⚡ Performance and Efficiency Benchmarks

This section reports the performance of Qwen 3.5 on NPU with FastFlowLM (FLM).

Note:

  • Results are based on FastFlowLM v0.9.38.
  • 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
Qwen3.5-0.8B NPU (FLM) 44.83 43.84 40.3 36.5 30.21 22.8
Qwen3.5-2B NPU (FLM) 28.92 28.33 27.18 25.34 22.3 17.98
Qwen3.5-4B NPU (FLM) 15.9 15.6 15.02 14.03 12.35 9.98
Qwen3.5-9B NPU (FLM) 9.9 9.8 9.56 9.15 8.41 7.24

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

Model HW 1k 2k 4k 8k 16k 32k
Qwen3.5-0.8B NPU (FLM) 851.21 1285.09 1744.47 2007.97 2084.64 1912.63
Qwen3.5-2B NPU (FLM) 625.99 929.93 1198.04 1367.84 1443.05 1381.91
Qwen3.5-4B NPU (FLM) 322.37 445.63 543.15 599.95 616.35 571.6
Qwen3.5-9B NPU (FLM) 248.76 328.1 391.6 420.5 436.25 416.65

🚀 Prefill TTFT with Image Input (Seconds)

Prefill time-to-first-token (TTFT) for Qwen3.5-4B on NPU (FastFlowLM) with different image resolutions.

Mid Resolution Images:

Model HW 720p (1280×720) 1080p (1920×1080)
Qwen3.5-0.8B NPU (FLM) 1.3 2.6
Qwen3.5-2B NPU (FLM) 2.1 4.6
Qwen3.5-4B NPU (FLM) 3.4 6.8
Qwen3.5-9B NPU (FLM) 4.4 9.4

High Resolution Images:

Model HW 2K (2560×1440) 4K (3840×2160)
Qwen3.5-0.8B NPU (FLM) 5.0 13.3
Qwen3.5-2B NPU (FLM) 9.1 29.1
Qwen3.5-4B NPU (FLM) 12.6 36.1
Qwen3.5-9B NPU (FLM) 16.2 47.3

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