Qwen NVFP4 Stress Tested on Four 5060Ti GPUs
joorklee · reddit · 2026-07-11
This is an inference stress test log for unsloth/Qwen3.6-27B-NVFP4, conducted on 4x 5060 Ti GPUs using vLLM and pipeline parallel size=4.
Test Configuration
- Concurrency: 1, 4, 8, 12, 16
- Input length: 6144 tokens
- Output length: 512 tokens
- Max batch tokens: 16384
- Enabled options like prefix caching, chunked prefill, and async scheduling
Purpose
The author primarily wants to observe how prefill / TTFT degrades as concurrency increases in a 4x 5060 Ti setup, using benchmark results to evaluate the actual throughput and latency of the new NVFP4 quantization.
Related event: Unsloth Brings Faster NVFP4 Quantization to Qwen3.6(3 posts)→
More from Infra
- RTX 5080 price jumps from $1658 to $2318 in ten days on Amazon — Frizzy-MacDrizzle · 2026-10-27
- ComfyUI on 12GB VRAM: measured peaks show offloading turns 24s into 11 minutes — Impossible-Mail-4619 · 2026-10-27
- Neon rebuilds its docs search entirely on its own stack: Lakebase Search, Functions, AI Gateway — matei_zaharia · 2026-10-27
- PyTorch details 2.12x faster on-device streaming ASR on Arm SME2 via ExecuTorch — PyTorch · 2026-10-27
- NVIDIA shows speculative decoding speeding up LLM inference on Jetson with Qwen3.5 9B — NVIDIA Developer · 2026-10-27
- Revenue per GPU can rise as token prices fall, argues Gavin Baker — GavinSBaker · 2026-10-27