On 64GB combined memory, is there anything better than Qwen 3.8 27B yet?
xdcfret1 · reddit · 2026-11-05
A Reddit user with 32GB VRAM + 32GB RAM asks whether anything meaningfully beats Qwen 3.8 27B at Q6 quantization on their hardware. They can also run larger MoE models by offloading experts to CPU, as long as the total footprint stays within 64GB. Despite many recent releases, they haven't found a meaningful upgrade within these limits and ask for community recommendations and real-world results. The attached image is AI-generated (noted by the author).
More from Infra
- oMLX lets a second Mac boost prefill up to 2.9x on big-memory Macs — pcuenq · 2026-11-05
- FineWeb-30M speedrun record falls to 14.2s on one 8xH100 node, a 1.64x speedup — RobertTLange · 2026-11-05
- HydraDB: a Rust graph database that keeps state in S3 and treats query workers as disposable compute — JeremyCMorgan · 2026-11-05
- OpenSSH 10.6p2 released with a workaround for a critical vulnerability — Animux · 2026-11-05
- Microsoft rejects multibillion-dollar AI deal with UK's Nscale, FT reports — nordicinst · 2026-11-05
- Microsoft Details Running Coding AI Locally on 128GB RTX Spark Windows PCs Without Token Fees — lee_stott · 2026-11-05