Go into account settings - computing preferences - and
set it to run "No more than 98%" of CPUs
That should Free Up one of your CPU cores and, your GPU
will run much better - (Probably : )
If you were going to do this, it would be better to do it locally through BOINC Manager and not on the website. Otherwise, his other hosts that don't have GPUs would all lose a crunching core - unless he went to the trouble of putting them in a different location (venue) where the change didn't apply.
In any case, he needs to sort out what's crippling the machine and then he may not need this change at all.
The utilization factor of 0.5 would be a real bonus for volunteers with AMD GPUs, even those with a 'little less than mid-range' type of unit, but particularly for the higher end.
AMD's now long standing problem with all their current batch of GCN 1.2 cards require this set to 1.0, until it get fixed (has this been fixed?)
It is quite amazing that boinc can actually take a vast variety of OSs, CPUs, GPUs and their drivers, memory, disk space, projects and their vast variety of apps then blending this unique collection of ingredients into a node of supercomputer (or supercomputers depending on how you see it) for no software cost to the participants - it is quite a feat.
Arecibo GPU tasks go on like clockwork. It is the Parkes GPU tasks that fail and the CPU tasks taking long runs, while they run OK on the Linux boxes. I've limited CPU usage to 55% on the Challenge task of vLHC@home running on the Windows PC, while CMS-dev, also a VirtualBox task on the same PC, runs perfectly.
Tullio
AMD's now long standing problem with all their current batch of GCN 1.2 cards require this set to 1.0, until it get fixed (has this been fixed?)
@AgentB
short answer ... NO
Some over at Seti were claiming for awhile that they could run
2X on Hawaii etc. with new AMD Vulkan Drivers. Now, they have recanted
that and said they are getting errors running more than one : (
AMD's now long standing problem with all their current batch of GCN 1.2 cards require this set to 1.0, until it get fixed (has this been fixed?)
Yes, quite right, I should have qualified the 'high end' comment by excluding the particular cards suffering from this problem. I was thinking about the time, hopefully soon, when AMD fix that. Then, as GPU apps become universal for all science runs here (which is the trend), and more people start using GPUs, new volunteers with AMD GPUs might have an easier entry.
Unfortunately, they don't. When I posted the link there were 6 failed Arecibo GPU tasks. Now there are 7. Unfortunately, that machine is not reacting well to the workload you are giving it and all types of Einstein work seems to be having difficulty and not just Parkes BRP6.
I see 23 valid Arecibo tasks out of 28. All Parkes tasks failed save one. I have reduced the load of VBox tasks. All CPU gamma-ray tasks have completed even in a longer time than on the Linux boxes, one of them 32-bit. Cheers.
Tullio
PS I never had a higher RAC on any BOINC project than this.
RE: Go into account
)
If you were going to do this, it would be better to do it locally through BOINC Manager and not on the website. Otherwise, his other hosts that don't have GPUs would all lose a crunching core - unless he went to the trouble of putting them in a different location (venue) where the change didn't apply.
In any case, he needs to sort out what's crippling the machine and then he may not need this change at all.
Cheers,
Gary.
RE: The utilization factor
)
AMD's now long standing problem with all their current batch of GCN 1.2 cards require this set to 1.0, until it get fixed (has this been fixed?)
It is quite amazing that boinc can actually take a vast variety of OSs, CPUs, GPUs and their drivers, memory, disk space, projects and their vast variety of apps then blending this unique collection of ingredients into a node of supercomputer (or supercomputers depending on how you see it) for no software cost to the participants - it is quite a feat.
Arecibo GPU tasks go on like
)
Arecibo GPU tasks go on like clockwork. It is the Parkes GPU tasks that fail and the CPU tasks taking long runs, while they run OK on the Linux boxes. I've limited CPU usage to 55% on the Challenge task of vLHC@home running on the Windows PC, while CMS-dev, also a VirtualBox task on the same PC, runs perfectly.
Tullio
RE: AMD's now long standing
)
@AgentB
short answer ... NO
Some over at Seti were claiming for awhile that they could run
2X on Hawaii etc. with new AMD Vulkan Drivers. Now, they have recanted
that and said they are getting errors running more than one : (
Bill
RE: Some over at Seti were
)
Thanks Bill, guess you mean Crimson, although anything is possible at Seti!
RE: RE: Some over at
)
It's a new name for what they used to call OpenGL .... I think )
http://support.amd.com/en-us/kb-articles/Pages/Radeon-Vulkan-Beta.aspx
RE: It's a new name for
)
Thanks Bill, i missed seeing that beta release, only a few weeks ago. Interesting times ahead.
RE: AMD's now long standing
)
Yes, quite right, I should have qualified the 'high end' comment by excluding the particular cards suffering from this problem. I was thinking about the time, hopefully soon, when AMD fix that. Then, as GPU apps become universal for all science runs here (which is the trend), and more people start using GPUs, new volunteers with AMD GPUs might have an easier entry.
Cheers,
Gary.
RE: Arecibo GPU tasks go on
)
Unfortunately, they don't. When I posted the link there were 6 failed Arecibo GPU tasks. Now there are 7. Unfortunately, that machine is not reacting well to the workload you are giving it and all types of Einstein work seems to be having difficulty and not just Parkes BRP6.
Did you actually check the link for yourself?
Cheers,
Gary.
I see 23 valid Arecibo tasks
)
I see 23 valid Arecibo tasks out of 28. All Parkes tasks failed save one. I have reduced the load of VBox tasks. All CPU gamma-ray tasks have completed even in a longer time than on the Linux boxes, one of them 32-bit. Cheers.
Tullio
PS I never had a higher RAC on any BOINC project than this.