CUDA Toolkit 12.8 what GPU is 'sm_120'?

[FrankM_72]

Yes, I faced that problem too.

Registers management 8.9 vs 9.0+

Hello community, I think I understand the conceptual changes in register file organization between 8.9 and 9.0 and I understand that different allocation granularity can cause more leaks. But… Is there a way to fix this or at least try to avoid this problem? The same problem also occurs in Blackwell. The same code compiled for 8.9 with a register limit of 128 does not leak, but compiled for 120 leaks ± 120 bytes; leaks 112 bytes using ccap100a and leaks 180 bytes when compiled for ccap90. Of …

But I’m not sure it is issue just with sm_100,… did you try to compile your code for Hopper sm_90?

The compiler for sm_90 is forced to allocate registers in larger “chunks” (for example, in pairs or in fixed blocks) compared to sm_89. Even if your kernel only uses a given number of logical registers, the physical allocation on sm_90 might be higher because of these coarser allocation units. This means that code which fits comfortably into the register file when compiled for sm_89 may exceed the available physical registers on sm_90, leading to heavy spilling.