Since this is not really a "bug" (but is an officially proposed use case from Intel), it's not likely to provide a test case. But from HPE's testing of the test kernel from comment 2, there is improvement to performance on a Sapphire Rapids CPU with HBM.
As described in https:/ /community. intel.com/ t5/Blogs/ Products- and-Solutions/ HPC/Enabling- High-Bandwidth- Memory- for-HPC- and-AI- Applications- for/post/ 1335100,
for Intel Sapphire Rapids CPUs with HBM, it is possible to utilize the HBM as cache (so called "2LM mode"). It works in the way to create N fake NUMA domains in a size aligned the HBM so that HBM can act like direct-mapped L4 cache.
Since this is not really a "bug" (but is an officially proposed use case from Intel), it's not likely to provide a test case. But from HPE's testing of the test kernel from comment 2, there is improvement to performance on a Sapphire Rapids CPU with HBM.