For NVIDIA GPU users on Windows, the choice between TCC (Tesla Compute Cluster) WDDM (Windows Display Driver Model)
nvidia-smi -Lnvidia-smi -i <GPU_ID> -dm 1 or nvidia-smi -i <GPU_ID> -gtt 0 (exact flags vary by driver).At its core, the choice is between a mode that shares your GPU with your screen and one that reserves it entirely for math. tcc wddm better
WDDM (Windows Display Driver Model): This is the default mode for almost all consumer GeForce GPUs. It is designed to handle the Windows desktop, 3D gaming, and user interface rendering. For NVIDIA GPU users on Windows, the choice
Efficient Memory Usage: TCC is optimized for headless rendering and AI training, allowing for better GPU memory utilization without the interference of desktop display requirements. WDDM vs. TCC Comparison WDDM (Windows Display Driver Model) TCC (Tesla Compute Cluster) Primary Use Desktop display, gaming, graphics AI, HPC, headless compute Graphics APIs Supports DirectX and OpenGL Disabled (no display output) Overhead High (commands are batched) Low (direct access) Hardware Supported on all NVIDIA GPUs Mostly restricted to Quadro/Tesla OS Priority High (OS manages resources) Low (GPU dedicated to task) Key Constraints and Considerations List GPUs: nvidia-smi -L Set TCC mode (example):
TCC (Timeline Compensation Clock) is a hardware clock mechanism in NVIDIA GPUs (starting with Turing architecture) designed for real-time, low-latency workloads.
In TCC mode, the card is "headless"—it has no display output. Therefore, no memory is reserved for rendering the Windows desktop. The entire frame buffer is available for your compute workload. In memory-bound tasks (like large matrix multiplications or 3D rendering), this extra overhead can be the difference between "Out of Memory" errors and a successful run.
This is the standard graphics architecture used by Windows since Vista. It handles all desktop rendering, window management, and 3D graphics. While it supports compute APIs like CUDA, it is subject to the Windows Watchdog Timer