llamaperf

Qwen3.8 125B (6B active) Flash-Next

on GTX 1070 8GB · Strata

Nov 14, 2026
Throughput
13.0 t/s gen
System RAM
64 GB
VRAM reported
8 GB

Summary

User reports Qwen3.8 Flash-Next 125B at 13 t/s output on an HP Z640 with an NVIDIA GTX 1070 8GB, 64GB DDR4 and a 4-core Xeon. Setup is Strata, with models loaded from an NVMe drive over a PCIe adapter; the 125B model cannot fit in 8GB of VRAM, so weights run from system memory or disk. DeepSeek V4.1 Flash runs at about 1.5 t/s output on the same machine. User asks whether a Tesla P100 16GB or an M40 24GB would be the better second card.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

Qwen3.8 125B · 6B active at 4-bit by hardware

Bars are medians on one scale, 0 to 232.0 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Every Qwen3.8 report →