mersenneforum.org gpuOwL: an OpenCL program for Mersenne primality testing
 Register FAQ Search Today's Posts Mark Forums Read

2020-08-06, 00:53   #2388
storm5510
Random Account

Aug 2009
U.S.A.

2×751 Posts

Quote:
 Originally Posted by paulunderwood It looks like the cofactor is given: 765044109502655639249
I reserved two more to actually run. "First PRP tests on Mersenne cofactors." I remember factors being in quotes for ECM's. I had an idea that the number above was similar in purpose.

Quote:
 Originally Posted by Google A Cofactor, in mathematics, is used to find the inverse of the matrix, adjoined. The Cofactor is the number you get when you remove the column and row of a designated element in a matrix, which is just a numerical grid in the form of rectangle or a square.
Alright, I will take their word for it.

 2020-08-06, 01:09 #2389 Uncwilly 6809 > 6502     """"""""""""""""""" Aug 2003 101×103 Posts 2×4,297 Posts 15 = 5 * 3 5 = factor 3 = cofactor 1260 = 7 * 5 * 36 7 = prime factor 5 = prime factor 36 = composite cofactor (will not be a PRP prime.)
 2020-08-06, 01:11 #2390 paulunderwood     Sep 2002 Database er0rr 339410 Posts https://www.thefreedictionary.com/cofactor "1. One of two or more contributing factors."
2020-08-06, 08:23   #2391
preda

"Mihai Preda"
Apr 2015

100101001112 Posts

Quote:
 Originally Posted by storm5510 The worktodo line said "PRP" only. The result line says "PRP-3." Below is a test I pulled from Primenet for demonstration only. I unreserved it later. Code: PRP=xxxx,1,2,10369241,-1,99,0,"765044109502655639249" The assignment in my account also reads "PRP." The reservation was from "Double-check tests on Mersenne cofactors." I do not know how a determination is made based on the work line as to the type.
It seems GpuOwl accepts the PRP line with the cofactor present, but GpuOwl not being aware of or using the cofactor in any way may be sub-optimal. GpuOwl will simply run a classic PRP test, which guess what will turn out "not prime", but that was already known at the start.

MPrime OTH may make some use of the cofactor by running a PRP-CF test.

(best use GpuOwl for "first time PRP" or similar)

Last fiddled with by preda on 2020-08-06 at 08:24

2020-08-06, 13:05   #2392
storm5510
Random Account

Aug 2009
U.S.A.

2·751 Posts

Quote:
 Originally Posted by preda It seems GpuOwl accepts the PRP line with the cofactor present, but GpuOwl not being aware of or using the cofactor in any way may be sub-optimal...
In simpler terms, the cofactor is ignored.

Quote:
 Originally Posted by Uncwillly 5 = 5 * 3 5 = factor 3 = cofactor 1260 = 7 * 5 * 36 7 = prime factor 5 = prime factor 36 = composite cofactor (will not be a PRP prime.)
Leave it to Google to complicate something simple.

2020-08-07, 16:10   #2393
R. Gerbicz

"Robert Gerbicz"
Oct 2005
Hungary

25618 Posts

Quote:
 Originally Posted by preda It seems GpuOwl accepts the PRP line with the cofactor present, but GpuOwl not being aware of or using the cofactor in any way may be sub-optimal. GpuOwl will simply run a classic PRP test, which guess what will turn out "not prime", but that was already known at the start. MPrime OTH may make some use of the cofactor by running a PRP-CF test. (best use GpuOwl for "first time PRP" or similar)
Basically that prp-cf test it the same as the prp test, just in the end there is a big division.
Even if the server already accepted such runs from your gpuowl (is it/was it possible?), then it is still not a mistake, if we stored the res2048 residue, see: https://www.mersenneforum.org/showthread.php?t=23462 . As if d=1 then prp=prp cf.

2020-08-07, 17:09   #2394
storm5510
Random Account

Aug 2009
U.S.A.

2·751 Posts

Changing test types for just a minute. I ran this:

Quote:
 PFactor=xxxx,1,2,99771479,-1,77,2
The bounds, 600,000 and 30,000,000 respectively. Suggested by preda, I reduced the B1 and left B2 at it default. Below was the result of trying to begin Stage 2:

Quote:
 Exception gpu_error: MEM_OBJECT_ALLOCATION_FAILURE clEnqueueCopyBuffer(queue.src, dst, 0, 0, size, 0, NULL, NULL) at clwrap.cpp:339 copybuf
I found the issue, maxAlloc. I had it too high, 7,000. I dropped it to 6,000 and it went on the Stage 2. Caveat: For a B2 of this size, at least 12 hours, perhaps more.

Some Nvidia GPU's are better suited for running specific tasks, while different models may be better with others. Mine, it has something to do with integer math, or type of integer math. It is great at running TF. Anything else, not so much.

2020-08-19, 13:37   #2395
storm5510
Random Account

Aug 2009
U.S.A.

2·751 Posts

When I got up this morning, I found gpuOwl has stopped running during the early AM. There was no visual indication on the screen, but my widget showed the GPU as being at idle temperature. The GPU's RAM was still allocated. I looked at the system logs. There was nothing relative at the time it stopped.

I attempted a stop by using Ctrl-C. gpuOwl appeared to resume running as it displayed another output line on the screen and the GPU began to heat up. A second Ctrl-C stopped it properly. I restarted it and finished the test.

The stoppage happened most of the way through Stage 2 of a P-1 test. Below are some particulars:

Quote:
 M99785863 (Completed and submitted) B1 = 600,000 B2 = 20,000,000 maxAlloc 6,000 of a possible 8,000 gpuOwl version 6.11-364-g36f4e2a Windows 10 Pro x64 v1909 Command Prompt console, (not PowerShell).
One thing prior to this yesterday evening was a requested restart by Window 10 to apply update(s). There were no problems.

I restarted again as I have two additional assignments very close in size to the one where the stoppage occurred. I changed no parameters. I want to see if there will be a repeat of the problem.

 2020-08-21, 04:45 #2396 paulunderwood     Sep 2002 Database er0rr 2×1,697 Posts 379 & 3.7.0 I have just upgraded to the latest commit and Rocm 3.7.0 and am seeing a ~0.5% speed up.
2020-08-21, 06:13   #2397
preda

"Mihai Preda"
Apr 2015

3×397 Posts

Quote:
 Originally Posted by paulunderwood I have just upgraded to the latest commit and Rocm 3.7.0 and am seeing a ~0.5% speed up.
Upgraded from which ROCm? 2.2 or 2.3?

2020-08-21, 12:02   #2398
paulunderwood

Sep 2002
Database er0rr

339410 Posts

Quote:
 Originally Posted by preda Upgraded from which ROCm? 2.2 or 2.3?
Rocm 3.5.1. I have to hack the Makefile, now to Rocm 3.7.0.

 Similar Threads Thread Thread Starter Forum Replies Last Post Bdot GPU Computing 1634 2020-09-10 21:40 xx005fs GpuOwl 0 2019-07-26 21:37 1260 Software 17 2015-08-28 01:35 CRGreathouse Computer Science & Computational Number Theory 18 2013-06-08 19:12 Unregistered Information & Answers 4 2006-10-04 22:38

All times are UTC. The time now is 01:31.

Sun Sep 20 01:31:47 UTC 2020 up 9 days, 22:42, 0 users, load averages: 1.60, 1.32, 1.25