Qwen3-14B AWQ on RTX 3090
awq INT4_G128_ZP+BF16
Engines
vLLM
0.31.0 ↗measured 2026-11-09 17:40 UTCnewer release at the time: 0.33.0
82.4 tok/s
78% of the roofline105.4 tok/s possible
- now, a token
- 12.13 ms
- floor
- 9.49 ms
- read, a token
- 8.52 GB
More
- memory, measured
- 898 GB/s
What was run
python vllm_decode_drive.py --model Qwen3-14B-AWQ --contexts 512 --batches 1 --tokens 64 --reps 1 --once 64 --max-num-batched-tokens 584 --model-context-limit 584 --capture-pass pairedSpeed at context
128 tokens
512 tokens
Where the time goes
- Now, a token
- 12.13 ms
- Roofline floor, a token
- 9.49 ms
- Gap to the floor
- 2.65 ms
- Kernels a token
- 460
Per role ms a token. floor (bytes ÷ measured memory speed) above the floor
Per kernel (27): ms now / ms at the roofline, launches a token
No floor = the kernel reads no weights (norms, conversions, copies): all of its time is gap.
Run qwen3-14b-awq-nvidia_geforce_rtx_3090_24g-001
Tested on
- GPU
- RTX 3090 · 24 GB
- System RAM
- 252 GB
- CPU threads
- 80
- Memory, measured
- 898 GB/s
- Temp, max
- 56 °C
- Power
- 134 W mean of 280 W
- SM clock
- 1746 MHz, throttled
- Driver
- 595.71.05
- Measured
- 2026-11-09 17:40 UTC · context 512 tokens · measured, in-model (nsys): split by role from the engine's own labels, 5 weight roles
Qwen3-14B on other GPUs
H200data center166.9 tok/s · 30%RTX 5090consumer151.5 tok/s · 76%H100 80GBdata center147.6 tok/s · 39%H200 NVLdata center135.0 tok/s · 26%RTX 4090consumer97.7 tok/s · 86%RTX 3090consumer82.4 tok/s · 78%L40Sdata center77.7 tok/s · 87%
Best engine and quant on each GPU, one user. The percentage is that GPU's own roofline.