Found the way to run IntelGPU task at full speed without free cpu core. Just set priority to "Realtime" in Process Explorer, and Intel gpu usage became 96% instead of usual 15-20%. Run time decreased to 12 minutes.
Up until Feb 9, your long running Intel GPU tasks were (mostly) validating. Since that time, virtually all your short running (10-12 minute) tasks have been ending up as validate errors. Sure, there's lots of 'pendings' but as soon as the validator actually looks at them (when the companion task is returned), they seem to fail the sanity check.
Found the way to run IntelGPU task at full speed without free cpu core. Just set priority to "Realtime" in Process Explorer, and Intel gpu usage became 96% instead of usual 15-20%. Run time decreased to 12 minutes.
Up until Feb 9, your long running Intel GPU tasks were (mostly) validating. Since that time, virtually all your short running (10-12 minute) tasks have been ending up as validate errors. Sure, there's lots of 'pendings' but as soon as the validator actually looks at them (when the companion task is returned), they seem to fail the sanity check.
Doesn't really seem to be a viable solution.
Don't forget that Bernd made changes to the validator around 5 February to eliminate "scientifically unusable results" arising from driver changes.
Sure, but there were no validate errors showing up in that host's BRP4 (Arecibo) results between Feb 5 and Feb 9 when the 'realtime solution' was proposed. Between Feb 9 and present there are 41 validate errors, the first ones appearing just a couple of hours after the post was made. If these errors were directly linked to Bernd's changes, why didn't they start on Feb 6?
Did you try setting priority to 'realtime' as suggested? I've never tried to use the Intel GPU in any of my hosts - those applicable all have external GPUs and I haven't been game to risk upsetting things until there is evidence that the problems are sorted and there is a clear benefit to be had.
Did you try setting priority to 'realtime' as suggested? I've never tried to use the Intel GPU in any of my hosts - those applicable all have external GPUs and I haven't been game to risk upsetting things until there is evidence that the problems are sorted and there is a clear benefit to be had.
Like Alex, I had a go, but Task Manager wouldn't accept realtime priority. I thought at first it was saying "can't set priority" because of a user/group permission thing and got diverted into exploring that. I'm using the same OS as den777 (Win 7 Pro 64) so we might ask him what tool he used.
If I don't get sidetracked today, I'll download a copy of Process Lassoo and see how I get on with that.
OK, I'm posting this from the Haswell (host 8864187), which has been running at 100% CPU usage and 'real time' priority for the intel_gpu app under Process Lasso for the last couple of hours or so - since shortly before 13:00 UTC today.
No sign of any screen lag, either when doing some normal full-screen browsing (at 1,920 x 1.200 px) on a monitor attached to the HD 4600 graphics, or when playing simple-minded Windows games like Solitaire. Run time for the OpenCL app seem to be the same as for the normal 75% CPU, low priority, mode - about 11 minutes - and I haven't seen any validation problems yet. I'm using the 3621 driver.
Seems to be safe to leave it running like this a while longer.
I played with RT priority only when I wrote this in forum. Because I changed priority for each task manually. All other short-time tasks since that was crunched with free core.
Also I've updated intel video drivers after that, probably that's the reason...
Quote:
I tried it, but my win7 does not allow that. 'High' is the max I can set.
Use Process Explorer or task manager under admin account (right click, run as administrator)
Confirmed. Driver 3621 and ProcessLasso for realtime priority produces valid results. http://einsteinathome.org/host/8805680/tasks&offset=0&show_names=1&state=0&appid=19
No errors since today.
The only side effect is video&audio lags for 1 second every 12 minutes when new tasks start, because ProcessLasso assigns RT priority for them instantly.
Agreed - I was just going to mention the momentary stutter every time one task finishes and the next one starts. And the Process Lasso setting to make the program re-start with the current configuration at machine re-start works a treat.
I played with RT priority only when I wrote this in forum. Because I changed priority for each task manually. All other short-time tasks since that was crunched with free core.
Also I've updated intel video drivers after that, probably that's the reason...
Quote:
I tried it, but my win7 does not allow that. 'High' is the max I can set.
Use Process Explorer or task manager under admin account (right click, run as administrator)
Thank you for the infos.
