CUDA version on a rented machine
Can I install whatever CUDA version I want on the rented machine?
- Views
- 2
- Replies
- 10
- Created
- 2026-09-01 09:45
- Last reply
- 2026-09-05 21:16
You have full SSH access to the dedicated machine, but 'whatever version' is too broad.
The CUDA/driver combination still has to be compatible with the GB10 platform and the host driver environment.
Jose
Staff · The Archivist
Edited 2026-09-23 11:28:50
Posted 2026-09-01 11:48
Edited 2026-09-23 11:28:50
Well.. it's not that simple but Yes. You get root (sudo) in your env. so you can install any CUDA toolkit. apt, conda, pip.
You won't be able to change the GPU driver though. We stick to driver 580 for now, seen some pretty bad situations with newer more unstable drivers. This is the latest stable by Nvidia for now. We have support for Cuda 13.x so it should be fine for most work.
How much context does Qwen3.8 have on AxForge?
The current AxForge serverless pricing/docs page lists 131,072 tokens for Qwen3.8 27B.
Do not assume the upstream model's maximum context is the same as the hosted service limit; the service limit is what matters for API calls.
Can I use my existing OpenAI SDK with AxForge?
Yes. The AxForge serverless API is OpenAI-compatible.
Typically you change the base URL to api.axforge.ai/v1, use your AxForge API key and select the AxForge model name.
Jose
Staff · The Archivist
2026-09-05 21:16
The site also lists streaming, tool calls and usage reporting on the OpenAI-compatible API shape.
Sign in to reply.
Posting guidelines
- Be helpful and respectful
- Search before you post
- Use a clear title and tags
- Add code or a screenshot when it helps
- Keep to the topic
Accounts that break these can lose forum access — paid plans included. Read the full guidelines →
Related topics
- GPU for AI agents 7 replies · 2 views
- Models on the API right now 3 replies · 2 views
- Lots of concurrent API users 4 replies · 2 views
- Bringing my own model 6 replies · 1 views