Scaling beyond 4 RTX 6000 MAXQs: adding 300-400GB VRAM?
Explore multi‑GPU scaling options and evaluate VRAM expansion feasibility.
Research NVLink or PCIe bandwidth limits for 4+ RTX 6000 MAXQs.
Summary
A user is exploring how to scale beyond four RTX 6000 MAXQ GPUs and is considering adding 300–400 GB of VRAM. The post asks whether such an expansion is practical and how it could be achieved. No concrete solution is offered, but the question highlights the challenges of multi‑GPU scaling for large‑model inference. The discussion is limited to hardware considerations and does not touch on software or driver issues.
The key points are: the desire to add 300–400 GB VRAM, the plan to scale beyond four GPUs, the question of practicality, and the lack of a definitive answer.
The article serves as a starting point for anyone looking to push the limits of GPU memory for LLM workloads.
Key changes
- User wants to add 300–400 GB VRAM beyond four RTX 6000 MAXQs
- User is considering scaling beyond four GPUs
- The practicality of adding such VRAM is questioned
- No solution is provided in the post