GPU RDMA Memory Ordering Limitations

A severe limitation of current GPUDirect RDMA is lack of sub-kernel level memory ordering with GPUDirect RDMA peers. Specifically, memory ordering between a peer device and a GPU kernel is only enforced at kernel boundaries. This is described in section 2.7 of the GPUDirect RDMA documentation:

Several papers:

Is there any update to this limitation that I’ve missed, or plans to address this limitation in future hardware?

Hello ryan.walton,

Thank you for posting to the NVIDIA Developer Forums.

We recommend reaching out to our Sales and Solutions Engineering team with this inquiry - they will be able to address this for you.

You can reach out to our sales team at the following link:
https://www.nvidia.com/en-us/contact/sales/

Best regards,
NVIDIA Enterprise Experience

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.