Ministral 3 14B VRAM Requirements
Ministral 3 14B needs about 9.7 GB of VRAM at Q4_K_M with an 8K context: 7.6 GB for weights, 1.3 GB for the KV cache and 0.8 GB of runtime overhead. At Q8_0 it needs 15.4 GB and at full FP16 precision 27.2 GB. The smallest GPU that runs it fully is the Intel Arc B570 10GB.
9.7 GB VRAM
Ministral 3 14B at Q4_K_M with 8K context and the KV cache in FP16.
- Weights 7.6 GB
- KV cache 1.3 GB
- Runtime overhead 0.8 GB
Which GPUs can run it
- RTX 3060 12GBFits~26 tok/s
- RTX 4060 Ti 16GBFits~21 tok/s
- RTX 3090 24GBFits~67 tok/s
- Radeon RX 7900 XTX 24GBFits~58 tok/s
- RTX 4090 24GBFits~72 tok/s
- RTX 5090 32GBFits~125 tok/s
- RTX 6000 Ada 48GBFits~69 tok/s
- A100 80GBFits~141 tok/s
- H100 80GB SXMFits~224 tok/s
- Ryzen AI Max+ 395 128GBFits~17 tok/s
- Apple M4 Max 128GBFits~39 tok/s
- Apple M5 Max 128GBFits~44 tok/s
- RTX PRO 6000 Blackwell 96GBFits~125 tok/s
- NVIDIA DGX Spark 128GBFits~18 tok/s
- Apple M3 Ultra 512GBFits~59 tok/s
Ministral 3 14B 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 | 7.6 GB | 9.7 GB | 13.5 GB | 12 GB |
| Q5_K_M | 5.69 | 8.9 GB | 11.0 GB | 14.8 GB | 12 GB |
| Q6_K | 6.59 | 10.4 GB | 12.4 GB | 16.2 GB | 16 GB |
| Q8_0 | 8.50 | 13.4 GB | 15.4 GB | 19.2 GB | 20 GB |
| FP16 / BF16 | 16.00 | 25.1 GB | 27.2 GB | 31.0 GB | 32 GB |
| Q3_K_M | 3.91 | 6.1 GB | 8.2 GB | 12.0 GB | 10 GB |
| Q2_K | 3.00 | 4.7 GB | 6.8 GB | 10.5 GB | 8 GB |
VRAM by context length
At Q4_K_M. The KV cache grows linearly with context: 0.16 GB per 1,000 tokens in FP16.
| Context | KV cache (FP16) | KV cache (Q8_0) | Total, FP16 cache | Total, Q8_0 cache |
|---|---|---|---|---|
| 4K tokens | 0.6 GB | 0.3 GB | 9.1 GB | 8.8 GB |
| 8K tokens | 1.3 GB | 0.7 GB | 9.7 GB | 9.1 GB |
| 32K tokens | 5.0 GB | 2.7 GB | 13.5 GB | 11.1 GB |
| 128K tokens | 20.0 GB | 10.6 GB | 28.5 GB | 19.1 GB |
| 256K tokens | 40.0 GB | 21.3 GB | 48.5 GB | 29.7 GB |
GPUs that can run Ministral 3 14B
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 |
|---|---|---|---|---|---|
| Intel Arc B570 10GB | 10.0 GB | Tight fit | IQ4_XS | 6K | ~21 tok/s |
| RTX 3080 10GB | 10.0 GB | Tight fit | IQ4_XS | 6K | ~55 tok/s |
| Apple M4 16GB | 10.7 GB | Fits | Q4_K_M | 10K | ~8.8 tok/s |
| RTX 2080 Ti 11GB | 11.0 GB | Fits | Q4_K_M | 12K | ~44 tok/s |
| RTX 2060 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~24 tok/s |
| RTX 3060 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~26 tok/s |
| Radeon RX 6700 XT 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~24 tok/s |
| Radeon RX 7700 XT 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~27 tok/s |
| Intel Arc B580 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~25 tok/s |
| RTX 4070 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~36 tok/s |
| RTX 4070 Super 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~36 tok/s |
| RTX 4070 Ti 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~36 tok/s |
| RTX 5070 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~48 tok/s |
| RTX 3080 12GB | 12.0 GB | Fits | Q5_K_M | 18K | ~65 tok/s |
| RTX 3050 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~11 tok/s |
| RTX 3060 Ti 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~14 tok/s |
| RTX 3070 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~14 tok/s |
| RTX 3070 Ti 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~15 tok/s |
| RTX 4060 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~12 tok/s |
| RTX 4060 Ti 8GB | 8.0 GB | 78% on GPU | Q2_K | – | ~12 tok/s |
About Ministral 3 14B
The largest Ministral 3: a dense 14B model with vision, aimed at 12–16 GB GPUs. Its 40 layers with 8 KV heads cost 160 KB of cache per token in FP16, so context length is the main thing to watch on a 12 GB card.
| Spec | Value |
|---|---|
| Parameters | 13.5B |
| Layers | 40 |
| Hidden size | 5120 |
| Attention heads / KV heads | 32 / 8 |
| Head dimension | 128 |
| Vocabulary | 131,072 |
| Max context (used here) | 256K tokens |
| KV cache per token (FP16) | 160 KB |
| Released | 2025-12 |
| License | Apache 2.0 |
Ministral 3 14B VRAM: common questions
Is 12GB enough for Ministral 3 14B?
Yes at Q4_K_M with an 8K–16K context. For 32K or more, use a Q8_0 cache or a 16 GB card.
How much VRAM does Ministral 3 14B need?
About 9.7 GB at Q4_K_M with an 8K context, 15.4 GB at Q8_0 and 27.2 GB at FP16. Longer contexts add 0.16 GB per 1,000 tokens with an FP16 cache.
What is the cheapest GPU that can run Ministral 3 14B?
The smallest listed device that fits it fully at Q4_K_M with 8K context is the Intel Arc B570 10GB (10.0 GB usable).