how much ram is being used on each GPU for these GW VelaJr tasks? as reported by nvidia-smi.
if it's more than 830MB or so, then that's probably the reason that the tasks don't work on 3GB and lower GPUs, since you can only use ~27% of the available memory for OpenCL on nvidia cards.
Bernd posted that the day after I had my last error and took the 3GB card out of Einstein processing. While the problem is probably fixed, Nvidia GPU BOINC processing isn't entirely clear cut, as Ian and Zalster pointed out. Right now I'm a little gun shy after 66 errors and a day in the penalty box. Next week I'll be brave and undo the exclusion when I'll have time to keep an eye on it.
Just a quick update to let everyone know that after I excluded the 1060 3GB from Einstein there have been no more errors. Every VelaJr1 task on the 1060 3GB failed, every one on the 1050Ti 4GB or the 1660 6GB succeeded.
Have you tried running these task on the gpu?
Gamma-ray pulsar binary search #1 on GPUs v1.22 () windows_x86_64
Admittedly they are not the tasks you ran before but they seem to run better on my gpu's.
VelaJr running in Nvidia 1060 6gb is showing (nvidia-smi) 1787 MiB memory in use. One at a time, of course. Several are in the validated tasks tables and none in the invalid lists so they appear to be flowing through normally.
(Can't find anything about video RAM used in stderr output. Max card memory - yes; amount used - no.)
Gamma-ray pulsar binary search #1 on GPUs v1.22 () windows_x86_64
Admittedly they are not the tasks you ran before but they seem to run better on my gpu's.
Yes, I am currently running both. It was that specific type if task (VelaJr1) on the 3GB card that failed, but now that I have excluded the 3GB card from running GW O2 tasks I have no new errors.
VelaJr running in Nvidia 1060 6gb is showing (nvidia-smi) 1787 MiB memory in use. One at a time, of course. Several are in the validated tasks tables and none in the invalid lists so they appear to be flowing through normally.
(Can't find anything about video RAM used in stderr output. Max card memory - yes; amount used - no.)
at the bottom of the Nvidia-smi output you can see how much the app is using rather than the whole card. Some of the used memory will be taken up by the desktop manager. not usually more than a couple hundred MB though
I am also getting the CL_MEM_OBJECT_ALLOCATION_FAILURE errors, running the VelaJr tasks. NVidia GeForce GTX 1050 with 2GB RAM. This is the host: https://einsteinathome.org/host/12599623
The errors started at the same time as the last Windows Update (April 15), so I thought that might have something to do with it. I guess not, though?
I have just updated the driver for my GPU, in case that caused it, but that didn't fix it. I am now going to deselect this project for a while so that this computer can get back to work.
I am going to change the
)
I am going to change the entry to only exclude einstein_O2MDF instead of all Einstein work from the 1060 3GB. I'll let you know if I mess anything up.
I don't think that is
)
I don't think that is necessary if I understand Bernd's post.
https://einsteinathome.org/goto/comment/176698Betreger wrote:I don't think
)
Did you read my reply to his statement?
https://einsteinathome.org/goto/comment/176723how much ram is being used on
)
how much ram is being used on each GPU for these GW VelaJr tasks? as reported by nvidia-smi.
if it's more than 830MB or so, then that's probably the reason that the tasks don't work on 3GB and lower GPUs, since you can only use ~27% of the available memory for OpenCL on nvidia cards.
_________________________________________________________________________
Betreger wrote:I don't think
)
Bernd posted that the day after I had my last error and took the 3GB card out of Einstein processing. While the problem is probably fixed, Nvidia GPU BOINC processing isn't entirely clear cut, as Ian and Zalster pointed out. Right now I'm a little gun shy after 66 errors and a day in the penalty box. Next week I'll be brave and undo the exclusion when I'll have time to keep an eye on it.
halfempty wrote:Just a quick
)
Have you tried running these task on the gpu?
Gamma-ray pulsar binary search #1 on GPUs v1.22 () windows_x86_64
Admittedly they are not the tasks you ran before but they seem to run better on my gpu's.
VelaJr running in Nvidia 1060
)
VelaJr running in Nvidia 1060 6gb is showing (nvidia-smi) 1787 MiB memory in use. One at a time, of course. Several are in the validated tasks tables and none in the invalid lists so they appear to be flowing through normally.
(Can't find anything about video RAM used in stderr output. Max card memory - yes; amount used - no.)
mikey wrote:Have you tried
)
Yes, I am currently running both. It was that specific type if task (VelaJr1) on the 3GB card that failed, but now that I have excluded the 3GB card from running GW O2 tasks I have no new errors.
Eugene Stemple wrote:VelaJr
)
at the bottom of the Nvidia-smi output you can see how much the app is using rather than the whole card. Some of the used memory will be taken up by the desktop manager. not usually more than a couple hundred MB though
_________________________________________________________________________
I am also getting the
)
I am also getting the CL_MEM_OBJECT_ALLOCATION_FAILURE errors, running the VelaJr tasks. NVidia GeForce GTX 1050 with 2GB RAM. This is the host: https://einsteinathome.org/host/12599623
The errors started at the same time as the last Windows Update (April 15), so I thought that might have something to do with it. I guess not, though?
I have just updated the driver for my GPU, in case that caused it, but that didn't fix it. I am now going to deselect this project for a while so that this computer can get back to work.