AI Workbench Support Matrix#
Overview#
- Use this reference to check system requirements and support for AI Workbench.
Find information on system requirements, e.g. storage and RAM. Find information on what’s expected to work vs what isn’t validated.
- Support levels define the degree of validation and testing.
Supported vs Not-supported depends on what we explicitly test in QA. Not supported configurations are untested or have known issues.
- AI Workbench has no inherent dependencies on GPUs, drivers, or CUDA versions.
The primary dependency is on the target operating system.
Key Concepts#
- Supported
Configurations that are validated and tested during QA by NVIDIA. These platforms receive full support and regular testing.
- Expected to Work
Configurations that should function correctly but are not rigorously tested. These platforms are compatible but may not receive extensive validation.
- Not Supported
Configurations that are untested or have known compatibility issues. Use these platforms at your own risk.
Desktop App Support and Requirements#
The Desktop App is supported on all three major operating systems.
All builds of Windows 11 (build 22631+)
Windows 10 22H2 (build 19045.4052+)
MacOS 14 (Sonoma) or higher, and
Ubuntu 22.04 and 24.04.
The Desktop App has the same system requirements on each of the three major operating systems.
500 MB disk space for the application
At least 8 GB of memory
No GPU required
No container runtime required
Full Local Install Support and Requirements#
Full local install is supported on the exact same operating systems as the Desktop App.
All builds of Windows 11 (build 22631+)
Windows 10 22H2 (build 19045.4052+)
MacOS 14 (Sonoma) or higher, and
Ubuntu 22.04 and 24.04.
- Software dependencies are standard but vary somewhat by operating system.
See the software dependency information here.
- System requirements for full local install are essentially the same on all three operating systems.
The main difference is in the storage requirements for Windows and MacOS due to the virtualization required to run and store containers.
Windows 10/11#
Windows has storage requirements that depend on WSL in addition to the container runtime.
Requirements |
Windows (Docker Desktop) |
Windows (Podman) |
|---|---|---|
Application Space |
AI Workbench: 500 MB WSL2 distro: 2.5 GB Docker Desktop: 2 GB Total: 5 GB |
AI Workbench: 500 MB WSL distro: 2.5 GB Podman: 500 MB GB Total: 3.5 GB |
Container Space Examples |
Python Basic container (2.5 GB): ~8 GB total CUDA container (8.6 GB): ~14 GB total PyTorch container (36 GB): ~41 GB total |
Python Basic container (1.5 GB): ~4 GB total CUDA container (5.2 GB): ~12 GB total PyTorch container (24.3 GB): ~46 GB total |
System RAM |
16 GB minimum, 32 GB recommended |
16 GB minimum, 32 GB recommended |
MacOS 14 or higher#
MacOS has storage requirements that depend on the container runtime.
Requirements |
MacOS (Docker Desktop) |
MacOS (Podman) |
|---|---|---|
Application Space |
AI Workbench: 500 MB Docker Desktop: 2 GB Total: 2.5 GB |
AI Workbench: 500 MB Podman: 2 GB Total: 2.5 GB |
Container Space Examples |
Python Basic container (2.5 GB): ~8 GB total CUDA container (8.6 GB): ~14 GB total PyTorch container (36 GB): ~41 GB total |
Python Basic container (2.5 GB): ~8 GB total CUDA container (8.6 GB): ~14 GB total PyTorch container (36 GB): ~41 GB total |
System RAM |
16 GB minimum, 32 GB recommended |
16 GB minimum, 32 GB recommended |
Ubuntu 22.04 or 24.04#
- Ubuntu uses native container runtimes without virtualization.
Container storage is more straightforward than Windows or macOS.
Requirements |
Ubuntu (Docker or Podman) |
|---|---|
Application Space |
AI Workbench: 500 MB Container Runtime: ~150 MB Total: ~650 MB |
Container Space |
Containers stored directly on filesystem Add container size to total (1:1 ratio) |
System RAM |
16 GB minimum, 32 GB recommended |
Remote Install Support and Requirements#
- Remote install is supported on Ubuntu 22.04 and 24.04.
These versions of Ubuntu are tested during QA.
- Remote install is expected to work on Jetpack 6.1.
This is not tested during QA but is know to work with some adjustments.
GPU Support#
- AI Workbench does not require a GPU to work.
There is no direct dependency on the type of GPU. The dependency is between the CUDA version in the container and the drivers on the host.
- vGPU support requires specific hypervisor, GPU, and driver configurations.
AI Workbench does not manage vGPU configuration. Refer to the NVIDIA vGPU Product Support Matrix for compatibility requirements.
CUDA Support#
- AI Workbench passes through GPU access from host to container.
GPU drivers are required for GPU access, but AI Workbench itself doesn’t depend on them. Container CUDA versions must be compatible with host GPU drivers.
- If a container cannot access the GPU, check driver compatibility.
Verify your host driver version supports the container’s CUDA version. Refer to NVIDIA’s driver-to-CUDA compatibility documentation. AI Workbench does not validate or manage this compatibility.