Single Intel B65 with XeStrata Flash-Next 4-bit hits 60 tok/s on Qwen3.8-Flash-Next
Original title: Intel 1x B65 with xeStrata Flash-Next 4-bit - 60tok/s - What a game changer for single card users
The author benchmarked a single Intel B65 (32GB) running Qwen3.8-Flash-Next with XeStrata Flash-Next 4-bit, offloading expert weights to host RAM, reaching a median 62.4 tok/s with 2.1s TTFT for a single user. Throughput drops to 48.7 tok/s aggregate at 2 concurrent users and 44.6 at 4. Compared against an RTX A6000 with llama.cpp MTP-2 (52.8 tok/s single-user, 75.2 at 2 users), the author argues the Intel setup is competitively priced and worth a second look for entry-level builds. Full data and interactive charts: https://b65.thekaiser.us/by-model/qwen38-flash-next/xestrata-20261030/
First-hand benchmarks of a single Intel B65 with XeStrata running Qwen3.8-Flash-Next, with throughput and concurrency data comparable to an A6000 setup.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.