As Richard posted, I tried it with PL, this one worked. Also gives the possibility to set all tasks of this kind to 'Realtime'.
My 'Rocky1' is running now for ~ 24hrs in this mode, no negative effects found.
Process Lasso has calculated an average cpu-load of 0.39% for the intel gpu app.
Your solution is much more effective than leaving a core free!
RE: Found the way to run
)
Up until Feb 9, your long running Intel GPU tasks were (mostly) validating. Since that time, virtually all your short running (10-12 minute) tasks have been ending up as validate errors. Sure, there's lots of 'pendings' but as soon as the validator actually looks at them (when the companion task is returned), they seem to fail the sanity check.
Doesn't really seem to be a viable solution.
Cheers,
Gary.
RE: RE: Found the way to
)
Don't forget that Bernd made changes to the validator around 5 February to eliminate "scientifically unusable results" arising from driver changes.
http://einsteinathome.org/node/197052&nowrap=true#138129
RE: Don't forget that Bernd
)
Sure, but there were no validate errors showing up in that host's BRP4 (Arecibo) results between Feb 5 and Feb 9 when the 'realtime solution' was proposed. Between Feb 9 and present there are 41 validate errors, the first ones appearing just a couple of hours after the post was made. If these errors were directly linked to Bernd's changes, why didn't they start on Feb 6?
Did you try setting priority to 'realtime' as suggested? I've never tried to use the Intel GPU in any of my hosts - those applicable all have external GPUs and I haven't been game to risk upsetting things until there is evidence that the problems are sorted and there is a clear benefit to be had.
Cheers,
Gary.
RE: Did you try setting
)
Like Alex, I had a go, but Task Manager wouldn't accept realtime priority. I thought at first it was saying "can't set priority" because of a user/group permission thing and got diverted into exploring that. I'm using the same OS as den777 (Win 7 Pro 64) so we might ask him what tool he used.
If I don't get sidetracked today, I'll download a copy of Process Lassoo and see how I get on with that.
OK, I'm posting this from the
)
OK, I'm posting this from the Haswell (host 8864187), which has been running at 100% CPU usage and 'real time' priority for the intel_gpu app under Process Lasso for the last couple of hours or so - since shortly before 13:00 UTC today.
No sign of any screen lag, either when doing some normal full-screen browsing (at 1,920 x 1.200 px) on a monitor attached to the HD 4600 graphics, or when playing simple-minded Windows games like Solitaire. Run time for the OpenCL app seem to be the same as for the normal 75% CPU, low priority, mode - about 11 minutes - and I haven't seen any validation problems yet. I'm using the 3621 driver.
Seems to be safe to leave it running like this a while longer.
I played with RT priority
)
I played with RT priority only when I wrote this in forum. Because I changed priority for each task manually. All other short-time tasks since that was crunched with free core.
Also I've updated intel video drivers after that, probably that's the reason...
Use Process Explorer or task manager under admin account (right click, run as administrator)
RE: Also I've updated intel
)
You're currently running driver 10.18.14.4080, which is known to produce bad results on HD 4600 GPUs at Einstein.
To run with the current Einstein application, you need to revert to the 10.18.10.3621 driver, as discussed in the BRP4 Intel GPU app feedback thread.
Confirmed. Driver 3621 and
)
Confirmed. Driver 3621 and ProcessLasso for realtime priority produces valid results.
http://einsteinathome.org/host/8805680/tasks&offset=0&show_names=1&state=0&appid=19
No errors since today.
The only side effect is video&audio lags for 1 second every 12 minutes when new tasks start, because ProcessLasso assigns RT priority for them instantly.
Agreed - I was just going to
)
Agreed - I was just going to mention the momentary stutter every time one task finishes and the next one starts. And the Process Lasso setting to make the program re-start with the current configuration at machine re-start works a treat.
RE: I played with RT
)
Thank you for the infos.
As Richard posted, I tried it with PL, this one worked. Also gives the possibility to set all tasks of this kind to 'Realtime'.
My 'Rocky1' is running now for ~ 24hrs in this mode, no negative effects found.
Process Lasso has calculated an average cpu-load of 0.39% for the intel gpu app.
Your solution is much more effective than leaving a core free!
Alexander