Hi, lovely library you made; if I understood correctly, based on database/*.inc currently tensor core code is only emitted for FP16 and FP64 inputs; is TF32 support something that’s on the near-term roadmap?
Many thanks!
bergdorf
1
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Supported datatypes in cuTENSOR | 0 | 285 | October 28, 2020 | |
| Tensor Core utilization in cuDSS | 1 | 33 | March 12, 2025 | |
| [cuBLASDx] Support for MM where input type != output type? | 0 | 345 | February 3, 2024 | |
| Turing Arch - INT4 ops with tensor cores | 1 | 2042 | November 1, 2018 | |
| Question about tensor cores performance | 3 | 662 | October 12, 2021 | |
| Why tensor cores can't do FP32 arithmetic? | 4 | 259 | December 10, 2024 | |
| Just Released: cuSPARSELt v0.3 | 0 | 273 | June 21, 2022 | |
| Turing architecture compatibility with CUDA 8 | 2 | 480 | October 12, 2021 | |
| Dense LU Factorization with Double Precision | 1 | 401 | April 28, 2021 | |
| Tensor core boiler plate with cublas, can not compile | 3 | 20 | February 18, 2025 |