llamaperf

Qwen3.8 27B

on NVIDIA RTX 5090 · MegaCapybara · 800,000 ctx

Tone: positiveNVIDIA hardware
Oct 3, 2026
Throughput
500.0 t/s gen

Use cases

codingagenticlong-context

Summary

User announces MegaCapybara, a purpose-built engine for RTX 5090 focused on Qwen3.8 27B. Claims roughly twice the decode speed of Ninfer, reaching up to 650 t/s in small bursts and 2600 t/s with 12 concurrent agent slots at 800k context. Setup includes custom kernels, Dflash2 with DSpark headers, draft trees, and smart VRAM-RAM-DISC cache management.