Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Generalized Fermat Prime Search :
All of my GFN19 and above are being marked invalid, could use another set of eyes here.
Author |
Message |
|
Can AMD folks out there do a quick look at some of these failures.
GPU
Running on platform 'Apple', device 'AMD Radeon Pro Vega 56 Compute Engine', vendor 'AMD', version 'OpenCL 1.2 ' and driver '1.2 (Jan 13 2020 20:12:46)'.
56 computeUnits @ 1250MHz, memSize=8176MB, cacheSize=0kB, cacheLineSize=0B, localMemSize=64kB, maxWorkGroupSize=256.
http://www.primegrid.com/result.php?resultid=1070729960
I can see the process is failing to renice the process priority, not sure how that would be responsible for producing invalid results?
Example stderr
<core_client_version>7.14.2</core_client_version>
<![CDATA[
<stderr_txt>
geneferocl 3.3.3-2 (Apple-x86/OpenCL/64-bit)
Copyright 2001-2018, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
Copyright 2010-2012, Shoichiro Yamada, Ken Brazier
Copyright 2011-2014, Michael Goetz, Ronald Schneider
Copyright 2011-2018, Iain Bethune
Genefer is free source code, under the MIT license.
Command line: geneferocl_macintel_3.3.3-2 -boinc -q 3452136^524288+1 --device 0
Normal priority change failed (needs superuser privileges.
Checking available transform implementations...
OCL transform is past its b limit.
OCL4 transform is past its b limit.
A benchmark is needed to determine best transform, testing available transform implementations...
Testing OCL2 transform...
Running on platform 'Apple', device 'AMD Radeon Pro Vega 56 Compute Engine', vendor 'AMD', version 'OpenCL 1.2 ' and driver '1.2 (Jan 13 2020 20:12:46)'.
56 computeUnits @ 1250MHz, memSize=8176MB, cacheSize=0kB, cacheLineSize=0B, localMemSize=64kB, maxWorkGroupSize=256.
Testing OCL3 transform...
Testing OCL5 transform...
Benchmarks completed (2.287 seconds).
Using OCL5 transform
Starting initialization...
Initialization complete (1.856 seconds).
Testing 3452136^524288+1...
Estimated time for 3452136^524288+1 is 1:41:00
3452136^524288+1 is complete. (3427842 digits) (err = 0.0000) (time = 1:44:55) 14:21:30
14:21:30 (4603): called boinc_finish
</stderr_txt>
]]>
| |
|
|
I don't have the answer but I have Vega 56 GPU's. I'm happy to run a few GFN 21 tasks after TdP if it helps solve your issue.
Lets see what others come up with first.
____________
**Please boycott BOINC project Rakesearch as it's a Russian project! ** | |
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13780 ID: 53948 Credit: 343,968,988 RAC: 10,814
                              
|
Everything below GFN-19 runs OCL 2, and that works.
GFN-19 runs OCL5 and that fails.
DYFL runs OCL5 and that fails.
GFN-20 runs OCL4, and it fails too.
GFN-21 and GFN-22 run OCL, but there's no tasks in those projects on your computer.
I suspect one of two things are causing this: either a bug in Apple's video driver (we've seen similar behavior in the past), or some sort of other issue preventing larger tasks from running.
I'll ask the devs to take a look, but I'm not sure there's anything they can do.
____________
My lucky number is 75898524288+1 | |
|
Yves GallotVolunteer developer Project scientist Send message
Joined: 19 Aug 12 Posts: 702 ID: 164101 Credit: 305,166,630 RAC: 523

|
The cause of the problem is AMD driver on Apple computers.
The same code runs with AMD GPUs on Windows and Linux. It also works on macOS with Nvidia GPUs.
OCL3, OCL4 and OCL5 fail but OCL and OCL2 do not.
Mark Rodenkirch and I are working on this problem but until now we don't understand: a large part of the code of OCL4 and OCL2 is identical except that OCL2 is more complex than OCL4 (3 primes vs 2 primes). Why the complex code is OK and the simple one is NOK?
In the meantime, one solution is to force OCL2 transform with the command line argument "-x OCL2". | |
|
rogueVolunteer developer
 Send message
Joined: 8 Sep 07 Posts: 1242 ID: 12001 Credit: 18,565,548 RAC: 0
 
|
The cause of the problem is AMD driver on Apple computers.
The same code runs with AMD GPUs on Windows and Linux. It also works on macOS with Nvidia GPUs.
OCL3, OCL4 and OCL5 fail but OCL and OCL2 do not.
Mark Rodenkirch and I are working on this problem but until now we don't understand: a large part of the code of OCL4 and OCL2 is identical except that OCL2 is more complex than OCL4 (3 primes vs 2 primes). Why the complex code is OK and the simple one is NOK?
In the meantime, one solution is to force OCL2 transform with the command line argument "-x OCL2".
I'm going to away from my iMac for about three weeks starting on Sunday due to work travel, so I hope to spend some time on this before I leave. | |
|
|
Starting a GFN-21 for a sanity test. Looks like it correctly selected OCL here. One thing to keep in mind is that it looks like there was a driver update with the Jan 2020 build Catalina (10.15.3 (19D76))
geneferocl 3.3.3-2 (Apple-x86/OpenCL/64-bit)
Copyright 2001-2018, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
Copyright 2010-2012, Shoichiro Yamada, Ken Brazier
Copyright 2011-2014, Michael Goetz, Ronald Schneider
Copyright 2011-2018, Iain Bethune
Genefer is free source code, under the MIT license.
Command line: geneferocl_macintel_3.3.3-2 -boinc -q 555658^2097152+1 --device 0
Normal priority change failed (needs superuser privileges.
Checking available transform implementations...
A benchmark is needed to determine best transform, testing available transform implementations...
Testing OCL transform...
Running on platform 'Apple', device 'AMD Radeon Pro Vega 56 Compute Engine', vendor 'AMD', version 'OpenCL 1.2 ' and driver '1.2 (Jan 13 2020 20:12:46)'.
56 computeUnits @ 1250MHz, memSize=8176MB, cacheSize=0kB, cacheLineSize=0B, localMemSize=64kB, maxWorkGroupSize=256.
Testing OCL2 transform...
Testing OCL3 transform...
Testing OCL4 transform...
Testing OCL5 transform...
Benchmarks completed (13.685 seconds).
Using OCL transform
Starting initialization...
Initialization complete (23.125 seconds).
Testing 555658^2097152+1...
Estimated time for 555658^2097152+1 is 15:40:00 | |
|
rogueVolunteer developer
 Send message
Joined: 8 Sep 07 Posts: 1242 ID: 12001 Credit: 18,565,548 RAC: 0
 
|
I suggest that if you are running on a Mac with AMD that you run geneferocl options 3 and 4 (primality and residue tests) for each ocl variant. If you have issues with any variants, then it tells me that you have a potentially buggy driver. If anyone can run both on their Mac with AMD and have zero issues with any variant, please let me know.
Yves and I have narrowed down the cause and can get ocl4 transforms working, but more code changes will be necessary.
I'm also working with Apple because this really appears to be an issue with Apple's AMD OpenCL driver. | |
|
|
Replying to this as it seems to be potentially linked to an issue I'm having. I'm new to crunching and to PrimeGrid and one of my devices is my MacBook Pro. Had two "successful, marked invalid" results today on the MacBook.
They can be found here and here.
It only seems to happen on Genefer tasks. Just waiting for the first one to finish on my iMac to see if it has the same issue with them. | |
|
Message boards :
Generalized Fermat Prime Search :
All of my GFN19 and above are being marked invalid, could use another set of eyes here. |