Prefill and token generation speed
2× RTX 3090 48GB (NVLink) with Qwen3.6 35B-A3B (MoE): prefill and generation speed
Last updated
1872 GB/s combined
Active 3B / Total 35B
79.2 ~ 248.6 tok/sDecode speed
Model fully loaded (21.28GB / 45GB)
First token (TTFT)
0.63 to 1.4 secondsContext 4,096 input tokens included
Prefill speed
2,863 ~ 6,573 tok/s0.62 to 1.4 seconds (Estimation based on public benchmarks)
Total time (E2E)
Approximately 1.1 to 2.9 seconds117 output tokens included
Press Play to process the input and generate the answer.
Ready
Expand settings Settings4-bit · 4K · Baseline decodeCollapse
QuantizationNeed 21.28GB
Input context4K tokens
Baseline
Baseline
Sample question:
117 output tokensCheck configuration
2× RTX 3090 48GB (NVLink) · Memory 48GB
Qwen3.6 35B-A3B (MoE) Q4_K_M configuration. Check that the retailer lists the same memory option.
Already own it? Open the setup guideWould another device feel faster?
Compare the same model including prefill, or just token generation.
What changes with a longer document?Expand input length, waiting time, speed and memory
Detailed analysis and all 60 hardware configurations
Popular configurations
Speed simulations by device and model
Mac mini M4 Pro 64GB × Qwen3.6 35B-A3B (MoE)
19.6 ~ 46.1 tok/s · 2.7 ~ 6.3 s
Studio M4 Max 128GB × Qwen3.6 35B-A3B (MoE)
39.2 ~ 81.9 tok/s · 1.5 ~ 3.1 s
DGX Spark 128GB × Qwen3.6 35B-A3B (MoE)
61.2 ~ 111.1 tok/s · 1.5 to 3.1 seconds
Gemma 4 12B on Mac mini M4 24GB: chat speed
11.6 ~ 12.9 tok/s · 12.8 ~ 31.1 s