How do I fold with Two GPUs in One box under Windows 11?

appepi · Post by **appepi** » Mon Jun 30, 2025 12:29 pm

OK, I have just finished the tests I said I would do. On Z4G41-11, starting with the working 1080 in the upper PCI3 (x16) slot as usual, I added a P1000 in the lowest slot -PCI3 (x8). I got the usual result from Windows: the 1080 disabled with the complaint about driver problems. I fired up 7.6.21. The 1080 was disabled, the P1000 got a WU. I went to configure slots. There was a -1 (whatever it means) against one of the GPUs. I deleted it. I added a GPU. The 1080 still disabled. Nothing changed. I shut down, and installed 8.4.9. It did the same thing. the 1080 was seen but greyed out. No drivers were shown in the info. Nothing was supported. The P1000 got a WU. I shut everything down. I removed the P1000. When I fired up, the 1080 was still in the same state. I reinstalled the 1080 drivers. All became well and the 1080 is completing a WU under 8.4.9. I will then revert to 7.6.21. The problem is not a Fah problem and not solved by 8.4.9. It is Windows refusing to let me have two concurrently enabled GPUs. The only references I can find in the net for doing this are with laptops and integrated GPUs plus a separate GPU, and then it is about assigning apps to run on one OR the other.

muziqaz · Post by **muziqaz** » Mon Jun 30, 2025 12:37 pm

You need to find out the driver version which still supports both of your GPUs, it seems the latest drivers are not providing required CUDA libraries for older Nvidia GPUs or something.
Linux might be an option.
I doubt this is windows OS fault.
How are both GPUs seen in device manager?

appepi · Post by **appepi** » Mon Jun 30, 2025 1:09 pm

There is certainly some e-literature suggesting that it is better two have two GPUs supported by the same driver. But the latest NVIDIA driver (576.80 Studio version) that suits the 1080 runs perfectly well under 7.6.21 and 8.4.9 as long as it is alone and NVIDIA says it spans the 10xx through 50xx series,so it would apply to a large proportion of folding GPUs. On the other hand the latest P1000 driver is 475.14 and NVidia says it is a "superset" of the same version of Studio driver:

NVIDIA RTX Enterprise Production Branch Driver

Release 470 is a Production Branch (PB) release of the NVIDIA RTX Enterprise Driver. PB drivers are designed and tested to provide long-term stability and availability, making these drivers ideal for enterprise customers and other users who require application and hardware certification from ISVs and OEMs respectively.
The PB driver is a superset of the NVIDIA Studio Driver and provides all the benefits of the Studio Driver of the same version, in addition to NVIDIA RTX-specific enhancements and testing.

So ... if I go back to version 475.14 for BOTH .... Tomorrow night's experiment.

How are both GPUs seen in device manager?

One of them (the 1080 this evening) has a yellow triangle indicating a problem and then the error message that I quoted in one of my earlier posts on this topic. The other one is normal for a working GPU and its driver.

Post by **Joe_H** » Mon Jun 30, 2025 1:21 pm

When Nvidia says a driver version supports a range of GPUs, often they are only referring to support as a video card. You sometimes have to dig down into the notes to find that an older card doesn't get full support for compute use with OpenCL and/or CUDA support dropped or broken.

muziqaz · Post by **muziqaz** » Mon Jun 30, 2025 2:37 pm

Check your motherboard manual about multi GPU support. Some of them act funky if you are using nvme ssds next to secondary pcie slot.
Driver 475 might be a good candidate

muziqaz · Post by **muziqaz** » Mon Jun 30, 2025 2:40 pm

Quick Google of "code31 on secondary Nvidia GPU" shows driver conflict as a cause

appepi · Post by **appepi** » Mon Jun 30, 2025 2:47 pm

You sometimes have to dig down into the notes

... and sometimes it depends on which NVIDIA site you are on when you do your manual search for drivers. The last post was based on what https://www.nvidia.com/en-us/geforce/drivers/ had to say was OK for a GTX1080 (576.80 Studio driver at https://www.nvidia.com/en-us/drivers/details/247854/ ) and for a Quadro P1000 (475.14 PB driver). The difference in numbers seemed very large given that these GPUs are more or less contemporary with one another: GTX1080 (GP104) released May 2016; Quadro P1000 (GP107) released Feb 2017 (according to Techpowerup). And the stated coverage of 475.14 made no sense. By contrast, the same manual search for a Quadro P1000 at https://www.nvidia.com/en-us/drivers/ produced a much more recent PB driver released on the same date (17 Jun 2025) as Studio Driver 576.80, namely PB driver R570 U6 (573.42) at https://www.nvidia.com/en-us/drivers/details/247852/.

