base kernel driver installs with WARNING but will not load on Dell R750 server running OpenSuse Leap 15.3 with Linux kernel 5.3.18
I am installing the ROCm all open driver stack 5.2.1 with the --usecase=openclsdk, but the same thing happens if I don't use this option and install everything. I am running on a Dell R750 server with Intel Ice Lake CPUs and Linux kernel 5.3.18-15300.59.63-default. I downloaded the ROCm 5.2.1 version of amdgpu-install and installed it using zypper on OpenSuse Leap 15.3. The installation succeeds, but it gives the warning: WARNING: nomodeset detected in kernel parameters, amdgpu requires KMS The rocminfo app is not included in the installation for some reason, so we used clinfo to check if everything is working. clinfo reports
Number of platforms: 1
Platform Profile: FULL_PROFILE
Platform Version: OpenCL 2.1 AMD-APP (3452.0)
Platform Name: AMD Accelerated Parallel Processing
Platform Vendor: Advanced Micro Devices, Inc.
Platform Extensions: cl_khr_icd cl_amd_event_callback
Platform Name: AMD Accelerated Parallel Processing
Number of devices: 0
dkms status reports
amdgpu, 5.16.9.22.20-1447096, 5.3.18-150300.59.63-default, x86_64: installed (original_module exists)
When we try to edit he boot script using grub to remove 'nomodeswitch', the server will not boot at all, and we have to use grub to reinsert 'nomodeswitch' before we can get it to boot.
Is this a known incompatibility with Dell servers, since they seem to require 'nomodeswitch', and is there any way around it?
I do not understand why we need nomodeswitch or KMS, since we are running this as a headless server, except when we log onto the network KVM system console. Since KMS is associated with setting display driver modes during boot up, as I understand it, is there some way we can inhibit the need for this, inasmuch as we don't want the GPU to display anything. We are using it for computation.
Cordially,
Walter F. Kailey, Ph.D. Performance Engineer Senior Software Development Engineer Ricoh USA