FreeToken Engine: Run 35B Models at 39.3 t/s on 8GB GPUs
rohanpaul_ai · x · 2026-08-29
Researchers from Berkeley and UT Austin released FreeToken, a new engine designed to efficiently run large open models on consumer hardware by optimizing expert caching.
- Performance: An 8GB gaming laptop can run a 35B model at 39.3 tokens per second, outpacing production Codex speeds.
- Key Innovation: Unlike static engines, FreeToken allows the GPU cache to dynamically follow the router's requests, reducing missed experts from 62% (in llama.cpp) to 16%.
- Workflow: It dynamically splits remaining experts between the GPU and system RAM based on real-time PCIe and memory speeds, ensuring neither side idles while waiting for the other.
Related event: FreeToken Lets 8GB GPUs Run 35B MoE Models On-Device(2 posts)→
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