Yesterday (3/27/17) at 4:30 pm MST, I had this task that was ready to go: http://bit.ly/2mMFJT2

When I got in this morning (3/28/17) at 8:00 AM MST, it still hadn't ran.

I have my tasks set up to run for 60 minutes at a time.

I had to suspend all the running GPU tasks simultaneously for it to run.

It finished in 00:28:34.

Is there a priority scheduling problem with multiple GPUs on the same WU?
Seems like it would be the same as a Cosmology or other (4CPU or 6CPU task)

Please let me know any info you have. Or if I should post this on the general Boinc forum.

Thanks, -Doug

P.S. How many Nvidia GPUs will be utilized simultaneously by a task like this?
I can have up to 7 cards on my MOBO.

Would 3 AMD cards work as well?

Could Nvidia and AMD be used on the same task?

As far as I know, MooWrapper specifically runs each work unit on an individual GPU, even when you have multiple cards of the same model in the machine, and there is no viable way around this (technically there IS a potential work-around via editing the "master" copy of the .ini file but it messes up the wrapper assign work code if you try to do so).

If you're using a mixed-card machine, this actually makes a LOT of sense as it prevents one WU tying up all your GPUs while there is only enough left for one to work on, while having little if any downside on a machine with all of the GPUs in it the same.

The base DNet client itself does not assign one key to multiple GPUs to work on at the same time, even when running it as a multi-GPU setup it still assigns a key to a single GPU, so there is no actual benefit to trying to force MooWrapper to assign one WU to all the GPUs in your system.

Yesterday (3/27/17) at 4:30 pm MST, I had this task that was ready to go: http://bit.ly/2mMFJT2

When I got in this morning (3/28/17) at 8:00 AM MST, it still hadn't ran.

I have my tasks set up to run for 60 minutes at a time.

I had to suspend all the running GPU tasks simultaneously for it to run.

It finished in 00:28:34.

Is there a priority scheduling problem with multiple GPUs on the same WU?
Seems like it would be the same as a Cosmology or other (4CPU or 6CPU task)

Please let me know any info you have. Or if I should post this on the general Boinc forum.

Thanks, -Doug

P.S. How many Nvidia GPUs will be utilized simultaneously by a task like this?
I can have up to 7 cards on my MOBO.

Would 3 AMD cards work as well?

Could Nvidia and AMD be used on the same task?

Boinc is not setup to run multiple gpu's on the same workunit at this time, some of the individual projects have figured out how to do that for the cpu's but not all of them even do that. If you do not game in fact it can be faster crunching if you remove any cables connecting the two gpu's together. You can then treat each gpu independently and run one gpu at project a and the other gpu at project b, or both at the same project of course.

My rig already had 3 Nvidia cards, so I thought I'd try my hand at adding an AMD card (HD 7470). I run a server build and Lubuntu dektop. Needless to say I was SOL and just replaced the AMD card with a fourth Nvidia card. I have 3 more cards on the way and I'll have a total of 7 GPUs.

Here is my client output: http://bit.ly/2oRiRBc

They are all happily crunching along. My previous comment about scheduling seems to be debunked except for when a (multi GPU) Moo! Wrapper task is pending, it won't DL any new GPU tasks until the Moo! Wrapper task has ran.



So Mikey, are you saying even though it says it's using 3 or 4 Nvidia GPUs for one task it's actually not?

Here is the first completed 4 GPU Moo Wrapper WU: http://moowrap.net/result.php?resultid=54926647

My rig already had 3 Nvidia cards, so I thought I'd try my hand at adding an AMD card (HD 7470). I run a server build and Lubuntu dektop. Needless to say I was SOL and just replaced the AMD card with a fourth Nvidia card. I have 3 more cards on the way and I'll have a total of 7 GPUs.

Here is my client output: http://bit.ly/2oRiRBc

They are all happily crunching along. My previous comment about scheduling seems to be debunked except for when a (multi GPU) Moo! Wrapper task is pending, it won't DL any new GPU tasks until the Moo! Wrapper task has ran.



So Mikey, are you saying even though it says it's using 3 or 4 Nvidia GPUs for one task it's actually not?

Here is the first completed 4 GPU Moo Wrapper WU: http://moowrap.net/result.php?resultid=54926647

No I'm saying Boinc doesn't provide support for doing that, it's up to each project and most do not, and being project specific it means that Moo does and Einstein or PrimeGrid, as examples, do not. If it were built into Boinc it would be possible on every project that has gpu workunits.

It wouldn't be practical on Moo Wrapper to run one work unit on multiple GPUs even if BOINC specifically supported it, as Moo is nothing more than a "wrapper" for the Distributed.net client that client does not support running one "block" on more than one GPU at a time.

It's possible to have AMD and NVida GPUs mixed on a Linux machine - but an AMD GPU MUST be the one used for video output, you MUST install the AMD drivers first, THEN install the NVidia drivers from the .run package MANUALLY using the "--no-opengl-files" command line switch - otherwise the NVidia opengl stuff overwrites the AMD opengl stuff and completely borks the AMD drivers to the point the AMD GPUs flat out won't work at all.

In theory you can hand-install the applicable files but that's a royal pain at best and will leave you with an unusable system if you miss even ONE of them, or use ONE wrong file.

Mixed GPUs is one of the very few things Windows does better than Linux, and IMO one of only 2 things it does a LOT better (the other being "support games").