clear heron
I got everything installed but it's only using CPU-
Fresh Ubuntu 20.04 VM
GPU passed through
Followed linux install instructions https://github.com/invoke-ai/InvokeAI/blob/main/docs/installation/INSTALL_LINUX.md (though ln -sf on the model caused problems so I just copied it instead.
pip install -r requirements-lin-AMD.txt
Notably, I have not installed any amd drivers or anything like that yet. Wanted input before I start installing every driver and library I can find :p
lspci
00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
00:01.0 VGA compatible controller: Red Hat, Inc. QXL paravirtual graphic card (rev 05)
00:02.0 PCI bridge: Red Hat, Inc. QEMU PCIe Root port
[etc]
00:1f.0 ISA bridge: Intel Corporation 82801IB (ICH9) LPC Interface Controller (rev 02)
00:1f.2 SATA controller: Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH) 6 port SATA Controller [AHCI mode] (rev 02)
00:1f.3 SMBus: Intel Corporation 82801I (ICH9 Family) SMBus Controller (rev 02)
01:00.0 Ethernet controller: Red Hat, Inc. Virtio network device (rev 01)
02:00.0 Communication controller: Red Hat, Inc. Virtio console (rev 01)
03:00.0 SCSI storage controller: Red Hat, Inc. Virtio block device (rev 01)
04:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT] (rev c1)
05:00.0 Audio device: Advanced Micro Devices, Inc. [AMD/ATI] Navi 10 HDMI Audio
06:00.0 USB controller: Advanced Micro Devices, Inc. [AMD] Zeppelin USB 3.0 Host controller
lsmod | grep amdgpu
amdgpu 9805824 1
iommu_v2 24576 1 amdgpu
gpu_sched 45056 1 amdgpu
i2c_algo_bit 16384 1 amdgpu
drm_ttm_helper 16384 2 qxl,amdgpu
ttm 86016 3 qxl,amdgpu,drm_ttm_helper
drm_kms_helper 307200 3 qxl,amdgpu
drm 618496 9 gpu_sched,drm_kms_helper,qxl,amdgpu,drm_ttm_helper,ttm
quasi coyote
I guess the starting point would be do you have rocm installed ?
sharp moss
Yeah, there's definitely a requirement on the Ubuntu side to have some kind of driver
clear heron
$ rocm-smi
======================= ROCm System Management Interface =======================
WARNING: No AMD GPUs specified
================================= Concise Info =================================
GPU Temp AvgPwr SCLK MCLK Fan Perf PwrCap VRAM% GPU%
================================================================================
============================= End of ROCm SMI Log ==============================
I'm getting an issue with amdgpu-install where it doesn't like dkms, but even with --no-dkms I still get the error
$ sudo apt install rocm-opencl-runtime
Reading package lists... Done
Building dependency tree
Reading state information... Done
rocm-opencl-runtime is already the newest version (5.1.0.50100-36).
0 upgraded, 0 newly installed, 0 to remove and 1 not upgraded.
1 not fully installed or removed.
After this operation, 0 B of additional disk space will be used.
Do you want to continue? [Y/n] y
Requesting to save current system state
Successfully saved as "autozsys_dhvg96"
Setting up amdgpu-dkms (1:5.13.20.5.1.50100-1395274) ...
Removing old amdgpu-5.13.20.5.1-1395274 DKMS files...
------------------------------
Deleting module version: 5.13.20.5.1-1395274
completely from the DKMS tree.
------------------------------
Done.
Loading new amdgpu-5.13.20.5.1-1395274 DKMS files...
Building for 5.15.0-50-generic
Building for architecture x86_64
Building initial module for 5.15.0-50-generic
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms-firmware.0.crash'
Error! Bad return status for module build on kernel: 5.15.0-50-generic (x86_64)
Consult /var/lib/dkms/amdgpu/5.13.20.5.1-1395274/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Errors were encountered while processing:
amdgpu-dkms
gonna try downgrading my kernel 🙃
or not- ```
Command failed.
Adding boot menu entry for UEFI Firmware Settings
done
Errors were encountered while processing:
amdgpu-dkms
ZSys is adding automatic system snapshot to GRUB menu
E: Sub-process /usr/bin/dpkg returned an error code (1)
I don't understand this error. I'm telling it specifically not to do anything with dkms and yet
$ sudo amdgpu-install --usecase=rocm --no-dkms
Hit:1 http://us.archive.ubuntu.com/ubuntu focal InRelease
Get:2 http://us.archive.ubuntu.com/ubuntu focal-updates InRelease [114 kB]
Fetched 397 kB in 1s (400 kB/s)
Reading package lists... Done
Reading package lists... Done
Building dependency tree
Reading state information... Done
rocm-dev is already the newest version (5.1.0.50100-36).
0 upgraded, 0 newly installed, 0 to remove and 1 not upgraded.
1 not fully installed or removed.
After this operation, 0 B of additional disk space will be used.
Do you want to continue? [Y/n] y
Requesting to save current system state
Successfully saved as "autozsys_tnrh66"
Setting up amdgpu-dkms (1:5.13.20.5.1.50100-1395274) ...
Removing old amdgpu-5.13.20.5.1-1395274 DKMS files...
------------------------------
Deleting module version: 5.13.20.5.1-1395274
completely from the DKMS tree.
------------------------------
Done.
Loading new amdgpu-5.13.20.5.1-1395274 DKMS files...
Building for 5.15.0-50-generic
Building for architecture x86_64
Building initial module for 5.15.0-50-generic
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms-firmware.0.crash'
Error! Bad return status for module build on kernel: 5.15.0-50-generic (x86_64)
Consult /var/lib/dkms/amdgpu/5.13.20.5.1-1395274/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Errors were encountered while processing:
amdgpu-dkms
I haven’t touched dkms in years 🤷
carmine radish
Oh lord. Kernel module hell. It's been a while since I've run into it but I still have nightmares.
Try looking in make.log at the path mentioned in the error message. My guess is that you're going to find that there is a missing build requirement, or possibly that the code doesn't compile cleanly with your current kernel.
clear heron
It's the default kernel on ubuntu-20.04.05 or whatever the current 20.04 is
That said I tore down the VM in frustration after I installed 3 kernels lol. I'll try again with 22.04 maybe tomorrow
carmine radish
Oh, you're running this in a VM on a Windows machine? You know for sure that the VM can use the GPU?
clear heron
No it's a VM on Unraid; along side my Windows VM, so I know things like GPU passthrough are working correctly
clear heron
Lots of research later, it seems my gpu is just plain unsupported by rocm. It's a 5700xt/navi10/gfx1010
coral fog
I believe ROCm should work on Navi10. Do you remember where you read that?
You might have to compile pytorch yourself with ROCm on gfx1010.
I'll do some testing myself, but It'll have to wait till after work tomorrow. (1700 EST)
clear heron
@coral fog I'm not seeing Navi10 as being supported for ROCm/computehttps://github.com/RadeonOpenCompute/ROCm/issues/998
