John Hubbard

Hi, HMM functionality is not available because our kernel driver does not yet support it. We are working on that, but we routinely do not provide timelines or ETAs for...

I'm referring to the code in this repo. In fact, maybe I should have written "kernel drivers", because both nvidia.ko and nvidia-uvm.ko are involved in supporting HMM. As for accelerating...

I think that this issue might *be* the public issue that tracks HMM support. :) Sure, I'll make a note to update this as part of the "release HMM support"...

That blog post was in error. After receiving your question here, we have corrected the blog to reflect that HMM is not yet supported in our driver. Thanks for alerting...

Yes, an early version of HMM support is included in the r530 release. However, as I wrote here: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/main/kernel-open/nvidia-uvm/uvm_hmm.c#L42 , it is not ready for production use. That's why it...

HMM depends upon the open source version of the driver. The open source version of the driver, in turn, only works on Turing and later GPUs. As it says in...

Right, no Windows support exists. The HMM feature required OS kernel changes, in addition to changes in our driver stack here. The open source Linux kernel, and the kernel community,...

HMM support is disabled by default in the r530 driver. That's why your sample is failing as shown above.

HMM is now supported with CUDA 12.2. Please see https://developer.nvidia.com/blog/simplifying-gpu-application-development-with-heterogeneous-memory-management/ for more information.

Thank you for the bug report. I have filed NVIDIA internal bug 4121956 to investigate this.