Hardware compatibility · checked 2026-08-26
NVIDIA RTX 5090 desktop workstation
Best choice for CUDA-heavy local AI and generation workflows when power, heat, and desktop size are acceptable.
NVIDIA / system builder · discrete-GPU desktop
“Compatible” means the documented operating system/runtime path is plausible or official. It does not guarantee a particular model, speed, driver, thermals, or local inference quality. Always test the exact model and version before buying for a production workload.
What the official source says
NVIDIA lists 32GB GDDR7, an 850W minimum power recommendation, three-slot physical clearance, and a 304mm card length for the reference design. Actual add-in-card specifications vary.
Local-AI fit
Best mainstream single-GPU option here for CUDA-first local inference, image/video generation, and applications that benefit from dedicated VRAM. The 32GB VRAM ceiling still constrains unquantized and very large models.
Specifications that matter
- memory: 32GB dedicated VRAM; system RAM of 64GB is a sensible workstation target for local AI, containers, datasets, and browser/agent overhead.
- storage: Use a fast NVMe SSD; 2TB is a practical starting point for multiple model families and containers.
- os: Windows and Linux are common deployment targets; verify driver, CUDA, WSL2, and application support for the chosen stack.
- thermal: Requires a properly ventilated case, substantial PSU, and adequate cooling. Check exact card dimensions, connector clearance, and sustained-load noise.
Service compatibility matrix
| Service | Compatibility interpretation |
|---|---|
| Hermes Agent | good on Linux/Windows paths when terminal, Docker, or WSL boundaries are configured; GPU use depends on the model backend |
| OpenClaw | gateway itself is lightweight; RTX workstation is useful when paired with a local model server |
| Ollama | strong fit for NVIDIA GPU workloads when current driver/runtime support is installed |
| LM Studio | official Windows guidance recommends at least 4GB dedicated VRAM; 32GB provides substantially more headroom |
| Open WebUI | good UI host paired with Ollama/compatible GPU backends; Docker CUDA image is documented |
Where to purchase
These are official manufacturer or official marketplace links. Prices, regional availability, configurations, and delivery dates can change.
Official evidence
Unresolved claims and re-checks
- NVIDIA's official page did not expose a reliable current street price in this check.
- Partner cards differ in dimensions, cooling, power connectors, and warranty.
- Application-specific CUDA/driver support should be checked before purchase.