Folding Forum

I'm hoping someone here will know how to solve this.

On CentOS 8

I have a machine with two NVIDIA cards: a GTX 1060 for folding, and a Quadro NVS 290 for running the display. I know the Quadro won't fold.

I managed to get the 1060 driver installed, and it's happily folding.

I THINK the 1060 installation configured X to use it instead of the Quadro. And I THINK when I install the Quadro driver, it'll configure X to use the Quadro, which is what I want. It had been running with nouveau.

I've got the run level at MultiUser (no graphics) and Folding@Home turned off so there should be no graphics driver active.

I go to install the Quadro driver, and I get this: I don't WANT to uninstall the 1060 driver.

There are a great many options on the installer script, many I do not understand, but none look like "Don't uninstall another driver because the system has two NVIDIA GPUs and we want them both."

Has anybody here solved this before?

Welcome to Folding@Home!

Out of curiosity, why don't you want to use the GTX 1060 as the display?

The Quadro NVS 290 supports

DirectX 11.1 (10_0)
OpenGL 3.3
OpenCL 1.1

As far as FAH is concerned, only OpenCL 1.1 matters and OpenCL 1.2 is the minimum required.

The GTX 1060 supports

DirectX 12 (12_1)
OpenGL 4.6
OpenCL 1.2
CUDA x.x

FAH will prefer CUDA but if for some reason it doesn't set up properly, it will revert to OpenCL 1.2.

When my system was configured to us nouveau, FAH could not use my NVidia GPU but I didn't have a second (older) GPU so my situation was different than yours.

I don't know what's involved with the use of DirectX or OpenGL for other applications on your system but it doesn't seem to matter for anything I use so I suppose either GPU display the desktop.

I assume you want to fold with the 1060.

When FAHClient initializes, it will detect the GPUs provided the drivers are recognized. That is reported about 100 lines from the top of FAH's log. Also, when a WU has been downloaded, it should invoke FAHCore_22 and the first parts of the log will be interesting. Please post those segments of the log.

> Out of curiosity, why don't you want to use the GTX 1060 as the display?

The most immediate reason is I don't have a DVI-D ==> VGA adapter for the KVM switch I don't have a straight DVI-D cable either. But I was poking around the X configuration, and I see it isn't configured to use the 1060 display after all.

But mainly, I'm trying to learn about Linux at this level. I have not fooled with Unix since the middle 1990s and I never had to mess with device drivers. Only a tiny bit of X configuration.

@bruce Yes, the 1060 is folding happily. I can't seem to install the NVIDIA driver for the Quadro NVS 290. I'm beginning to think it is impossible to install two different NVIDIA drivers on Linux.

NVIDIA say this:
"When using two or more NVIDIA cards in the same system, both cards need to use the same NVIDIA driver to avoid conflicts."
[URL=https://nvidia.custhelp.com/app/answers ... /a_id/2280]

They don't quite say "you can't install two different drivers"

A customer (MrInfinit3)
says this:
"you cannot install two differing driver packs as one will always override the other."
https://www.nvidia.com/en-us/geforce/fo ... -together/

A different customer (Robert_Crovella) says:
"You can only load one driver, so the driver you use must support both cards"
https://forums.developer.nvidia.com/t/q ... stem/42084

The driver that supports the GTX 1060 does not support the Quadro 290, so I guess I'm stuck.

I suppose I could continue using nouveau to drive the Q 290. I guess it won't try to snag the 1060 as well. (Yet another question)

trleith wrote:> Out of curiosity, why don't you want to use the GTX 1060 as the display?

The most immediate reason is I don't have a DVI-D ==> VGA adapter for the KVM switch I don't have a straight DVI-D cable either.

https://www.amazon.com/Gold-Plated-Comp ... B07X41K6NF

if you have a VGA to VGA cable already

https://www.amazon.com/Friencity-DVI-VG ... B079MD7CWF

@JimboPalmer Yes, thank you. I have one coming if I decide to give up entirely on the Q 290. Which seems likely.

trleith wrote:I suppose I could continue using nouveau to drive the Q 290. I guess it won't try to snag the 1060 as well. (Yet another question)

Many of the NV drivers support quite a range of hardware but it looks like the two you have are just too far apart. I guess the best option is to uninstall the NVS 290. If you want something to fill the slot, check the website's description of the driver and see what it supports.

Without looking, I'll bet anything >= a GT710 (Kepler) would be supported. If you're going to buy a second GPU, you can get a lot newer (more powerful for FAH) GPU than the 710 for almost the same price. Donate the NVS 290 to somebody who doesn't run FAH. That device came out in 2007 and was deprecated by FAH many years later but it can still drive a display.

Another option is to run the folding GPU or even your desktop virtualized, if your CPU and motherboard supports IOMMU redirection of PCIe lanes to a virtual guest OS running under KVM/Qemu.

@Bruce -- yes, exactly. The correct term (I have learned) is "Driver Pack". You can install only one NVIDIA Driver Pack and the one that includes the NVS 290 driver does not include the GTX 1060. I think what NVIDIA are doing is making sure their legacy drivers work with newer Linux Kernels, but they're not putting everything in one Grand Unified Driver Pack. I guess.

With the NOUVEAU driver, FaH does not recognize the NVS 290 as a GPU at all.

@gunnarree -- that is creative. I shall have to remember this possibility in case anybody wants me to do something absolutely nuts

Right now, I'm giving FaH 6 Xeon cores and the 1060, basically full time: I paid a hundred bucks for it. I don't really have a use for a GPU otherwise, so I won't likely buy a second GPU. If somebody has a spare 1060 or better lying about, you can ship it to me and I'll commit to running it for two years.

But today I got my DVI adapter, so I'm going to yank the NVS 290, move the 1060 to Slot 0 because I think that's where it has to be during boot, and see how it works. Then configure X because I want to be reminded how. Then figure out why the latest kernel doesn't work (I'm using the one prior). There is a great deal I do not understand about the boot process, but that hasn't anything to do with FaH.

Thanks to everyone.