zeInit() fails with ZE_RESULT_ERROR_UNINITIALIZED on s017-n001

[Brice1]

Hello

I am on node s017-n001. This is quad-GPU node where I am using oneAPI L0 for hwloc development. My code has been working fine on other nodes, but it fails in the very first  L0 API call on this node: zeInit() returns error ZE_RESULT_ERROR_UNINITIALIZED (0x78000001).  The documentation says it means the driver is not initialized but I am not sure if that means the kernel driver (i915 module is loaded) or some user-space L0 driver.

$ dpkg -l | grep level-zero | cut -c5-70
intel-level-zero-gpu 1.3.24420+i504~u20.04
level-zero 1.8.5+i504~u20.04

On a working node, packages are i815 instead i504, not sure if it matters.
$ dpkg -l | grep level-zero | cut -c5-70
intel-level-zero-gpu 1.3.24055+i815~u20.04
level-zero 1.8.5+i815~u20.04
level-zero-dev 1.8.5+i815~u20.04

By the way, level-zero-dev is missing on s017-n001. I reported the issue in the past and was told that it's only available on GPU-nodes, which should be the case here.

Thanks

[Rahila_T_Intel]

Hi,

Thank you for posting in Intel Communities.

We are able to reproduce the issue on the node s017-n001 and have informed the concerned team regarding this.

Could you please let us know the node numbers in which you are able to run the same successfully?

Thanks.

[Brice1]

Hello
It works on all non-NDA nodes with GPUs I could test recently, this includes s019-n012, s001-n157, s001-n141 this morning.
Brice

[Rahila_T_Intel]

Hi,

Sorry to inform you that the node s017-n001 is work in progress and it will not be accessible.

Thanks

[Rahila_T_Intel]

Hi,

Hope your query got clarified. Could you please confirm whether we can close the case?

Thanks

[Brice1]

Hello. I don't see where anything got clarified. Both s017n001 and n002 are both "state = down,offline" since I reported this issue. I know some oneAPI developers are looking at it but I am not aware of any fix yet. This issue is certainly not resolved yet.

[Rahila_T_Intel]

Hi,

We are working on this internally and will share you the updates.

Thanks

[Rahila_T_Intel]

Hi.

Please be informed that quad nodes will become available early Q1 2023.

Hope this will clarify your query.

Please let us know if we can close this case.

Thanks

[Brice1]

Ok, thanks.

[Rahila_T_Intel]

Hi,

Thanks for confirming. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Thanks