Qwen3.6 27B
Qwen/Qwen3.6-27B
Qwen · 2 recorded configurations
Measured setup
Hardware
Runtime and tool
Quantisation
MTP draft depth
Every option opens a recorded setup. Where the corpus recorded nothing at this page's other settings, the option names the fields it switches with it. Allocation and runtime detail stay in the technical record below.
2× RTX 3090 · llama.cpp
Q4_K_M · MTP off
PP · left axisTG · right axis
View exact measurements
| Context depth (tokens) | Fresh prefill PP (tok/s) | TG (tok/s) |
|---|---|---|
| 0 | 1,580.00 | 59.05 |
| 2,048 | 1,529.10 | 58.71 |
| 4,096 | 1,508.60 | 58.11 |
| 8,192 | 1,465.60 | 56.99 |
| 16,384 | 1,387.40 | 55.04 |
| 32,768 | 1,251.80 | 51.25 |
| 65,536 | 1,046.60 | 45.08 |
| 131,072 | 774.10 | 35.86 |
| 262,144 | 498.10 | 25.41 |
Concurrent-stream measurements
| Streams | Context depth (tokens) | Aggregate PP (tok/s) | Aggregate TG (tok/s) | TG / stream (tok/s) | Aggregate all (tok/s) |
|---|---|---|---|---|---|
| 1 | 0 | 1,580.00 | 59.05 | — | — |
| 1 | 0 | 1,483.50 | 59.22 | 59.22 | 164.53 |
| 2 | 0 | 1,607.40 | 98.37 | 49.19 | 262.93 |
| 4 | 0 | 1,658.20 | 128.72 | 32.18 | 334.28 |
Technical record and provenance
Run details
- Source run ID
- rtx3090x2-qwen3.6-27b-q4km-2026-08
- Hardware
- 2× RTX 3090
- Runtime
- llamaCpp · llama-bench fc6545d (build 50)
- Weight placement
- gpu-resident
- Split ratio
- 1:1
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Batch size
- 2,048
- Microbatch size
- 2,048
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Measured repetitions
- 3
- Notes
- Q4_K_M, batch 2048/2048, all layers on GPU, tensor split 1:1. MTP-capable GGUF with speculative decoding disabled (llama-bench does not exercise the draft path). Extended sweep to 64K/128K/256K. Concurrency rows from llama-batched-bench (aggregate rates).
- Source article
- /local-ai/rtx-3090-for-local-ai-inference/