mfaktc: a CUDA program for Mersenne prefactoring - Page 129

TheJudger · 2011-12-10, 00:13

Quote:

Originally Posted by TheJudger

Factory overclocked GTX 560Ti (1701MHz), barrett79 kernel, raw GPU speed (without sieving), M66362159 from 2⁶⁹ to 2⁷⁰

Code:

                 |  CUDA 3.2 | CUDA 4.1-RC1
mfaktc 0.17      | 260.94M/s |    261.93M/s
mfaktc 0.18-pre9 | 260.80M/s |    258.97M/s

Factory overclocked GTX 560Ti (1701MHz), barrett79 kernel, raw GPU speed (without sieving), M66362159 from 2⁶⁹ to 2⁷⁰

Code:

                  |  CUDA 3.2 | CUDA 4.1-RC2
mfaktc 0.17       | 260.94M/s |    261.93M/s
mfaktc 0.18-pre10 | 260.80M/s |    265.39M/s

A little bit better than before

but there are no changes in the code of the barrett79 kernel from -pre9 to -pre10...

Factory overclocked GTX 560Ti (1701MHz), barrett92 kernel, raw GPU speed (without sieving), M3321932839 from 2⁷⁹ to 2⁸⁰

Code:

                  | CUDA 4.1-RC2
mfaktc 0.17       |    170.62M/s
mfaktc 0.18-pre10 |    173.32M/s

A little bit faster, too. But the difference between compute capability 2.0 and 2.1 increases further...

Oliver

TheJudger · 2011-12-19, 23:26

Hello!

http://www.mersenneforum.org/mfaktc/mfaktc-0.18.tar.gz
http://www.mersenneforum.org/mfaktc/mfaktc-0.18.win.zip
http://www.mersenneforum.org/mfaktc/...linux64.tar.gz

The executables need at least a CUDA 4.0 capable driver (270 series driver or newer). The Windows zip archive contains both, the 32 bit and 64 bit version. I'll upload new executables once CUDA 4.1 is public available. The sources should compile with older CUDA version, too, but they might be slower. CUDA 4.1 will give another performance improvement for the barrett based kernels on compute capability 2.x GPUs (especially on 2.0).

Compared to mfaktc 0.17 there are "more than usuall" minor changes. Highlights from the Changelog.txt:

autoadjustment of SievePrimes is now less dependend on the gridsize and
absolute speed. Instead of measuring the absolute (average) time waited
per precessing block (grid size) now the relative time spent on waiting
for the GPU is calculated. In the per-class output "avg. wait" is replaced
by "CPU wait".
new commandline option: "-v" (verbosity) let the user decide how many
informations are printed
(suggested by aspen on www.mersenneforum.org)
"has a factor" result lines now contain informations (program name,
versions, bitlevel, ...) James Heinrich is working on this on the server
side. This should give more accurate credits for "has a factor" results
from the primenet server once this is fully implemented.
mfaktc no longer refuses to load a checkpoint file from a Linux version
with a Windows version of mfaktc and vice versa. Of course mfaktc still
refuses to load checkpoint files from other versions than itself
(identical version string!)
added a (simple) signal handler (captures SIGINT and SIGTERM).
1st ^C: mfaktc will exit after the currently processed class is finished.
2nd ^C: mfaktc will stop immediately
added a minimum delay between two checkpoint file writes. The user can set
the delay in mfaktc.ini (CheckpointDelay).
added a new code path to barrett79_mul32 and barrett92_mul32 kernels, CUDA
>= 4.1 features multiply-add with carry for compute capability >= 2.0.
On my GTX 470 (compute capability) this yields up to 15% for
barrett92_mul32 and up to 7% for barrett79_mul32 extra throughput.

As usuall: finish your current assignments with your current version and do the update after it, mfaktc 0.18 will refuse foreign checkpoint files.

Oliver

kladner · 2011-12-20, 00:43

Many thanks, sir! I am impatient for my current assignments to finish so that I can put this version into service.

Dubslow · 2011-12-20, 01:34

Would you mind posting the .dll/.so s on the mfatkc mirror? I'd rather not have to download the whole CUDA environment...

LaurV · 2011-12-20, 03:47

Quote:

Originally Posted by TheJudger

...mfaktc 0.18...

Output file (results.txt) customizable from the ini file? (including the path, for collecting all the results from all running processes of mfaktc in a single file).

diamonddave · 2011-12-20, 04:04

Quote:

Originally Posted by TheJudger

[*]"has a factor" result lines now contain informations (program name,
versions, bitlevel, ...) James Heinrich is working on this on the server
side. This should give more accurate credits for "has a factor" results
from the primenet server once this is fully implemented.

Many thanks! Can't wait to test this feature with a new exponent!

kladner · 2011-12-20, 05:05

The new version seems to be working well. At least, there have been no problems reported.

TheJudger · 2011-12-20, 11:13

Quote:

Originally Posted by Dubslow

Would you mind posting the .dll/.so s on the mfatkc mirror? I'd rather not have to download the whole CUDA environment...

They are included in the archives for the executables, aren't they?

