This repository has been archived by the owner on Jun 5, 2023. It is now read-only.
No OpenCL Devices are detected #111
Comments
|
Can you please check which kernel version you are running (eg output of "uname -a") ? Canonical pushed a kernel update out to some 20.04.1 users bumping it to 5.8, and the current DKMS kernel will not install on 5.8. |
|
Thank youf for you repl. So its a 5.4 Kernel. Thank you for your help so far. |
|
Hello? Any ideas how I can debug this further ? |
|
There are some environment variables that enable more debug output that can help diagnose the problem. I can never remember the one in OpenCL. This one is more low-level: HSAKMT_DEBUG_LEVEL=7 |
|
@fxkamd thank you for your help, what command are your refering to? Cliinfo has exactly the same output: |
|
Here is the rocminfo output: |
|
That's very strange. This should have produced some output on stderr. If not, it means it's not even getting to the low-level Thunk functions and failing somewhere higher up the stack. Or you could check one level lower to see what KFD is reporting: ls -lR /sys/class/kfd/kfd/topology Also check your kernel log with dmesg. |
|
@fxkamd Thank you, here are the outputs ls: ls: |
|
Here is your problem: KFD (more specifically the AQL firmware) on your Ellesmere GPUs requires PCIe v3 with support for atomic operations. Your mainboard doesn't seem to support that. On some mainboards, some slots support PCIe v3, while others do not. You could try experimenting with that. But given the age of your CPU, your chipset may just not have PCIe v3 support at all. |
|
I did tryied on every slot. My board is an ASRock H110 Pro BTC + Even amdpro drivers do not work. |
|
I looked up the spec or your main board: https://www.asrock.com/mb/Intel/H110%20Pro%20BTC+/index.asp |
|
The AMD Linux Pro driver should work though. As far as I know, it uses a legacy OpenCL implementation that does not depend on ROCm. In the 20.45 release it moved to a ROCm-based OpenCL implementation, but only for Vega and later GPUs. |
Hi @alfredopalhares, I was having the same issue and the Hope it works out for you. All the best! |
Further investigated this. The command is referring to the mesa version( |
|
@ALL , When ran: But factually there are two AMD gpus on my computer. (When ran: When ran: What could be the cause and how to solve it? |
I have exactly same situation with all driver versions. |
|
I have the same problem, since 4.5 my gpu is no longer detected with OpenCL |
|
Same issue here. I am using an RX 580 on Ubuntu 20.04 And this is my rocminfo output: I installed the drivers using this: https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#installing-a-rocm-package-from-a-debian-repository |
|
I have the same issue, on rocm 4.5.2, as well as all the earlier versions i've tried: Motherboard is set to use PCIe 3.0 on the slot of the 5700xt gpu. Motherboard is P8Z77-V DELUXE. From dmesg: kfd kfd: amdgpu: skipped device 1102:731f, PCI rejects atomics 142<145 |
|
Same issue here
|
|
Hello tried installing amd drivers some installers work other result in error the last attempt which outputs are listed above were rendered with sudo dpkg -i amdgpu-install_22.20.50200-1_all.deb |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Hello, I have 4 AMD 480
On Ubuntu 20.04 all update, followed the install intructions
But cliinfo does not detect any cards:
No OpenCL applications work, no devices are detected. Can you help me diagnose this problem ?
The text was updated successfully, but these errors were encountered: