Intel® DevCloud
Help for those needing help starting or connecting to the Intel® DevCloud
1626 Discussions

oneAPI with GPU (GEN9 and IRIS_XE) not working

garsanca
Novice
1,843 Views

Hi to all,

I dont know if anyelse has notice but queues with GPUs (gen9 and iris_xe_max) may have changed OpenCL driver or any configuration but are not working anymore.

Nodes in gen9 queue seems to work with oneAPI but the job doesnt finish. In fact I have tested the well-known vector-adition and the behaviour is the same. It also happens in the same nodes when cpu_selector{} is selected in a oneapi source-code.

 

Nodes in iris_xe_max (both dual_gpu and quad_gpu) doesnt work by a problem with OpenCL runtime:

Running on Intel(R) Graphics [0x4905]
terminate called after throwing an instance of 'cl::sycl::runtime_error'
what(): Level 0 error occurred 1879048196 (Unknown OpenCL error code)
Aborted

 


Has anyone observed similar behavior or managed to run a job with oneAPI?

 

Regards,

 

Carlos

0 Kudos
1 Solution
garsanca
Novice
1,699 Views

Hi Gopika,

 

we have been able to run our applications in the nodes you mention.

 

Thanks a lot,

 

Carlos

View solution in original post

0 Kudos
7 Replies
Gopika_Intel
Moderator
1,823 Views

Hi,

 

Thank you for posting in Intel Devcloud forum and reporting this issue. We’ll be informing this to the concerned team. We’ll get back to you when we get an update.

 

Regards

Gopika


0 Kudos
Gopika_Intel
Moderator
1,801 Views

Hi,

 

We’ve reported the issues with gen9 and iris_xe_max. Here are the list of iris_xe_max nodes in which you can work in the meantime. Vector addition was working in these nodes.

 

Dual

 

s011-n001

s011-n002

s011-n003

s011-n004

s011-n007

 

Quad

s012-n003

s012-n004

 

Regards

Gopika


0 Kudos
garsanca
Novice
1,798 Views

Hi Gopika_Intel,

 

I have tested the Iris Dual GPU's nodes (listed in your mail) and seems to be run as expected.

Thank you for your efforts.

 

Kind regards,

 

Carlos

0 Kudos
Gopika_Intel
Moderator
1,788 Views

Hi,

 

The nodes are subjected to device changes, HW changes, driver changes or some nodes might even be offline. So we suggest you try using the below methods that will guarantee that any available iris_xe_max or gen9 nodes will run the workloads, instead of going for specific node numbers.

 

qsub -l nodes=1:ppn=2:iris_xe_max -d . run.sh
qsub -l nodes=1:ppn=2:gen9 -d . run.sh

 

Hope this helps.

Regards

Gopika

 

Gopika_Intel
Moderator
1,765 Views

Hi,

 

We haven’t heard from you. Did the above-mentioned generic method work for you? Can we discontinue monitoring this thread? Let us know your updates.

 

Regards

Gopika


0 Kudos
Gopika_Intel
Moderator
1,708 Views

Hi,

We haven’t heard from you in a while. We’re assuming the solution provided worked for you. We are no longer monitoring this thread. For further queries and issues, please raise a new thread. We’d be glad to help you.

Regards

Gopika


0 Kudos
garsanca
Novice
1,700 Views

Hi Gopika,

 

we have been able to run our applications in the nodes you mention.

 

Thanks a lot,

 

Carlos

0 Kudos
Reply