Question 1

Can I run NemoClaw on WSL1 instead of WSL2?

Accepted Answer

No. WSL1 does not provide a real Linux kernel — it translates Linux syscalls to Windows NT calls. NemoClaw requires kernel features (Landlock, seccomp, namespaces, cgroups v2) that only exist in a real Linux kernel. WSL2 runs an actual Linux kernel in a lightweight VM, which is why it works at all. WSL1 will fail immediately at the onboarding stage.

Question 2

Will NVIDIA officially support WSL2?

Accepted Answer

GitHub #305 is the tracking issue, and NVIDIA has not committed to official WSL2 support. The alpha release targets native Linux. Community workarounds exist (this guide documents them), but you should not expect NVIDIA engineering resources dedicated to WSL2-specific issues during the alpha phase. Plan your production deployment on native Linux regardless.

Question 3

Should I run Ollama on Windows or inside WSL2?

Accepted Answer

Running Ollama inside WSL2 eliminates the cross-VM networking problem entirely. The tradeoff: GPU access from within WSL2 may have slightly higher overhead than native Windows Ollama, and you manage Ollama within the Linux environment rather than as a Windows service. For NemoClaw evaluation, running Ollama inside WSL2 is simpler. For production inference, run Ollama on a dedicated Linux server with native GPU access.

Question 4

How much RAM do I need for NemoClaw on WSL2?

Accepted Answer

16 GB minimum. WSL2 itself consumes 2-4 GB, Docker Desktop adds another 2 GB of overhead, and NemoClaw requires 5-6 GB for the gateway, sandbox, and policy engine. With Ollama running a 7B parameter model, total memory usage reaches 12-14 GB. Set your .wslconfig to allocate at least 12 GB to WSL2 and configure 8 GB of swap as a safety net. See our Troubleshooting Guide for the swap configuration.

Tag: WSL2

NemoClaw on WSL2: The Complete Workaround Guide