Delta Inference Engine Achieves 41 tok/s for DeepSeek-V4.1-Flash-1017 on Single 3090 GPU

Original title: DeepSeek-V4.1-Flash-1017 at 41 tok/s on one 3090

The Delta Inference Engine (built on Xolotl and Strata) achieves local execution of 552B parameter DeepSeek-V4.1-Flash-1017 (2-bit experts) on consumer PCs. Benchmarks show writing speeds up to 41 tok/s and long prompt reading speeds up to 610 tok/s on an NVIDIA RTX 3090 with 128 GB+ of RAM.

The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.