Both GPUs fold happily with their current drivers, and indeed have done so with other recent ones, so long as they are the only card in town. They also behave normally on Geekbench 6 compute tasks using OpenCL. But I was not testing drivers systematically then, so for my next experiment I will see if I can get away with installing ONLY PB Driver 573.42 and selecting it for both. That is, if the attic and I survive the 110Km/hr wind gusts headed our way according to http://www.bom.gov.au/products/IDN21037.shtml

appepi · Post by **appepi** » Mon Jun 30, 2025 3:05 pm

"code31 on secondary Nvidia GPU" shows driver conflict as a cause

Aha moment! Very helpful, unlike most MS error codes. It is still a puzzle that W11 would regard the 1080 as "secondary" when it's in the catbird seat up there at PCIe3 (x16) #1 and the P1000 is down on the bottom, although it is true that when they arrived the P1000 was the only GPU. Also, every device in the whole box has its own driver, so you would think that each adaptor x driver combination would be allowed to be unique and no conflict should arise.

But certainly all hints are tending towards using the latest driver likely to be common? As per my proposed experiment.

muziqaz · Post by **muziqaz** » Mon Jun 30, 2025 3:33 pm

At one point one of the GPUs is secondary. I don't think you mentioned exact pcie slots used for both GPUs.
Some motherboards share pcie lanes between some pcie slots and nvme slots

appepi · Post by **appepi** » Mon Jun 30, 2025 4:50 pm

From the desk of Z4G41-t11 at 2:40am: The NVIDIA Installer for PB Driver 573.42 says that it cannot continue because "... This graphics driver could not find compatible hardware." The only available and relevant hardware was the GTX1080.

Some further net searches found a user stating bluntly that the second driver loaded over-writes the first so only one can be "current". My previous installs of this driver before would have been while there was a P1000 in the box, so I never encountered problems installing it before. But with only the 1080 there, the incompatibility becomes obvious.

The take-home message seems to be that it may be possible to run two GPUs if the same driver will serve both. Perhaps I will test it with a Dual P1000 setup or pairs selected from 1080,1080,1070,1060. But not right now.

muziqaz · Post by **muziqaz** » Mon Jun 30, 2025 4:58 pm

Talk about nVidia "just works"...

appepi · Post by **appepi** » Tue Jul 01, 2025 4:06 am

"just works"...

if you install ONE driver that works for both GPUs, at least in the special cases of identical GPUs below when ...

... you put 2xP1000 GPUs in the 2 PCIe3 x 16 slots of Z4G42-11 and install PB driver R570 U6 (573.42) from https://www.nvidia.com/en-us/drivers/details/247852/ or...

.... you put 2xGTX1080 GPUs in (a) the top PCIe3 x 16 slot and (b) the second-from-bottom PCIe 3 (x4) slot of Z4G41-11 because you'd have to disconnect the SATA ports to get it in the bottom PCIe3 (x8) slot and the second 1080 would completely block the air intake of the first if you put it in the second PCIe3(x16) slot even if you could force it in there with a hammer, and once you discover that the 6+2 pin cable from the 1000W PSU fits if you try it upside down in the MSI/Dell 1080 relative to the ASUS one and you leave the door open because the airflow for both of them is now woeful. Oh, and if you install 576.80 Studio driver from https://www.nvidia.com/en-us/drivers/details/247854/.

For FaH the two P1000's installed easily enough with 7.6.21, but it struggled with the 2xGTX1080 setup and one was disabled, so I tried 8.4.9 and was able to set up both.

Z4G42-11 is now producing at 0.627 MPPD from the two P1000's and Z4G41-11 is now producing 3.15M PPD from the two RTX 1080s.

Thanks for all input and perhaps we have produced a useful guide for those afflicted with W11.

PS: It looks like this: https://docs.google.com/document/d/e/2P ... C8uVG3/pub

Folding Forum

How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?

Re: How do I fold with Two GPUs in One box under Windows 11?