search

intel / intel-device-plugins-for-kubernetes

Collection of Intel device plugins for Kubernetes
Apache License 2.0
47 stars 204 forks source link

Intel arc support #1109

Closed atotalnoob closed 2 years ago

atotalnoob commented 2 years ago

Hi,

Since the arc a380s are now available, if/when will arc be supported for kubernetes workloads? Does it follow the normal Intel discrete GPU setup?

uniemimu commented 2 years ago

From the point of view of this project, a380 should already work with the latest Intel GPU-plugin. Workload containers can get the GPU device, if you add the GPU resource request to the workload POD specification.

eero-t commented 2 years ago

@atotalnoob If your kernel supports / exposes given (Intel GPU) device to user-space, GPU plugin supports it.

Whether you can then use the GPU exposed to your k8s container workload by the GPU plugin, depends on the user-space (Intel GPU) driver(s) included to that container.

Intel provides pre-built driver packages: https://dgpu-docs.intel.com/installation-guides/index.html

But currently it has only user-space drivers, not kernel one.

If you want to build drivers from upstream sources...

[1] In case of current upstream kernel, Intel dGPU support is not enabled by default quite yet, you need to use kernel "i915.force_probe=GPU-PCI-ID" probe option to enable it (PCI ID is listed e.g. in "lspci" output). This is because things may not work optimally yet, due to some final parts listed here still missing from upstream: https://www.kernel.org/doc/html/latest/gpu/rfc/index.html

Upstream user-space drivers do have support for the new HW, but you need to check release notes for level of HW support included to given release (and whether it's enabled in given project release binaries):

mythi commented 2 years ago

labeled as a 'question' which also got answered.