Project Maya runs DeepSeek-V4.1-Flash fully in 120.6 GB on a 128 GB Ryzen AI Max
DotaMate · reddit · 2026-10-23
Project Maya v1.1.22 (Maya-XS) fits DeepSeek-V4.1-Flash entirely in a 128 GB Ryzen AI Max PC: all 384 experts resident in GPU memory at 120.6 GB, tensor-parallel across multiple GPUs, with much faster drafting on a single GPU. A notable open-source option for running large MoE models locally.
More from Infra
- AI stocks fall on strong earnings; the hard part is guessing supply, not demand — TiernanRayTech · 2026-10-23
- Carrier to invest over $600 million in new plant for data-center cooling equipment — pstAsiatech · 2026-10-23
- Besi beats order forecasts, says ASML is not a hybrid bonding rival — pstAsiatech · 2026-10-23
- Huawei secures $2.3 billion offshore loan as chip spending soars — pstAsiatech · 2026-10-23
- MiniMax 3.1 Flash runs locally on MLX as quant releases become the norm — pcuenq · 2026-10-23
- Boom Supersonic CEO: most data center water use cools turbines, and his new turbine needs zero — BrentM_BuildGuy · 2026-10-23