I know that thread block clustering for accessing shared memory of different thread blocks has been a concept introduced in the Hopper architecture.
I wonder if thread block clustering is not supported in the Ada architecture that is parallel to Hopper. To be specific, the AD102 GPU.
Ada is compute capability 8.9, Hopper is 9.0. Let’s see, if it will be there with Blackwell.
Thread block clustering requires cc9.0 (or higher, perhaps). The Ada architecture is cc8.9. Thread block clusters are not supported on Ada.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.