System has been crunching for an hour and has gotten to a few of the new FGRP Units. The MAC is successfully crunching 2 Units at a time per GPU card.
Thanks, everyone, for their input. Thank you Keith Myers for the assistance with creating a new app_config.xml file.
I will continue monitoring the MAC for abnormalities and errors. So far, it seems to be crunching 2 FGRP Units at a time in the Estimated Time Window of 39 Min. per Unit. A couple Units actually finished in less than 19 Min.
TL
[EDIT:]
Just also noticed that in addition to the FGRP 1.12 Units; I've just picked up 1.13 Units as well.
I will continue monitoring the MAC for abnormalities and errors. So far, it seems to be crunching 2 FGRP Units at a time in the Estimated Time Window of 39 Min. per Unit. A couple Units actually finished in less than 19 Min.
<SNIP>
Those times don't look right at all. Granted I'm using a 280x, but runtimes should be in the hundreds of seconds, not thousands (mine go through in around 230-240s). Maybe try running one WU per GPU - if two WU do't fit in your GPU memory then running multiple WU/GPU is *very* counterproductive.
I will continue monitoring the MAC for abnormalities and errors. So far, it seems to be crunching 2 FGRP Units at a time in the Estimated Time Window of 39 Min. per Unit. A couple Units actually finished in less than 19 Min.
<SNIP>
Those times don't look right at all. Granted I'm using a 280x, but runtimes should be in the hundreds of seconds, not thousands (mine go through in around 230-240s). Maybe try running one WU per GPU - if two WU do't fit in your GPU memory then running multiple WU/GPU is *very* counterproductive.
Kailee.
Well, another concern that I have is that these new FGRP Units are OpenCL; and as posted in Number Crunching at SETI by TBar, "Darwin 15.4.0" onward have "issues" Validating properly with OpenCL. At SETI, these OpenCL Units on MAC "usually" end up in the "Inconclusive" Column; until a Third cruncher completes said Units and Validates against my machine and the first Wingman.
However; as I've noticed with the change to Drupal for the Einstein Website, the "Inconclusive" Column is no longer available for view. So right now, I have no clue what's happening with my completed FGRP Units... I'm assuming that they are currently "Inconclusive"; but, am NOT 100% sure about this.
I will continue monitoring, and if these Units turn out to NOT validate, then I'll drop to crunching 1 Unit at a time per GPU card on the MAC.
My FGRP 1.12 Units are starting to show up in Pending. I should be on to the 1.13 Units in the next couple of days. One of the completed 1.12 Units is "Inconclusive"; but, was working against a Windows system... I haven't heard of any new news about FGRP on Windows; so, I must conclude that the Windows system is in Error, and my Unit should validate against the new host.
I will keep monitoring. So far, I see no issues running multiple WUs on my MAC; except as noted by others that my system takes longer to complete the Units... (No big loss, as looking at the Pendings I have; the Wingmen haven't started work on these Units yet.)
One of the completed 1.12 Units is "Inconclusive"; but, was working against a Windows system... I haven't heard of any new news about FGRP on Windows; so, I must conclude that the Windows system is in Error, and my Unit should validate against the new host.
That WIndows system crunched a CPU version (v1.05, FGRPSSE) of that same work. There's plenty of that CPU-version running under this FGRPB1G, not just GPU tasks. That Windows-host in question has a history of zero invalids or errors with those tasks. Interesting to see how it goes.
One of the completed 1.12 Units is "Inconclusive"; but, was working against a Windows system... I haven't heard of any new news about FGRP on Windows; so, I must conclude that the Windows system is in Error, and my Unit should validate against the new host.
That Windows system crunched a CPU version (v1.05, FGRPSSE) of that same work. There's plenty of that CPU-version running under this FGRPB1G, not just GPU tasks. That Windows-host in question has a history of zero invalids or errors with those tasks. Interesting to see how it goes.
I didn't realize that the Windows Unit was CPU; makes sense now that I look at the FGRPSSE... I learn something new every day.
The new/third host, (a Linux system), validated against the Windows system. So, my Unit was marked Invalid. This is a known issue with OpenCL at SETI with "Darwin 15.4.0" and above. If this continues, I now expect to receive more, (but a low percentage), of Invalids and a higher number of Inconclusives.
Again, I'll keep monitoring. For SETI, the answer was installing and using TBar's CUDA75 App. However; I don't foresee something like that happening here.
Also, noticed that my one remaining Beta BRP4G Unit got crunched with a couple other FGRP Units at the same time. The Estimated Time to Complete for the BRP4G went WAY UP to 5+ Hours... I then Suspended all FGRP work and let the BRP4G Unit complete on its own. It quickly shot up from 27% Complete to 100% Complete in 20+ Minutes... I then Resumed the FGRP work and those Units continued completing in 38 Min or less.
Hmm, I just got a bunch of FGRPopencl-ati-mav 1.13 tasks, they ran in only 235 to 238 seconds, but they were validated.
Funny how they're validating from an El Capitan machine with an R9 280x (flashed) running 7.6.33, but on another Mac with a R9 380 (unflashed but supported) running Sierra I got computing errors..
I've found with my Hackintosh that OS X fully leverages my NVIDIA card and will run one work unit at a time to the full potential of the card. I don't know if it's advantageous to run more than one or not, but it certainly is when I run it on Linux. The implementation seems to work completely differently under Linux, where I need to run about 5 or 6 work units, which individually run slower, but collectively keep pace with the Mac running one at a time. Maybe somebody here with more experience will point out where I am an idiot and I will learn something about why it works that way.
