-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Compatibility
Matthew Nicely edited this page Dec 15, 2021
·
4 revisions
CUTLASS requires a C++11 host compiler and performs best when built with the CUDA 11.5 Toolkit. It is also compatible with CUDA 11.0, CUDA 11.1, CUDA 11.2, CUDA 11.3, and CUDA 11.4.
We have tested the following environments.
| Operating System | Compiler |
|---|---|
| Windows 10 | Microsoft Visual Studio 2015 |
| Microsoft Visual Studio 2017 | |
| Ubuntu 18.04 | GCC 7.5.0 |
| Ubuntu 20.04 | GCC 10.3.0 |
Additionally, CUTLASS may be built with clang. See these instructions for more details.
CUTLASS runs successfully on the following NVIDIA GPUs, and it is expected to be efficient on any Volta-, Turing-, or NVIDIA Ampere- architecture NVIDIA GPU.
For all GPUs, we recommend compiling with the CUDA 11.5 Toolkit for best performance.
| GPU | CUDA Compute Capability | Minimum CUDA Toolkit | CUDA Toolkit Enabling Native Tensor Cores |
|---|---|---|---|
| NVIDIA Tesla V100 | 7.0 | 9.2 | 10.1 |
| NVIDIA TitanV | 7.0 | 9.2 | 10.1 |
| NVIDIA GeForce RTX 2080 TI, 2080, 2070 | 7.5 | 10.0 | 10.2 |
| NVIDIA Tesla T4 | 7.5 | 10.0 | 10.2 |
| NVIDIA A100 | 8.0 | 11.0 | 11.0 |
| NVIDIA A10 | 8.6 | 11.1 | 11.1 |
| NVIDIA GeForce 3090 | 8.6 | 11.1 | 11.1 |