Tag AMD Instinct MI350P

AI & Compute

Qwen 3.8 27B Needs Two RTX 5090s for Full Context

Three horizontal bars drawn to scale from one shared axis: two short gray bars for the 32,768 and 37,740 tokens of context a single RTX 5090 reached under the published vLLM and SGLang recipes, against a blue bar roughly eight times longer for Qwen 3.8 27B's full 262,144-token window.

Weights that fit in 32GB still left the RTX 5090 waiting half an hour for a first token, and the fix was a different inference engine rather than a different card.

Saya
September 9, 2026