If you run a deep learning workstation with two monitors and a single RTX A6000, stay with WDDM for convenience, but be ready for occasional TDR frustrations. If you manage a 8x A100 server in a rack, switch to TCC immediately and never look back.
TCC is the only mode supported for GPU pass-through in Hyper-V (DDA – Discrete Device Assignment) and for Windows Server running HPC workloads. tcc wddm
: The user gets maximum rendering performance and more usable VRAM headroom, while the GTX 1660 keeps the UI responsive. Key Differences at a Glance If you run a deep learning workstation with
nvidia-smi -g 0 -dm 0 # 0 = TCC, 1 = WDDM : The user gets maximum rendering performance and
For further reading, consult the official NVIDIA documentation: “WDDM vs TCC Mode for NVIDIA GPUs” (NVIDIA Developer Blog) and the CUDA Toolkit documentation for cudaSetDeviceFlags .