[QUOTE=mattmill30;438348]I've been running mfakto (and mfaktc) on a number of workstations, but when GPUSieve=1 the workstation graphics become heavily latented.


What other tasks than sieving does mfakto perform?


Would it be possible to implement something similar to the CUDALucas Polite setting, or otherwise opportunistic GPU sieving, so that when GPU resources are underutilised mfakto switches to GPU sieving.[/QUOTE]

You might want to have a play with this bit if using mfaktc or it's equivalent if any in mfakto - only have nvidia cards so can't help with that

# GPUSieveProcessSize defines how far many bits of the sieve each TF block
# processes (in K bits). Larger values may lead to less wasted cycles by
# reducing the number of times all threads in a warp are not TFing a
# candidate. However, more shared memory is used which may reduce occupancy.
# Smaller values should lead to a more responsive system (each kernel takes
# less time to execute). GPUSieveProcessSize must be a multiple of 8.
#
# Minimum: GPUSieveProcessSize=8
# Maximum: GPUSieveProcessSize=32
#
# Default: GPUSieveProcessSize=16

Thanks kladner and Gordon. [QUOTE=mattmill30;438348]I've been running mfakto (and mfaktc) on a number of workstations, but when GPUSieve=1 the workstation graphics become heavily latented.


What other tasks than sieving does mfakto perform?


Would it be possible to implement something similar to the CUDALucas Polite setting, or otherwise opportunistic GPU sieving, so that when GPU resources are underutilised mfakto switches to GPU sieving.[/QUOTE]

Since upgrading to Windows 10 on my device which has an Intel HD 4000 graphics card (laptop so probably integrated), I have noticed that the graphics latency has significantly reduced. Interestingly with this change, when I checked Task Manager the CPU now registers full utilisation of a single core by mfakto (with GPUSieve=1) whereas on Windows 7 the mfakto was reported as utilising 1-5%.

I wonder whether the issue is that the Intel HD 4000 GPU is integrated into the CPU, along with the OpenCL component, in a way which Windows 7 is not capable of recognising and so the Windows task scheduler is effectively over provisioning the CPU to 125%.

In contrast if Windows 10 can recognise the integrated GPU/CPU OpenCL component, then the scheduler isn't over provisioning the CPU resources (including the GPU), by not assuming the GPU OpenCL component is a separate resource and provisioning as such.

I'm just going to jump in here without reading the previous 124 pages of this thread...
I just got a new [url=https://ca.msi.com/Graphics-card/Radeon-RX-480-GAMING-X-8G.html]RX 480[/url] so I'm (finally!) trying out mfakto. Perhaps it should be noted up front that I still have my GTX 670 in the other slot and am simultaneously running mfaktc there.
Windows7pro64, lastest drivers from AMD (non-whql-win7-64bit-radeon-software-crimson-16.11.4-nov15)

I grabbed a copy of [url=http://download.mersenne.ca/mfakto/mfakto-0.15pre6/]mfakto-0.15pre6[/url] and fired it up with no parameters, and its internal small selftest shows:[code]Started a simple selftest ...
ERROR: selftest failed for M50752613 (cl_barrett15_73_gs)
no factor found
Selftest statistics
number of tests 30
successful tests 29
no factor found 1
selftest FAILED![/code]So I ran the more extensive -st and got:[code]Selftest statistics
number of tests 34026
successful tests 33961
no factor found 65
selftest FAILED!
ERROR: selftest failed, exiting.[/code]Is this a known issue? Why would it mostly-work but a tiny fraction of them fail?

[QUOTE=James Heinrich;447628]I'm just going to jump in here without reading the previous 124 pages of this thread...
I just got a new [url=https://ca.msi.com/Graphics-card/Radeon-RX-480-GAMING-X-8G.html]RX 480[/url] so I'm (finally!) trying out mfakto. Perhaps it should be noted up front that I still have my GTX 670 in the other slot and am simultaneously running mfaktc there.
Windows7pro64, lastest drivers from AMD (non-whql-win7-64bit-radeon-software-crimson-16.11.4-nov15)

I grabbed a copy of [url=http://download.mersenne.ca/mfakto/mfakto-0.15pre6/]mfakto-0.15pre6[/url] and fired it up with no parameters, and its internal small selftest shows:[code]Started a simple selftest ...
ERROR: selftest failed for M50752613 (cl_barrett15_73_gs)
no factor found
Selftest statistics
number of tests 30
successful tests 29
no factor found 1
selftest FAILED![/code]So I ran the more extensive -st and got:[code]Selftest statistics
number of tests 34026
successful tests 33961
no factor found 65
selftest FAILED!
ERROR: selftest failed, exiting.[/code]Is this a known issue? Why would it mostly-work but a tiny fraction of them fail?[/QUOTE]

On linux, this was the Selftest landscape for a long time. Only updating to the AMDGPU driver included in the final 4.8 kernel yielded a stable card.

[QUOTE=airsquirrels;447629]On linux, this was the Selftest landscape for a long time. Only updating to the AMDGPU driver included in the final 4.8 kernel yielded a stable card.[/QUOTE]So mfakto on RX-4xx on Windows is broken until AMD fixes something? :cry:

Playing with it a little more, if I repeatedly start mfakto, it will pass 30/30 selftests about half the time. Always that same one (M50752613) that (sometimes) fails. Does this mean that if I can get it to start, mfakto may [i]probably[/i] (but [i]definitely[/i]) find the factor in the exponent I'm working on?

I just ran mfakto 0.14 on my new IdeaPad FLEX 4 (with a Radeon R7 M460 GPU) and got similar results. Upgrading the Radeon drivers to the latest version didn't resolve the issue.

Sorry for the double post, but I have an update: I downloaded version 0.15pre6 instead, and the self-tests initially passed. However, there are two more issues:

1. mfakto only ran for a short time before it stopped progressing. There were no error messages; the output simply got stuck at 4% for several minutes. After I quit mfakto and launched it again, one of the self-tests failed.
2. I got a Blue Screen of Death with the error code [FONT="Courier New"]THREAD_STUCK_IN_DEVICE_DRIVER[/FONT] while mfakto was running. I don't know if the mfakto was the culprit, but the issue hasn't happened again after I removed the program.

The "4%" and the associated BSoD sound like known to us.. we hit the 4% issue few times in the past with both AMD and Nvidia cards (i.e under mfaktc too) and updating the drivers (or sometimes rolling them back) solved the problem every time. Unfortunately we do not have new and fancy cards therefore we don't know how to help with your particular problem. Bdot may level into it, please...

We learned for 100th time that "if it works do not fix it" and we try to avoid upgrading drivers and software as long as the old one works with the old hardware we have...

I guess I should mention that I've submitted several weeks worth of TF results from RX480 cards already...

I was aware from the [URL="http://www.mersenneforum.org/showthread.php?t=21353&page=5"]RX480 thread[/URL] that there were issues so I did check a batch of exponents in parallel with a R9 270 and ran -st and -st2 several times, including after driver updates.

That all passed running [URL="http://www.mersenneforum.org/showpost.php?p=423772&postcount=1338"]mfakto-0.14p[/URL] (0.14-MGW) on Windows 10 x64.

Just passed the selftests on mfakto-0.14-MGW with driver 16.11.3 (not the latest) but replicating the issues mentioned for mfakto-0.15pre6.

I just tried that build and got the same failures.

[QUOTE=ixfd64;447869]I just tried that build and got the same failures.[/QUOTE]

My experience has been that the issue was a scheduler/barrier timing issue when I tried to debug it. I could get several successful ST in a row and the have one fail. Usually it was an issue with the output of the sieve, not the factoring itself.