Unable to select GPU node with qsub

[atuft]

Hello,

I'm having some difficulty selecting a GPU node on DevCloud. When I run this command to request a GPU node:

qsub -I -l nodes=1:gpu:ppn=2,walltime=01:00:00

I find that the node I am allocated doesn't actually have a GPU device, despite pbsnodes claiming that it does e.g. for the s001-n166 node:

pbsnodes | grep -A6 "^s001-n166"
s001-n166
     state = job-exclusive
     power_state = Running
     np = 2
     properties = xeon,cfl,e-2176g,ram64gb,net1gbe,gpu,gen9
     ntype = cluster
     jobs = 0-1/1926896.v-qsvr-1.aidevcloud

The output from clinfo does not show a GPU:

clinfo | grep " Device Name "
  Device Name                                     Intel(R) FPGA Emulation Device
  Device Name                                     Intel(R) Xeon(R) E-2176G CPU @ 3.70GHz
    Device Name                                   Intel(R) FPGA Emulation Device
    Device Name                                   Intel(R) FPGA Emulation Device
    Device Name                                   Intel(R) FPGA Emulation Device

I cannot see a GPU device from within SYCL when running on a device with the gpu property - see this example code:

#include <CL/sycl.hpp>
#include <iostream>

int main() {
    for (auto &device : sycl::device::get_devices()) {
        std::cout << device.get_info<sycl::info::device::name>() << std::endl;
    }
}

Output when executing this on s001-n167:

Intel(R) FPGA Emulation Device
Intel(R) Xeon(R) E-2176G CPU @ 3.70GHz
SYCL host device

Is there an issue with the qsub properties at the moment, or am I doing something wrong?

Thanks,

Adam

[AlekhyaV_Intel · Accepted solution]

Hi Adam,

Thank you for posting in Intel Communities. We would like to know which DevCloud you're using. Are you using DevCloud for oneAPI/FPGA/Edge?

We tried to access GPU nodes, i.e. s001-166 & s001-167 on DevCloud for oneAPI and we could see that the graphics card is present. We have attached the screenshots below.

Please try once again with the below commands to access the above mentioned nodes & see if the issue stills persists. If yes, please provide error screenshots to us so that we can debug your issue further.

To access a specific node:

qsub -I -l nodes=s001-166:ppn=2 -d .

qsub -I -l nodes=s001-167:ppn=2 -d .

Run the command which you provided:

clinfo | grep " Device Name "

Regards,

Alekhya

[atuft]

Thanks for your reply! I have no idea what has changed since yesterday but I can now see the GPUs when I request a GPU node.

[AlekhyaV_Intel]

Hi,

Glad to know that your issue is resolved. If you need any further assistance, please post a new question as this thread will no longer be monitored by Intel.

Regards,

Alekhya