Quote:

Originally Posted by LaurV

Output file (results.txt) customizable from the ini file? (including the path, for collecting all the results from all running processes of mfaktc in a single file).

Well, I'm still unsure about this feature. Personally I don't like it but it seems that you and some others want it. Bdot (mfakto) tries to convince me, too.

So I guess I'll add this for 0.19?

Oliver

James Heinrich · 2011-12-20, 13:55

Quote:

Originally Posted by TheJudger

Well, I'm still unsure about this feature. Personally I don't like it but it seems that you and some others want it. Bdot (mfakto) tries to convince me, too.

I also think it would be good to have as a configurable option. Naturally you'll need to lock the file for writing for the split second it takes to write the result line so two instances don't try and write at the same time.

Along the same lines, a unified worktodo.txt would also be nice, perhaps split into [Worker #1], [Worker #2], etc sections. This is of course a little more work than a configurable results.txt, but lets us just deal with one in and one out for each machine, in a format that's already familiar to us from Prime95.

Even better would be to optimize/thread the sieving such that we'd only ever need to run a single mfaktc instance (sieving would spread across as many CPU cores as needed to feed the GPU(s). But that's a whole other set of complications for a much later release.

Chuck · 2011-12-20, 14:47

Great! Thanks for the update. I've got two instances running now.

kladner · 2011-12-20, 15:07

This was rather a quick test, showing the difference between mfaktc .17 and .18. V.18 did eventually drop to SievePrimes 5000, though the time didn't really change that much.

EDIT: These were run with the same exponent in single instances.

2011-12-19, 23:26	#1410
TheJudger "Oliver" Mar 2005 Germany 2127₈ Posts	mfaktc 0.18 Hello! http://www.mersenneforum.org/mfaktc/mfaktc-0.18.tar.gz http://www.mersenneforum.org/mfaktc/mfaktc-0.18.win.zip http://www.mersenneforum.org/mfaktc/...linux64.tar.gz The executables need at least a CUDA 4.0 capable driver (270 series driver or newer). The Windows zip archive contains both, the 32 bit and 64 bit version. I'll upload new executables once CUDA 4.1 is public available. The sources should compile with older CUDA version, too, but they might be slower. CUDA 4.1 will give another performance improvement for the barrett based kernels on compute capability 2.x GPUs (especially on 2.0). Compared to mfaktc 0.17 there are "more than usuall" minor changes. Highlights from the Changelog.txt: autoadjustment of SievePrimes is now less dependend on the gridsize and absolute speed. Instead of measuring the absolute (average) time waited per precessing block (grid size) now the relative time spent on waiting for the GPU is calculated. In the per-class output "avg. wait" is replaced by "CPU wait". new commandline option: "-v" (verbosity) let the user decide how many informations are printed (suggested by aspen on www.mersenneforum.org) "has a factor" result lines now contain informations (program name, versions, bitlevel, ...) James Heinrich is working on this on the server side. This should give more accurate credits for "has a factor" results from the primenet server once this is fully implemented. mfaktc no longer refuses to load a checkpoint file from a Linux version with a Windows version of mfaktc and vice versa. Of course mfaktc still refuses to load checkpoint files from other versions than itself (identical version string!) added a (simple) signal handler (captures SIGINT and SIGTERM). 1st ^C: mfaktc will exit after the currently processed class is finished. 2nd ^C: mfaktc will stop immediately added a minimum delay between two checkpoint file writes. The user can set the delay in mfaktc.ini (CheckpointDelay). added a new code path to barrett79_mul32 and barrett92_mul32 kernels, CUDA >= 4.1 features multiply-add with carry for compute capability >= 2.0. On my GTX 470 (compute capability) this yields up to 15% for barrett92_mul32 and up to 7% for barrett79_mul32 extra throughput. As usuall: finish your current assignments with your current version and do the update after it, mfaktc 0.18 will refuse foreign checkpoint files. Oliver

2011-12-20, 00:43	#1411
kladner "Kieren" Jul 2011 In My Own Galaxy! 10158₁₀ Posts	Kudos! Many thanks, sir! I am impatient for my current assignments to finish so that I can put this version into service.

2011-12-20, 01:34	#1412
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3·29·83 Posts	Would you mind posting the .dll/.so s on the mfatkc mirror? I'd rather not have to download the whole CUDA environment... Last fiddled with by Dubslow on 2011-12-20 at 01:34

2011-12-20, 15:07	#1419
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	.17 vs .18 This was rather a quick test, showing the difference between mfaktc .17 and .18. V.18 did eventually drop to SievePrimes 5000, though the time didn't really change that much. EDIT: These were run with the same exponent in single instances. Attached Thumbnails Last fiddled with by kladner on 2011-12-20 at 15:09

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
mfakto: an OpenCL program for Mersenne prefactoring	Bdot	GPU Computing	1676	2021-06-30 21:23
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

2011-12-20, 05:05	#1415
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	The new version seems to be working well. At least, there have been no problems reported.

2011-12-20, 14:47	#1418
Chuck May 2011 Orange Park, FL 1565₈ Posts	Great! Thanks for the update. I've got two instances running now.