The implementation seems to work completely differently under Linux, where I need to run about 5 or 6 work units
It's all about Linux Power Management and it differs from distros to distros and kernels. Some distro does very agressive power management to maximize uptime when running on batteries.
Try using the "performance" mode instead of the "balanced/auto" mode. This will push GPU frequency even if the task is run at low priority (what boinc does)
Crap. I was wondering why my Linux machine with an NVIDIA GTX 960 wasn't getting any GPU tasks.
NVidia device (or driver) doesn't support OpenCL
I have the latest driver and CUDA 8 installed. Any quick references anyone can point me to before I run out of GPU tasks, please? When I try to install OpenCL with the Ubuntu package manager, it tries to remvove CUDA.
[Update:]System has been
)
[Update:]
System has been crunching for an hour and has gotten to a few of the new FGRP Units. The MAC is successfully crunching 2 Units at a time per GPU card.
Thanks, everyone, for their input. Thank you Keith Myers for the assistance with creating a new app_config.xml file.
I will continue monitoring the MAC for abnormalities and errors. So far, it seems to be crunching 2 FGRP Units at a time in the Estimated Time Window of 39 Min. per Unit. A couple Units actually finished in less than 19 Min.
TL
[EDIT:]
Just also noticed that in addition to the FGRP 1.12 Units; I've just picked up 1.13 Units as well.
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
TimeLord04
)
Those times don't look right at all. Granted I'm using a 280x, but runtimes should be in the hundreds of seconds, not thousands (mine go through in around 230-240s). Maybe try running one WU per GPU - if two WU do't fit in your GPU memory then running multiple WU/GPU is *very* counterproductive.
Kailee.
Kai Leibrandt
)
Well, another concern that I have is that these new FGRP Units are OpenCL; and as posted in Number Crunching at SETI by TBar, "Darwin 15.4.0" onward have "issues" Validating properly with OpenCL. At SETI, these OpenCL Units on MAC "usually" end up in the "Inconclusive" Column; until a Third cruncher completes said Units and Validates against my machine and the first Wingman.
However; as I've noticed with the change to Drupal for the Einstein Website, the "Inconclusive" Column is no longer available for view. So right now, I have no clue what's happening with my completed FGRP Units... I'm assuming that they are currently "Inconclusive"; but, am NOT 100% sure about this.
I will continue monitoring, and if these Units turn out to NOT validate, then I'll drop to crunching 1 Unit at a time per GPU card on the MAC.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
[Update:My FGRP 1.12 Units
)
[Update:]
My FGRP 1.12 Units are starting to show up in Pending. I should be on to the 1.13 Units in the next couple of days. One of the completed 1.12 Units is "Inconclusive"; but, was working against a Windows system... I haven't heard of any new news about FGRP on Windows; so, I must conclude that the Windows system is in Error, and my Unit should validate against the new host.
I will keep monitoring. So far, I see no issues running multiple WUs on my MAC; except as noted by others that my system takes longer to complete the Units... (No big loss, as looking at the Pendings I have; the Wingmen haven't started work on these Units yet.)
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
TimeLord04
)
That WIndows system crunched a CPU version (v1.05, FGRPSSE) of that same work. There's plenty of that CPU-version running under this FGRPB1G, not just GPU tasks. That Windows-host in question has a history of zero invalids or errors with those tasks. Interesting to see how it goes.
Richie_9 wrote:TimeLord04
)
I didn't realize that the Windows Unit was CPU; makes sense now that I look at the FGRPSSE... I learn something new every day.
The new/third host, (a Linux system), validated against the Windows system. So, my Unit was marked Invalid. This is a known issue with OpenCL at SETI with "Darwin 15.4.0" and above. If this continues, I now expect to receive more, (but a low percentage), of Invalids and a higher number of Inconclusives.
Again, I'll keep monitoring. For SETI, the answer was installing and using TBar's CUDA75 App. However; I don't foresee something like that happening here.
Also, noticed that my one remaining Beta BRP4G Unit got crunched with a couple other FGRP Units at the same time. The Estimated Time to Complete for the BRP4G went WAY UP to 5+ Hours... I then Suspended all FGRP work and let the BRP4G Unit complete on its own. It quickly shot up from 27% Complete to 100% Complete in 20+ Minutes... I then Resumed the FGRP work and those Units continued completing in 38 Min or less.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
Hmm, I just got a bunch
)
Hmm, I just got a bunch of FGRPopencl-ati-mav 1.13 tasks, they ran in only 235 to 238 seconds, but they were validated.
Funny how they're validating from an El Capitan machine with an R9 280x (flashed) running 7.6.33, but on another Mac with a R9 380 (unflashed but supported) running Sierra I got computing errors..
I've found with my Hackintosh
)
I've found with my Hackintosh that OS X fully leverages my NVIDIA card and will run one work unit at a time to the full potential of the card. I don't know if it's advantageous to run more than one or not, but it certainly is when I run it on Linux. The implementation seems to work completely differently under Linux, where I need to run about 5 or 6 work units, which individually run slower, but collectively keep pace with the Mac running one at a time. Maybe somebody here with more experience will point out where I am an idiot and I will learn something about why it works that way.
The implementation seems to
)
It's all about Linux Power Management and it differs from distros to distros and kernels. Some distro does very agressive power management to maximize uptime when running on batteries.
Try using the "performance" mode instead of the "balanced/auto" mode. This will push GPU frequency even if the task is run at low priority (what boinc does)
Crap. I was wondering why my
)
Crap. I was wondering why my Linux machine with an NVIDIA GTX 960 wasn't getting any GPU tasks.