Windows 10 suddenly stopped folding

[russell_heater]

I'm getting this right now for my AMD card.

02:38:59: GPU 0: Bus:3 Slot:0 Func:0 AMD:6 Navi 10 [Radeon RX 5600 OEM/5600 XT /
02:38:59: 5700/5700 XT]
02:38:59: CUDA: Not detected: Failed to open dynamic library 'nvcuda.dll': The
02:38:59: specified module could not be found.
...
02:38:59:WARNING:FS01:No CUDA or OpenCL 1.2+ support detected for GPU slot gpu:3:0 Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT]. Disabling.

[JimboPalmer]

russell_heater, welcome to Folding@Home!

You do not give us much information about why OpenCL is not loading. If you post the first 200 lines of your log, (in a thread about YOUR problem) we would know much more about your PC environment.
viewtopic.php?f=24&t=26036

[anandhanju]

@russell_heater,

1. Do you have the latest motherboard drivers or AMD drivers for the card?
2. What is your Windows version (as reported by the command winver)?

[Neil-B]

Perhaps @fah-toq can confirm whatcuda install worked? ... but obviously it might just not work for your issue even though it appears similar ... it is mad solution anyways !!

[russell_heater]

Responding to each point in order:

1.
- Latest AMD drivers (according to the Radeon auto-updater, which says itself was released on Oct/16. I only just installed Windows on this computer two days ago.)
- Mobo drivers are latest delivered by non-"optional" Windows Update. I supposed I can try installing the optional Windows updates...

2. Latest Windows 10 Pro (20H2)

3. After my earlier experiment, I then installed the full CUDA suite. No dice. Also, I noticed there is no "nvcuda.dll" in either of the dirs created by the installer under "c:\Program Files", so I'm not sure this file even comes from nVidia.

[PantherX]

Please note that installation of CUDA Toolkit currently causes issues. Just installing the latest Nvidia Driver should work. In a future version of FahCore_22, that issue can hopefully be resolved.

[russell_heater]

The driver for which Nvidia card? I don't have any Nvidia cards. Should I just pick an arbitrary Nvidia card driver?

[PantherX]

Oh, sorry for the confusion russell_heater. I was simply commenting that currently having CUDA Toolkit on your system will cause FahCore_22 to fallback to OpenCL instead of CUDA. To use CUDA on Nvidia GPUs, you simply need to use the latest Nvidia drivers.

In your case, you have an AMD GPU so in theory, you don't need Nvidia Drivers or CUDA Toolkit. This error in the log file is always expected on an AMD GPU system:

Code: Select all

CUDA: Not detected: Failed to open dynamic library 'nvcuda.dll': The
15:55:46:               specified module could not be found.

[bruce]

The FAHClient should be enhanced to count the number of AMD GPUs and the number of nV GPus. Then if there are zero nVidia GPUs, inhibit printing the warning message. Nevertheless that's an unimportant enhancement.

To discover the number of nVidia GPUs, the client needs the drivers and the nV run-time code, so it has to process the information in the opposite order. If the nvidia run-time is not present, no nV GPUs will be detected anyway so it makes sense to warn you that it won't be able to detect nV GPus if you happen to have any.

For those of you who have AMD GPUs there is no reason to install CUDA. That message does NOT indicate an error. For anyone with an nVidia GPU, that message explains why FAH may not detect your nV GPU.

[psaam0001]

I am glad that you solved the issue... however, that's a very unusual method and one that I wouldn't think was possible. However, who am I to question the IT Gods.

The nvcuda.dll error in your log is expected since you have and AMD GPU and that error is only for CUDA detection. Given that you don't have Nvidia GPU, we won't recommend you to install any Nvidia drivers or CUDA software package.

When I look at the Software Installation tool that comes with the Fedora distributions (not the 'dnf update' process as root user in Terminal), it will offer the NVidia drivers to install when I look under 'Utilities' --> 'Hardware Drivers' tab (even though that laptop has ATI/AMD legacy graphics support). I just ignore it altogether, as: 1) even if those drivers worked for the GPU series I have in it, F@H's requirements of support for OpenCL 1.2 are not met; and 2) I'm not messing with something that may cause system instability, to the point where the only recourse is a clean reinstall.

Hopefully in the future, AMD will unify their Workstation, Gaming/Average Use GPU card, and iGPU drivers into one package (as NVidia has done). And give clearer manual installation instructions for the major Linux distributions (on their site).

Paul

[bruce]

3. After my earlier experiment, I then installed the full CUDA suite. No dice. Also, I noticed there is no "nvcuda.dll" in either of the dirs created by the installer under "c:\Program Files", so I'm not sure this file even comes from nVidia.

You are not listening. CUDA is not for AMD GPUs; it is for nVidia GPUs. You said you have an AMD GPU. Ignore the message. It is a spurious message on systems with AMD GPUs.

[anandhanju]

@bruce, the error here: 02:38:59:WARNING:FS01:No CUDA or OpenCL 1.2+ support detected for GPU slot gpu:3:0 Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT]. Disabling.

There have been a few reports of AMD Radeon RX 5600 OEM/5600 XT / 5700/5700 XT cards not working due to this despite having the AMD drivers installed. What has confusingly worked in some cases is the installation of the CUDA tookit (I know, it sounds crazy) which has made this OpenCL detection failure go away.

I've compiled these reports into an issue in Github -- https://github.com/FoldingAtHome/fah-issues/issues/1589