Models / Nanbeige4.3 33B
Best home rig for Nanbeige4.3 33B
Quick answer
How much VRAM does Nanbeige4.3 33B need?
The weights are about 20.2 GB at Q4_K_M (33B params × 4.89 bits ÷ 8, before context/KV cache). Add memory for context (KV cache) and the runtime. The smallest-memory setup with a measured result here, fully in memory, is 2x RTX 3090 (48 GB) at 66.2 tok/s.
What is the fastest measured hardware for Nanbeige4.3 33B?
2x RTX 3090: 66.2 tok/s generation (vLLM TP2, 1 stream, W4A16 AutoRound), from https://github.com/tonyd2wild/Nanbeige4.3-33B-W4A16-2x-RTX-3090.
Last updated:
Current model33B dense2 measured resultsfastest plain decode: 66.2 tok/s model info ↗
Will it fit my GPU? Best value GPU per euro →
Real, measured generation speeds for Nanbeige4.3 33B on home hardware, fastest first. Different quants/engines aren't directly comparable.
| Build | Quant | Engine | tok/s | Prompt processing (tok/s) | Cost | Source |
|---|---|---|---|---|---|---|
| 2x RTX 3090: Nanbeige4.3 33B W4A16 AutoRound on vLLM TP2, 62.2 tok/s prose (Tech2wild) | W4A16 AutoRound | vLLM TP2, 1 stream | 66.2 | – | github.com ↗ | |
| 2x RTX 3090: Nanbeige4.3 33B W4A16 AutoRound on vLLM TP2, 62.2 tok/s prose (Tech2wild) | W4A16 AutoRound | vLLM TP2, 1 stream | 62.2 | – | github.com ↗ |
Prompt processing = prefill speed reported by the source, @ prompt length. * est. TTFT @8k is calculated from measured prefill, not measured. Why it matters.
Memory math: how much VRAM / RAM does Nanbeige4.3 33B need?
Nanbeige4.3 33B has 33B parameters (source: Nanbeige model card: 33B dense Llama-architecture, Apache 2.0). Weight size = parameters × bits-per-weight ÷ 8, before context (KV cache) and runtime overhead.
| Format | Bits/weight | Weights | Smallest common memory size that holds the weights alone |
|---|---|---|---|
| Q8_0 | 8.50 | 35.1 GB | 48 GB |
| Q6_K | 6.56 | 27.1 GB | 32 GB |
| Q5_K_M | 5.70 | 23.5 GB | 24 GB |
| Q4_K_M | 4.89 | 20.2 GB | 24 GB |
| Q3_K_M | 4.00 | 16.5 GB | 24 GB |
| Q2_K | 3.16 | 13.0 GB | 16 GB |
Bits-per-weight for GGUF quants are llama.cpp's own figures (llama.cpp quantize README). "As measured" rows apply the source's stated bits-per-weight to the parameter count; real files differ slightly because some layers stay at higher precision. Try other combinations in the VRAM calculator.
FAQ
How much VRAM does Nanbeige4.3 33B need?
The weights are about 20.2 GB at Q4_K_M (33B params × 4.89 bits ÷ 8, before context/KV cache). Add memory for context (KV cache) and the runtime. The smallest-memory setup with a measured result here, fully in memory, is 2x RTX 3090 (48 GB) at 66.2 tok/s.
What is the fastest measured hardware for Nanbeige4.3 33B?
2x RTX 3090: 66.2 tok/s generation (vLLM TP2, 1 stream, W4A16 AutoRound), from https://github.com/tonyd2wild/Nanbeige4.3-33B-W4A16-2x-RTX-3090.
Which GPUs have measured results for Nanbeige4.3 33B?
2x RTX 3090.
Can it run on… 2x RTX 3090
Related: what GPU do you need for 2026 models · all builds