UK BOINC Team Forum

I have two questions about multiple GPUs.

1) If I have an on-board GPU, Intel HD 4600 on my i7, and install a new GPU then it is my understanding it will turn the on-board one off?

Can I tell it not to do this, and instead allow the machine to use both; the HD 4600 is okay for Boinc so it makes more sense to allow Boinc to run using both GPUs.

2) When you have 2 GPUs, how does Boinc actually use these; do you still only have one GPU task, but it runs across both GPUs at twice the speed, or do you get two GPU tasks?

Thanks.

Hi Chris,

1) On my machines with an integrated GPU (internal to the i7) and an external GPU in a PCIe slot, when I first installed the external GPUs the machine did indeed default to using it although I don't think it actually turned the internal one off, just ignored it. In all cases, there was an option in the BIOSes to select which one(s) to use for the display. At present, since the integrated one is far less powerful, I use that for the display and the others for crunching. I believe BOINC will still detect all the GPUs (see the answer to 2) regardless of which is actually driving the display.

2) By default, BOINc will only use the most powerful GPU it finds and any other ones will be ignored so you'll only have one task running on one GPU. Add the following to your cc_config.xml file then reboot/restart BOINC/re-read configuration files and it'll use all the GPUs it finds. Some projects only support a limited range of GPUs types so even with everything enabled and detected, you may not get work for a particular GPU from that project. With no other changes you'll get one GPU task running on each GPU. (It is possible to run more than one task at a time per GPU but you can't combine GPUs to share a task)

Regards,

Mark

Thank you so much Mark, that makes perfect sense.

No problem Chris, I've been there before

Just as an update to this, it turned out that Boinc and the i7 actually recognised both cards without me having to actually do anything so I am guessing they have changed some of the coding re this. I did not change anything in the Bios, as nothing seemed appropriate on my Gigabyte MB, and before I had a chance to find the config file I noticed that 2 GPU tasks were running in Boinc client.

This only works for an onboard GPU + GPU card.

If you install two cards you will need to create the cc_config.xml file as explained above.

You need to save the cc_config.xml file to C:\ProgramData\BOINC - Its a hidden folder so just type that into Windows explorer.

Also, if your using Notepad to create the file make sure you save the file as "cc_config.xml" (including the quotes), and not as: cc_config.xml - else Windows will actually save it as: cc_config.xml.txt !!

The BIOS has settings to control whether the IGPU or PCIe GPU is the default GPU. You can also disable the IGPU if you wish. I've always been concerned about burning up the IGPU while crunching on it and trashing the mobo. Maybe that's not concern. Been a long time since I did and don't recall temps. But the credits weren't high enough to risk it for me.

Just seen this message; I agree, no point crunching on the iGPU.

Since this thread has been resurrected:

It is now possible to run one GPU task on multiple GPUs, Moo does this. They still need to be the same family though (Nvidia or AMD) and I'm not sure if the work is equally spread or depends on the capabilities of the GPUs in question as my setups have matched GPUs.

Hopefully they share depending on their capabilities as it wouldn't be beneficial from a crunching viewpoint to have a high end GPU paired with a low end one (or the internal IGPU) and have it sitting idling while waiting for the lesser GPU to finish it's 50% share of the work.

I haven't tried more than two GPUs per host so I don't know if more than two GPUs can be combined. There is an app_config setup to revert to one task per GPU.

Mark

Hi Mark,

Does it complete the WU more than 50% quicker? (eg is it worth it).

Is it only Moo that you know has this?

Hi Chris,

I think it's reasonably balanced, no great advantage or disadvantage. My host with a single 970 takes 2240-2270 seconds for a WU, the host with dual 970s takes 1210-1230 per WUs although the single GPU setup is the more powerful card and this include setup and reporting time so I expect actual crunching times to be very similar. Both for 6144 credits.

Moo is the only GPU project that I'm aware of that has multi GPU tasks although I've been concentrating on tht for a while so things may have changed.

(Just noticed I have the top rated host on Moo )

Mark

Woodles wrote: (Just noticed I have the top rated host on Moo )

Thanks for the info on running 2 GPUs on one WU.

I've also noticed that the other top rated hosts are limiting tasks to one GPU at a time so that implies it's better. If I have time I'll see what the difference actually is.

Mark

Personally I would keep one task per GPU, as that way you can move around projects. Its one reason that I quite like having 2/3 970's over a single 1080, even though TDP is much lower. Perhaps this is a factor in people's decision.