Ministral 3 3B VRAM Requirements
Ministral 3 3B needs about 3.5 GB of VRAM at Q4_K_M with an 8K context: 1.9 GB for weights, 0.8 GB for the KV cache and 0.8 GB of runtime overhead. At Q8_0 it needs 5.0 GB and at full FP16 precision 8.0 GB. The smallest GPU that runs it fully is the RTX 3050 8GB.
3.5 GB VRAM
Ministral 3 3B at Q4_K_M with 8K context and the KV cache in FP16.
- Weights 1.9 GB
- KV cache 0.8 GB
- Runtime overhead 0.8 GB
Which GPUs can run it
- RTX 3060 12GBFits~91 tok/s
- RTX 4060 Ti 16GBFits~73 tok/s
- RTX 3090 24GBFits~228 tok/s
- Radeon RX 7900 XTX 24GBFits~199 tok/s
- RTX 4090 24GBFits~244 tok/s
- RTX 5090 32GBFits~413 tok/s
- RTX 6000 Ada 48GBFits~233 tok/s
- A100 80GBFits~463 tok/s
- H100 80GB SXMFits~707 tok/s
- Ryzen AI Max+ 395 128GBFits~60 tok/s
- Apple M4 Max 128GBFits~136 tok/s
- Apple M5 Max 128GBFits~152 tok/s
- RTX PRO 6000 Blackwell 96GBFits~413 tok/s
- NVIDIA DGX Spark 128GBFits~64 tok/s
- Apple M3 Ultra 512GBFits~201 tok/s
Ministral 3 3B VRAM by quantization
Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.
| Quantization | Bits/weight | Weights | Total at 8K | Total at 32K | Smallest tier |
|---|---|---|---|---|---|
| Q4_K_M | 4.85 | 1.9 GB | 3.5 GB | 6.0 GB | 4 GB |
| Q5_K_M | 5.69 | 2.3 GB | 3.9 GB | 6.3 GB | 6 GB |
| Q6_K | 6.59 | 2.6 GB | 4.2 GB | 6.7 GB | 6 GB |
| Q8_0 | 8.50 | 3.4 GB | 5.0 GB | 7.4 GB | 6 GB |
| FP16 / BF16 | 16.00 | 6.4 GB | 8.0 GB | 10.4 GB | 10 GB |
| Q3_K_M | 3.91 | 1.6 GB | 3.2 GB | 5.6 GB | 4 GB |
| Q2_K | 3.00 | 1.2 GB | 2.8 GB | 5.2 GB | 4 GB |
VRAM by context length
At Q4_K_M. The KV cache grows linearly with context: 0.10 GB per 1,000 tokens in FP16.
| Context | KV cache (FP16) | KV cache (Q8_0) | Total, FP16 cache | Total, Q8_0 cache |
|---|---|---|---|---|
| 4K tokens | 0.4 GB | 0.2 GB | 3.1 GB | 2.9 GB |
| 8K tokens | 0.8 GB | 0.4 GB | 3.5 GB | 3.2 GB |
| 32K tokens | 3.3 GB | 1.7 GB | 6.0 GB | 4.5 GB |
| 128K tokens | 13.0 GB | 6.9 GB | 15.7 GB | 9.6 GB |
| 256K tokens | 26.0 GB | 13.8 GB | 28.7 GB | 16.5 GB |
GPUs that can run Ministral 3 3B
At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.
| GPU | Usable memory | Status | Best quant | Max context | Speed |
|---|---|---|---|---|---|
| RTX 3050 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~57 tok/s |
| RTX 4060 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~69 tok/s |
| RTX 4060 Ti 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~73 tok/s |
| Radeon RX 7600 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~62 tok/s |
| RTX 3060 Ti 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~112 tok/s |
| RTX 3070 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~112 tok/s |
| RTX 5060 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~112 tok/s |
| RTX 5060 Ti 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~112 tok/s |
| RTX 3070 Ti 8GB | 8.0 GB | Fits | Q8_0 | 47K | ~151 tok/s |
| Intel Arc B570 10GB | 10.0 GB | Fits | FP16 / BF16 | 65K | ~74 tok/s |
| RTX 3080 10GB | 10.0 GB | Fits | FP16 / BF16 | 65K | ~187 tok/s |
| Apple M4 16GB | 10.7 GB | Fits | FP16 / BF16 | 72K | ~31 tok/s |
| RTX 2080 Ti 11GB | 11.0 GB | Fits | FP16 / BF16 | 74K | ~153 tok/s |
| RTX 2060 12GB | 12.0 GB | Fits | FP16 / BF16 | 84K | ~85 tok/s |
About Ministral 3 3B
The smallest Ministral 3, a dense edge model with vision and a 256K context window. It uses plain grouped-query attention, so long contexts are not free: 128K of cache in FP16 is larger than the model itself. Keep the context to what you need on small GPUs.
| Spec | Value |
|---|---|
| Parameters | 3.43B |
| Layers | 26 |
| Hidden size | 3072 |
| Attention heads / KV heads | 32 / 8 |
| Head dimension | 128 |
| Vocabulary | 131,072 |
| Max context (used here) | 256K tokens |
| KV cache per token (FP16) | 104 KB |
| Released | 2025-12 |
| License | Apache 2.0 |
Ministral 3 3B VRAM: common questions
Can Ministral 3 3B run on a 4GB GPU?
Yes at Q4_K_M with up to about 8K of context. Longer contexts need a Q8_0 cache or more memory.
How much VRAM does Ministral 3 3B need?
About 3.5 GB at Q4_K_M with an 8K context, 5.0 GB at Q8_0 and 8.0 GB at FP16. Longer contexts add 0.10 GB per 1,000 tokens with an FP16 cache.
What is the cheapest GPU that can run Ministral 3 3B?
The smallest listed device that fits it fully at Q4_K_M with 8K context is the RTX 3050 8GB (8.0 GB usable).