No "Collatz Sieve" for Nvidia Ion
log in

Advanced search

Message boards : Windows : No "Collatz Sieve" for Nvidia Ion

Author Message
Profile JohnMD
Avatar
Send message
Joined: 25 Mar 14
Posts: 29
Credit: 15,481,538
RAC: 0
Message 20919 - Posted: 29 Jul 2015, 10:00:50 UTC

Ion only supports Opencl 100. Collatz application reference does not say which Opencl level is required - can it be added ?

Profile JohnMD
Avatar
Send message
Joined: 25 Mar 14
Posts: 29
Credit: 15,481,538
RAC: 0
Message 20926 - Posted: 29 Jul 2015, 20:27:50 UTC - in response to Message 20919.

Solved !
By chance I saw that Collatz Sieve is a test application - and of course I had not checked the test-box :.-(

Profile JohnMD
Avatar
Send message
Joined: 25 Mar 14
Posts: 29
Credit: 15,481,538
RAC: 0
Message 20975 - Posted: 11 Aug 2015, 19:36:31 UTC

Following stderr -
Using:
verbose=1
threads=64
sleep=1
lookup table size=32
reduction on CPU=no
kernels per reduction=2
Error: (-4)Memory object allocation failure at 1275 of collatzOpenCL::RunSteps
Error: GPU steps do not match CPU steps. Workunit processing aborted.
21:31:31 (6684): called boinc_finish

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 20977 - Posted: 11 Aug 2015, 20:34:22 UTC - in response to Message 20975.

Following stderr -
Using:
verbose=1
threads=64
sleep=1
lookup table size=32
reduction on CPU=no
kernels per reduction=2
Error: (-4)Memory object allocation failure at 1275 of collatzOpenCL::RunSteps
Error: GPU steps do not match CPU steps. Workunit processing aborted.
21:31:31 (6684): called boinc_finish


The issue is that nVidia doesn't always support the number of items being run per kernel. Some work, others don't. Your's is one that doesn't. I'm testing v1.07 of the sieve app and if it checks out, there will be a version in a few days that should run a without crashing. It will also take care of some of the video lag issues many have reported. Using a smaller sieve means less numbers per kernel but more numbers checked overall (e.g. To check 2^32 numbers using a 2^26 sieve size will calculate 64 million numbers. Using a 2^32 sieve needs to check only 40 million numbers. So, using 2^26 will take longer as it has to do more work but since the credit is based on total steps and there will be more of them because of more numbers being checked, the end result should be about the same amount of credit per hour regardless of the sieve size chosen. Using a higher sieve size will check more numbers faster and would be preferred, but if it crashes (runs out of memory), no science gets done. I just have to figure out what the best default size is for the various platforms.

Profile JohnMD
Avatar
Send message
Joined: 25 Mar 14
Posts: 29
Credit: 15,481,538
RAC: 0
Message 20978 - Posted: 11 Aug 2015, 22:34:44 UTC

Thanks Slicker - I'll be looking out for 1.07.
As you can see, I was using the absolute minimum values for parameters. Maybe sieve-size could also be a parameter - with some recommendation on it's use ?


Post to thread

Message boards : Windows : No "Collatz Sieve" for Nvidia Ion


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.