Hi all,
can we get benefited with half precision or int8 computation on cuda cores.In terms number of computations per GPU cycle.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Mixed-Precision Programming with CUDA 8 | 1 | 437 | February 23, 2017 | |
| How can i Using (Mixed Precision/ Automatic Mixed Precision) INT8(IMMA) | 0 | 609 | October 4, 2019 | |
| Roofline Tensor Core should be half but not float? | 3 | 1576 | May 29, 2024 | |
| Tensor WMMA INT8 vs FP16 processing speed | 1 | 1971 | February 14, 2019 | |
| Int8 is 30% slower than fp16 in cudnn_samples_v8/conv_sample | 4 | 854 | February 8, 2023 | |
| Forward looking GPU integer performance | 22 | 22259 | March 20, 2017 | |
| A question on single and double precision performance calculation with CUDA cores | 7 | 2530 | May 31, 2024 | |
| How can i Using (Mixed Precision/ Automatic Mixed Precision) INT8(IMMA) | 0 | 437 | October 4, 2019 | |
| Poor half performance | 13 | 2596 | June 19, 2025 | |
| cuBLAS INT8 tensor core mode vs. FP16 mode | 0 | 938 | February 15, 2019 |