boltbeam.fyi

Qwen3-14B AWQ on RTX 4090

awq INT4_G128_ZP+BF16

Engines

vLLM

0.31.0 ↗measured 2026-11-09 19:32 UTCnewer release at the time: 0.33.0

97.7 tok/s

86% of the roofline113.0 tok/s possible

now, a token
10.24 ms
floor
8.85 ms
read, a token
8.52 GB
More
memory, measured
962 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 paired

Speed at context

128 tokens
vLLM97.8 tok/s
512 tokens
vLLM97.7 tok/s of 113.0
Measured at short contexts only. The pink tick is the roofline at that context.

Where the time goes

Now, a token
10.24 ms
Roofline floor, a token
8.85 ms
Gap to the floor
1.39 ms
Kernels a token
460

Per role ms a token. floor (bytes ÷ measured memory speed) above the floor

attention0.30 / 0.09 ms 31%
ffn_gate_up INT4_G128_ZP4.05 / 3.85 ms 95%
attn_qo INT4_G128_ZP0.73 / 0.57 ms 77%
ffn_down INT4_G128_ZP2.09 / 1.93 ms 92%
attn_kv + attn_qo INT4_G128_ZP0.94 / 0.79 ms 84%
lm_head F161.63 / 1.62 ms 100%
flash::flash_fwd_splitkv_combine_kernel0.13 ms · no floor
triton_red_fused_fused_add_rms_norm_20.09 ms · no floor
kernel gaps0.08 ms · no floor
triton_red_fused_fused_add_rms_norm_00.08 ms · no floor
vllm::reshape_and_cache_flash_kernel0.07 ms · no floor
20 other small kernels0.17 ms · no floor
Per kernel (27): ms now / ms at the roofline, launches a token
Marlin7.82 / 7.14 ms ×160
gemm · lm_head F161.63 / 1.62 ms ×1
flash_fwd_splitkv_kernel0.30 / 0.09 ms ×40
flash::flash_fwd_splitkv_combine_kernel0.13 ms · no floor ×40
triton_red_fused_fused_add_rms_norm_20.09 ms · no floor ×40
triton_red_fused_fused_add_rms_norm_00.08 ms · no floor ×40
vllm::reshape_and_cache_flash_kernel0.07 ms · no floor ×40
triton_red_fused_30.05 ms · no floor ×39
triton_poi_fused_mul_silu_slice_10.05 ms · no floor ×40
_gumbel_sample_kernel0.02 ms · no floor ×1
at::native::index_elementwise_kernel0.01 ms · no floor ×3
_bias_kernel0.01 ms · no floor ×1
_gather_block_tables_kernel0.00 ms · no floor ×1
triton_red_fused__to_copy_embedding_rms…0.00 ms · no floor ×1
at::native::reduce_kernel0.00 ms · no floor ×1
_post_update_kernel0.00 ms · no floor ×1
_combine_sampled_and_draft_tokens_kernel0.00 ms · no floor ×1
_get_num_sampled_and_rejected_kernel0.00 ms · no floor ×1
at::native::vectorized_elementwise_kern…0.00 ms · no floor ×3
_compute_slot_mappings_kernel0.00 ms · no floor ×1
at::native::unrolled_elementwise_kernel0.00 ms · no floor ×1
at::native::_scatter_gather_elementwise…0.00 ms · no floor ×1
triton_red_fused_10.00 ms · no floor ×1
_prepare_pos_seq_lens_kernel0.00 ms · no floor ×1
<unnamed>::elementwise_kernel_with_index0.00 ms · no floor ×1
_apply_write_kernel0.00 ms · no floor ×0
concurrent kernel overlap (subtract)-0.01 / -0.01 ms ×0

No floor = the kernel reads no weights (norms, conversions, copies): all of its time is gap.

Run qwen3-14b-awq-nvidia_geforce_rtx_4090_24g-001

Tested on

GPU
RTX 4090 · 24 GB
System RAM
189 GB
CPU threads
96
Memory, measured
962 GB/s
Temp, max
59 °C
Power
85 W mean of 450 W
SM clock
2582 MHz
Driver
595.58.03
Measured
2026-11-09 19:32 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