⚡ Performance and Efficiency Benchmarks

This section reports the performance of Qwen 3 on NPU with FastFlowLM (FLM).

Note:

  • Results are based on FastFlowLM v0.9.31.
  • Under FLM’s default NPU power mode (Performance)
  • Newer versions may deliver improved performance.
  • Fine-tuned models show performance comparable to their base models.

Test System 1:

AMD Ryzen™ AI 7 350 (Kraken Point) with 32 GB DRAM; performance is comparable to other Kraken Point systems.


🚀 Decoding Speed (TPS, or Tokens per Second, starting @ different context lengths)

Model HW 1k 2k 4k 8k 16k 32k
Qwen 3 0.6B NPU (FLM) 66.5 57.5 44.5 31.0 19.6 14.1
Qwen 3 1.7B NPU (FLM) 40.2 35.8 30.8 23.7 16.4 12.5
Qwen 3 4B NPU (FLM) 19.6 18.1 16.3 13.7 10.6 8.5
Qwen 3 8B NPU (FLM) 11.9 11.5 11.1 10.4 8.7 7.2

🚀 Prefill Speed (TPS, or Tokens per Second, with different prompt lengths)

Model HW 1k 2k 4k 8k 16k 32k
Qwen 3 0.6B NPU (FLM) 1494 2003 2165 1981 1485 907
Qwen 3 1.7B NPU (FLM) 956 1263 1434 1411 1143 768
Qwen 3 4B NPU (FLM) 509 582 615 576 448 303
Qwen 3 8B NPU (FLM) 357 435 457 442 367 260

🚀 Prefill TTFT with Image Input (Seconds)

Prefill time-to-first-token (TTFT) for Qwen3-VL-4B on NPU (FastFlowLM) with different image resolutions.

Mid Resolution Images:

Model HW 720p (1280×720) 1080p (1920×1080)
Qwen3-VL-4B NPU (FLM) 3.3 7.4

High Resolution Images:

Model HW 2K (2560×1440) 4K (3840×2160)
Qwen3-VL-4B NPU (FLM) 13.7 41.2

This test uses a short prompt: “Describe this image.”