2× RTX 3090
187tok/sEveryday
1 request, prose prompt, prompt 450 tokens, no speculative decoding Source
Intelligenceno independent value
Local AI · Recipe · 2× RTX 3090
MiniCPM-V-4.7 35B-A3B INT4 with vLLM on 2× RTX 3090
MiniCPM-V-4.7 35B-A3B INT4 with vLLM on 2× RTX 3090: 199 tok/s according to github.com (dataset as of Nov 3, 2026).
by tonyd2wild
Intelligence (original model) · Artificial Analysisno independent value
Engine
- Engine
- vLLM 0.31.1rc1.dev296 (nightly fe4df3d, torchcodec removed)
- Quantization
- INT4 (compressed-tensors, OpenBMB) + DSpark draft BF16
- Model family
- MiniCPM-V-4.7
- Context
- 262,144
- Parameters
- 35B-A3B
- Creator
- tonyd2wild
- GitHub stars
- 2
- Repo updated
- Oct 13, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Prefill by context
Time to first token
Shorter is better.
What you need
- Hardware
- 2 × NVIDIA RTX 3090
- Weights
- Main weightsopenbmb/MiniCPM-V-4.7-35B-A3B-int4 Apache-2.0
- Baseopenbmb/MiniCPM-V-4.7-35B-A3B Apache-2.0
- Draft modelopenbmb/MiniCPM-V-4.7-35B-A3B-DSpark Apache-2.0
- Engine
- vLLM 0.31.1rc1.dev296 (nightly fe4df3d, torchcodec removed)
- Context
- 262,144 tokens
Notes
What matters before you rebuild it.
- vision
- tool-calling
Sources
- github.com/tonyd2wild/MiniCPM-V-4.7-35B-A3B-2x-RTX-3090#draft-length-7-beats-the-model-cards-15
- github.com/tonyd2wild/MiniCPM-V-4.7-35B-A3B-2x-RTX-3090#concurrency-sweep-with-the-draft-best-case-json
- github.com/tonyd2wild/MiniCPM-V-4.7-35B-A3B-2x-RTX-3090/blob/main/results/speed/prefill_len_minicpm47-2x3090-tp2-nvl-dspark7.json
- github.com/tonyd2wild/MiniCPM-V-4.7-35B-A3B-2x-RTX-3090#dspark-speculative-decoding
- github.com/tonyd2wild/MiniCPM-V-4.7-35B-A3B-2x-RTX-3090/blob/main/README.md
Related recipes
2× RTX 3090
179tok/sEveryday
1 request, prose prompt, no speculative decoding Source
Intelligence18with thinkingno thinking 15
- License unclear
2× RTX 3090
151tok/sEveryday
1 request, prose prompt, no speculative decoding Source
Intelligenceno independent value
2× RTX 3090
144tok/sEveryday
1 request, prose prompt, no speculative decoding Source
Intelligenceno independent value
2× RTX 3090
123tok/sEveryday
1 request, prose prompt, with Draft Source
Intelligence34with thinkingno thinking 20
2× RTX 3090
109tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence21with thinkingno thinking 20
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.