AI

Have 8 GB of VRAM and 32 GB of RAM? You can run Qwen 3.8 Flash Next at up to 900 t/s PP and 30 t/s TG!

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 26/10/2026, 15:47:25 UTC.

Briefings

A user achieved 900 t/s prompt processing speed and 30 t/s token generation speed by running Qwen 3.8 Flash Next on an RTX 5060 with 8 GB VRAM and 32 GB RAM using Strata. The setup involved specific parameter tuning like --vram-reserve-mib 500 and --vram-reserve-mib 900, and running Vision on the CPU. Results were compared against Qwen 3.6 35B A3B in llama.cpp, showing superior performance in benchmarks and real-world tests.

Key points

• Qwen 3.8 Flash Next achieved 900 t/s prompt processing speed and 30 t/s token generation speed on an RTX 5060 with 8 GB VRAM.
• Strata distributed workloads across VRAM, RAM, and SSD to enable running a 125B-parameter model with minimal VRAM requirements.

Context

The article details a successful benchmark run on Linux hardware using Strata to distribute model parameters. The user compared these results against previous Qwen 3.6 35B A3B fine-tunes and current setups, highlighting the effectiveness of the new configuration for agentic tasks.

Limitations

Performance data is specific to the tested hardware (RTX 5060, AMD Ryzen 9 8945HX) and Linux OS. Results may vary with different hardware configurations, quantization levels, or system settings. The article does not provide a direct link to the source material.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 26/10/2026, 16:46:42 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibxzqauk5mardtyipumcsiniikajlg4xb6tp5x3i3era5jseiehrq