[QUOTE=SELROC;494787]Yes. I do this, for a double-control I switch between GPUs, the issue is persistent. It is:

300M exponent with 1 error recorded and 200 block size: 16.80 ms/it

332M exponent with 0 errors and 400 block size: 16.50 ms/it

The difference is small, but persistent.

FFT size 18M[/QUOTE]

Ah, it's the block size that's different. Every blockSize iterations, an additional multiplication is done. Maybe this multiply is a bit slower than a normal iteration (let's say, 2x or 3x), that might stretching it a bit explain the difference.

[QUOTE=preda;494792]Ah, it's the block size that's different. Every blockSize iterations, an additional multiplication is done. Maybe this multiply is a bit slower than a normal iteration (let's say, 2x or 3x), that might stretching it a bit explain the difference.[/QUOTE]

Also every blockSize^2 iterations, you do the actual error check, right?

[QUOTE=axn;494794]Also every blockSize^2 iterations, you do the actual error check, right?[/QUOTE]

Yes, but that's accounted separately time-wise, not in the time-per-it.

[QUOTE=preda;494792]Ah, it's the block size that's different. Every blockSize iterations, an additional multiplication is done. Maybe this multiply is a bit slower than a normal iteration (let's say, 2x or 3x), that might stretching it a bit explain the difference.[/QUOTE]


Yes.


I started the 300M exponent with FFT 16M and blockSize 200, at 8% gave the first error, it was stopped and restarted with default FFT 18M.


The 332M exponent started with default FFT 18M and blockSize 400.

Latest Windows build from git(as of now...) untested!

Does gpuowl accept pull requests in github ?

[QUOTE=SELROC;494843]Does gpuowl accept pull requests in github ?[/QUOTE]

It depends: I'm open and welcome contributions, but I'd do a review (which might be considered strict), and I reserve the right to not accept it in the end.

[QUOTE=preda;494845]It depends: I'm open and welcome contributions, but I'd do a review (which might be considered strict), and I reserve the right to not accept it in the end.[/QUOTE]

Sorry for the wrong title, first time for me to open a pull request on github.

Builds posted
 

[QUOTE=kracker;494833]Latest Windows build from git(as of now...) untested![/QUOTE]Welcome back kracker to build posting!
How is your build at post 676 different from what I posted in [URL]http://www.mersenneforum.org/showpost.php?p=494759&postcount=666?[/URL]

FYI, James Heinrich has at my request put up several Windows build versions on his mirror site, at [URL]https://download.mersenne.ca/gpuowl[/URL], a selection from the larger list of links to executables or zip files at the bottom of [URL]http://www.mersenneforum.org/showpost.php?p=488539&postcount=4[/URL]

[QUOTE=SELROC;494848]Sorry for the wrong title, first time for me to open a pull request on github.[/QUOTE]

Thanks for this pull request!

[QUOTE=kriesel;494876]Welcome back kracker to build posting!
How is your build at post 676 different from what I posted in [URL]http://www.mersenneforum.org/showpost.php?p=494759&postcount=666?[/URL]

FYI, James Heinrich has at my request put up several Windows build versions on his mirror site, at [URL]https://download.mersenne.ca/gpuowl[/URL], a selection from the larger list of links to executables or zip files at the bottom of [URL]http://www.mersenneforum.org/showpost.php?p=488539&postcount=4[/URL][/QUOTE]


Hmm... Sorry, didn't notice you posted! Honestly I think it's very similar... only difference is that I don't use -O2 when compiling... don't think that flag affects gpu performance at all(?). Could be(probably) wrong though.