mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2020-12-10, 15:13   #1662
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

132448 Posts
Default

Quote:
Originally Posted by axn View Post
He's using standard Primenet conversion 1Gd/d = 2GFLOPS
Hmm, wouldn't that be based on Core2Duo instruction set and performance? By definition? https://www.mersenneforum.org/showpo...67&postcount=4

It might be considered standard, but it seems to me far from valid for application to a gpu with a completely different architecture and instruction set and its own performance constraints.
kriesel is offline   Reply With Quote
Old 2020-12-10, 18:15   #1663
axn
 
axn's Avatar
 
Jun 2003

19·271 Posts
Default

Quote:
Originally Posted by kriesel View Post
It might be considered standard, but it seems to me far from valid for application to a gpu with a completely different architecture and instruction set and its own performance constraints.
As long as the algorithm used is same-ish (i.e. IBDWT), the architectural differences are irrelevant.

Given that the algorithm is same, 1 iteration at a given FFT size takes x floating point operations, snd will get you y GHzD credit. This won't change because of the processor. Only thing that is affected by the processor is the time to complete that iteration; the faster a processor completes an iteration, the higher its GHzD/d rating, but also its GFLOPS rating. But the conversion factor between these two doesn't change and will remain independent of the processor.
axn is offline   Reply With Quote
Old 2020-12-10, 19:16   #1664
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22×32×7×23 Posts
Default

Mfakto is performing TF, by a variety of kernels, not FFT multiplication by IBDWT.
FFT almost always uses DP. TF kernels may use SP or int. Different hardware designs have different ratios among them. Gpus have vastly different DP/SP or I think DP/int32 ratios than cpus.
kriesel is offline   Reply With Quote
Old 2020-12-10, 19:40   #1665
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

138816 Posts
Default

So what?
Each TF bit takes some specific amount of work. You're complaining about how we measure work done, but not suggesting some alternative.
VBCurtis is offline   Reply With Quote
Old 2020-12-11, 02:45   #1666
axn
 
axn's Avatar
 
Jun 2003

19·271 Posts
Default

Quote:
Originally Posted by kriesel View Post
Mfakto is performing TF, by a variety of kernels, not FFT multiplication by IBDWT.
FFT almost always uses DP. TF kernels may use SP or int. Different hardware designs have different ratios among them. Gpus have vastly different DP/SP or I think DP/int32 ratios than cpus.
Well, whoops. I guess I don't remember the context of posts from two months ago that well :-(

All good points; you can inform OP to take those into consideration and adjust their calculations.
axn is offline   Reply With Quote
Old 2020-12-20, 16:54   #1667
DrobinsonPE
 
Aug 2020

25·3 Posts
Default

GB-BRi5H-8250, i508250U, UHD 620, 16GB DDR-4, SSD, Windows 10.

mfakto 0.15pre7

Code:
C:\Users\user\mfakto\015pre7>mfakto -st

Selftest statistics
  number of tests           34026
  successful tests          33288
  no factor found           738

selftest FAILED!
I need to find what exponent levels it is failing on and see if there is just a range to avoid.

Code:
C:\Users\user\mfakto\015pre7>mfakto
mfakto 0.15pre7-MGW (64bit build)
OpenCL device info
  name                      Intel(R) UHD Graphics 620 (Intel(R) Corporation)
  device (driver) version   OpenCL 2.1 NEO  (27.20.100.8681)
  maximum threads per block 256
  maximum threads per grid  16777216
  number of multiprocessors 24 (24 compute elements)
  clock rate                1100 MHz
Automatic parameters
  threads per grid          0
  optimizing kernels for    INTEL
selftest PASSED!
got assignment: exp=115746439 bit_min=73 bit_max=74 (16.53 GHz-days)
Starting trial factoring M115746439 from 2^73 to 2^74 (16.53 GHz-days)
Using GPU kernel "cl_barrett32_76_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Dec 20 08:18 |    0   0.1% | 60.505  16h07m |     24.58    81206    0.00%
Dec 20 08:19 |    5   0.2% | 60.985  16h13m |     24.39    81206    0.00%
Dec 20 08:20 |    9   0.3% | 61.054  16h13m |     24.36    81206    0.00%
Dec 20 08:21 |   12   0.4% | 61.219  16h15m |     24.30    81206    0.00%
Dec 20 08:22 |   17   0.5% | 61.204  16h14m |     24.30    81206    0.00%
DrobinsonPE is offline   Reply With Quote
Old 2020-12-22, 15:34   #1668
DrobinsonPE
 
