Trying GPUNetIO sample with BF2 and GPU (cc = 70), adding the following in meson.build:
nvcc_flags += [‘-gencode’, ‘arch=compute_70,code=sm_70’]
Showing the following when running application with any of doca_xxx function in kernel function:
[09:35:04:787461][DOCA][ERR][GPU_SEND_WAIT_TIME::KERNEL:97]: […/device/gpunetio_send_wait_time_kernel.cu:97] cuda failed with device kernel image is invalid
If only “printf” in CUDA kernel function, the sample completed successfully.
Does GPUNetIO in DOCA and DPU currently support GPU cc < 80?
Hi iceiilin,
Thank you for posting your query on NVIDIA Community.
In case by GPU “cc”, if you are referring to GPU Compute Capability, then I would like to inform that GPUNetIO has no minimum GPU cc requirement.
I would also like to bring to your notice that support related to Bluefield2 requires an active support entitlement.
In case you do not have an active entitlement, please reach out to contracts team at Networking-Contracts@nvidia.com
If you have an existing entitlement, please do not hesitate to open a support ticket as the current information provided is not sufficient to debug the addressed issue. The support ticket can be opened by emailing enterprisesupport@nvidia.com
Thanks,
Namrata.