GPU planning for RAG & agents
Size it before you buy it.
- The whole stack, not just the LLMLLM, embedders, rerankers, OCR, speech and guardrails: memory, speed and question-to-first-word time.
- The right number of GPUsSmallest count that fits across 93 GPUs, clusters with TP × PP, offloading.
- Saved, shared, comparedAutosaved plans, read-only links for your team, setup codes.
138 LLMs · 175 RAG & agent models · 93 GPUs