it should be working as I intended (but clearly it isn't). It's also difficult to test because that section of code will only get processed when a new factor is submitted (my logic works fine in my test environment, but something different is happening on the server). I have added a couple of debug lines that might help me track down the problem, if you see them next time you (collective "you", anyone reading this) submit a factor please email me either a copy-paste or screenshot of the output.

Here is another test case.

Neutron3529 · 2020-10-22, 12:46

Quote:

Originally Posted by James Heinrich

I've looked at the code again and clearly I'm missing something because I think it should be working as I intended (but clearly it isn't). It's also difficult to test because that section of code will only get processed when a new factor is submitted (my logic works fine in my test environment, but something different is happening on the server). I have added a couple of debug lines that might help me track down the problem, if you see them next time you (collective "you", anyone reading this) submit a factor please email me either a copy-paste or screenshot of the output.

processing: TF no-factor for M114899527 (274-275)
CPU credit is 33.2990 GHz-days.
processing: TF factor 23024239007594549773417 for M114899501 (274-275) [range fully factored]
DEBUG: TF.range-complete credit(1) = 9.5092580373447
DEBUG: TF.range-complete credit(2) = 33.29903727452
DEBUG: TF.range-complete credit(3) = 33.29903727452
DEBUG: TF.range-complete credit(4) = 33.29903727452
Please report these debug lines to james@mersenne.ca or post at https://www.mersenneforum.org/showthread.php?p=560500CPU credit is 9.5093 GHz-days.
processing: TF factor 21599873573633423090833 for M104186261 (274-275) [range fully factored]
DEBUG: TF.range-complete credit(1) = 7.1037506824902
DEBUG: TF.range-complete credit(2) = 36.723102738304
DEBUG: TF.range-complete credit(3) = 36.723102738304
DEBUG: TF.range-complete credit(4) = 36.723102738304
Please report these debug lines to james@mersenne.ca or post at https://www.mersenneforum.org/showthread.php?p=560500Already have factor 21599873573633423090833 for M104186261 CPU credit is 7.1038 GHz-days.
Done processing:
* Parsed 44 lines.
* Found 0 datestamps.

James Heinrich · 2020-10-22, 12:57

Thanks guys, I think (again) that I've found the problem. In this case my code was working as expected and overriding the correct amount of credit, but the message displayed to the user was pre-written elsewhere using the default credit amount, I just needed to also rewrite the user message. Can one of the people who've posted above (or anyone with a new TF factor in a fully-factored range reported in the last 8h or more recently) confirm if the GHz-days credit in your Account Result Details page shows the higher (correct) or lower (incorrect) amount of credit for the TF-F (range-fully-factored) result?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
mfakto: an OpenCL program for Mersenne prefactoring	Bdot	GPU Computing	1668	2020-12-22 15:38
The P-1 factoring CUDA program	firejuggler	GPU Computing	753	2020-12-12 18:07
gr-mfaktc: a CUDA program for generalized repunits prefactoring	MrRepunit	GPU Computing	32	2020-11-11 19:56
mfaktc 0.21 - CUDA runtime wrong	keisentraut	Software	2	2020-08-18 07:03
World's second-dumbest CUDA program	fivemack	Programming	112	2015-02-12 22:51

2020-10-22, 12:57	#3399
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 3244₁₀ Posts	Thanks guys, I think (again) that I've found the problem. In this case my code was working as expected and overriding the correct amount of credit, but the message displayed to the user was pre-written elsewhere using the default credit amount, I just needed to also rewrite the user message. Can one of the people who've posted above (or anyone with a new TF factor in a fully-factored range reported in the last 8h or more recently) confirm if the GHz-days credit in your Account Result Details page shows the higher (correct) or lower (incorrect) amount of credit for the TF-F (range-fully-factored) result?