Solver supported GPU cards per version
The following table lists the supported graphics processing unit (GPU) requirements for accelerated thermal radiation computation by solver version.
| TMG version | Supported CUDA version | GPU architecture | Compute capability | Note |
|---|---|---|---|---|
| 2412 - 2506.6 | NVIDIA CUDA 12.4 or later | Maxwell to Hopper | 5.0 - 9.0 | With CUDA 12.4, support for GPU architectures prior to Maxwell is deprecated. |
| 2506.7-2512 | NVIDIA CUDA 12.9 or later | Maxwell to Blackwell | 5.0 – 12.1 | With CUDA 12.9, support for GPU architectures prior to Maxwell is deprecated. |
| 2606 | NVIDIA CUDA 13.0.2 or later | Turing or later | 7.5 and up | With CUDA 13.0, support for GPU architectures prior to Turing is deprecated. |
Recommendations
On Windows, the operating system may terminate the GPU driver to prevent an OS crash when the GPU is busy, the GPU computation load is high, or rendering takes more than a few seconds. This can cause the thermal solver to crash. To help prevent these crashes, we recommend:
- Extending the Timeout Detection and Recovery (TDR) delay in the Windows registry to specify how long Windows waits before terminating the GPU driver.
- Dedicating one GPU to OS rendering and the other GPU to the software if your machine has two GPUs.