Aug 2020

25×3 Posts
Default

Quote:
Originally Posted by DrobinsonPE View Post
GB-BRi5H-8250, i508250U, UHD 620, 16GB DDR-4, SSD, Windows 10.

mfakto 0.15pre7

Code:
C:\Users\user\mfakto\015pre7>mfakto -st

Selftest statistics
  number of tests           34026
  successful tests          33288
  no factor found           738

selftest FAILED!
I need to find what exponent levels it is failing on and see if there is just a range to avoid.
All 738 -st errors are in the 61-62, 62-63, and 63-64 ranges.
DrobinsonPE is offline   Reply With Quote
Old 2020-12-22, 15:38   #1669
DrobinsonPE
 
Aug 2020

9610 Posts
Default

I5-4570T with HD4600 Graphics

mfakto 0.15pre7

Code:
C:\Users\user\mfakto>mfakto
mfakto 0.15pre7-MGW (64bit build)
Starting trial factoring M115801657 from 2^73 to 2^74 (16.52 GHz-days)
Using GPU kernel "cl_barrett32_76_gs_2"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Dec 22 06:48 |   12   0.4% | 89.923  23h52m |     16.53    81206    0.00%
DrobinsonPE is offline   Reply With Quote
Old 2021-06-30, 18:17   #1670
birtwistlecaleb
 
birtwistlecaleb's Avatar
 
Jun 2021

23×7 Posts
Default

Quote:
Originally Posted by James Heinrich View Post
Perhaps this one?
https://download.mersenne.ca/mfakto/mfakto-0.12-hd4000

edit: Although based on what Ken says below, you're likely fine with the normal latest-version (0.15-pre6):
https://download.mersenne.ca/mfakto/mfakto-0.15pre6
They both do not have a worktodo.txt, and it seems like they are broken because of that.
birtwistlecaleb is offline   Reply With Quote
Old 2021-06-30, 18:25   #1671
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

24·3·73 Posts
Default

Quote:
Originally Posted by birtwistlecaleb View Post
They both do not have a worktodo.txt, and it seems like they are broken because of that.
What do you mean? No program will come with worktodo.txt, that's what you supply with the assignments you're working on.
What error message(s) do you see when you run whichever version it is you're running?
James Heinrich is offline   Reply With Quote
Old 2021-06-30, 19:49   #1672
birtwistlecaleb
 
birtwistlecaleb's Avatar
 
Jun 2021

1110002 Posts
Default

Quote:
Originally Posted by James Heinrich View Post
What do you mean? No program will come with worktodo.txt, that's what you supply with the assignments you're working on.
What error message(s) do you see when you run whichever version it is you're running?
They both instantly close when they find that, so I can't see that. Am I supposed to manually add the file?
birtwistlecaleb is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
mfaktc: a CUDA program for Mersenne prefactoring TheJudger GPU Computing 3509 2021-10-22 11:54
gpuOwL: an OpenCL program for Mersenne primality testing preda GpuOwl 2733 2021-10-13 10:39
LL with OpenCL msft GPU Computing 433 2019-06-23 21:11
OpenCL for FPGAs TObject GPU Computing 2 2013-10-12 21:09
Program to TF Mersenne numbers with more than 1 sextillion digits? Stargate38 Factoring 24 2011-11-03 00:34

All times are UTC. The time now is 11:52.


Sat Oct 23 11:52:50 UTC 2021 up 92 days, 6:21, 0 users, load averages: 1.43, 1.36, 1.20

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.