Qwen3.8-Flash-Next Architecture: Estimated Local Memory ~82GB
pmv143 · reddit · 2026-08-26
A Reddit user analyzes the unreleased Qwen3.8-Flash-Next architecture (125B-A6B + 51B n-gram). Ideal 4-bit quantization requires 82GB VRAM (58GB main weights + 24GB n-gram tables). Due to sparse access, the n-gram table is an excellent candidate for system RAM offload, potentially making it very local-friendly upon release.
More from Infra
- Next-gen GraalVM benchmarks: 4x faster startup, 2x less memory, no build step needed — RealGeneKim · 2026-10-15
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark AI laptops — tomwarren · 2026-10-15
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark laptops — The Verge AI · 2026-10-15
- Global smartphone shipments plunge 22% in worst quarterly decline on record as AI demand fuels memory costs — Polymarket · 2026-10-15
- SK Hynix drops 11% again, now below 3x FY2027 earnings as AI memory selloff deepens — JOBhakdi · 2026-10-15
- ASML Sells Zero Lithography Machines into Europe as US Hosts 75%+ of AI Compute — NinaDSchick · 2026-10-15