Hardware requirements
What hardware do I need to run Ministral 3 14B?
13.9B
256K context
Apache 2.0
At Q4_K_M the weights are 7.8 GB; with an 8K KV cache and runtime overhead that is 9.7 GB, so it wants a 16 GB card to run with headroom — in practice 16 GB VRAM: GeForce RTX 5080 and 17 others. Every figure below is computed from the model’s published shape against each device’s usable memory; tokens per second are estimates. The method.
Minimum practical
10 GB VRAM
GeForce RTX 3080 — Q3_K_M, 8.2 GB of 8.8 GB, ~73 tok/s, up to 11K context.
the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running
Recommended
16 GB VRAM
GeForce RTX 5080 — Q4_K_M, 9.7 GB of 14.4 GB, ~74 tok/s, up to 38K context.
runs the recommended quantisation with headroom at 8K context
High quality
20 GB VRAM
Radeon RX 7900 XT — Q8_0, 15.6 GB of 18.4 GB, ~35 tok/s, up to 25K context.
near-lossless Q8_0 with headroom
Long context
40 GB VRAM
A100 40 GB — Q4_K_M, 28.4 GB of 38.4 GB, ~120 tok/s at 128K.
128K tokens (or the model’s maximum) at the recommended quantisation
Apple Silicon
24 GB unified
M2 · 24 GB — Q4_K_M, 9.7 GB of 18.0 GB, ~7.1 tok/s, up to 61K context.
smallest Mac that runs the recommended quantisation with headroom
CPU only
32 GB RAM
CPU only · DDR4 dual-channel — Q8_0, 15.6 GB of 25.6 GB, ~1.6 tok/s, up to 71K context.
no GPU — weights stream from system RAM
How much VRAM at each quantisation
| Quant | Weights | +KV 8K | +KV 32K | +KV 128K | Total @ 8K | Needs | Quality |
|---|---|---|---|---|---|---|---|
| F16 | 25.9 GB | 1.25 GB | 5.00 GB | 20.0 GB | 27.7 GB | 40 GB card | Reference |
| Q8_0 | 13.8 GB | 1.25 GB | 5.00 GB | 20.0 GB | 15.6 GB | 20 GB card | −0.1% ppl |
| Q6_K | 10.6 GB | 1.25 GB | 5.00 GB | 20.0 GB | 12.5 GB | 20 GB card | −0.4% ppl |
| Q5_K_M | 9.2 GB | 1.25 GB | 5.00 GB | 20.0 GB | 11.0 GB | 16 GB card | −0.8% ppl |
| Q4_K_M | 7.8 GB | 1.25 GB | 5.00 GB | 20.0 GB | 9.7 GB | 16 GB card | −1.9% ppl |
| Q3_K_M | 6.3 GB | 1.25 GB | 5.00 GB | 20.0 GB | 8.2 GB | 11 GB card | −5.4% ppl |
| Q2_K | 5.4 GB | 1.25 GB | 5.00 GB | 20.0 GB | 7.3 GB | 10 GB card | −15% ppl |
Totals add 0.6 GB runtime overhead and an f16 KV cache; "needs" is the smallest discrete card class whose usable memory (after display and driver reserve) leaves 15% headroom. A q8_0 cache roughly halves the KV columns.
NVIDIA · GeForce
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| GeForce RTX 5090 | 32 GB | Q8_0 | Q8_0 | 140K | 139 | tight | Check |
| GeForce RTX 5080 | 16 GB | Q5_K_M | Q6_K | 38K | 74 | no | Check |
| GeForce RTX 5070 Ti | 16 GB | Q5_K_M | Q6_K | 38K | 69 | no | Check |
| GeForce RTX 5070 | 12 GB | Q3_K_M | Q4_K_M | 13K | 52 | no | Check |
| GeForce RTX 5060 Ti 16 GB | 16 GB | Q5_K_M | Q6_K | 38K | 35 | no | Check |
| GeForce RTX 5060 | 8 GB | — | offload | — | ~19 | no | Check |
| GeForce RTX 4090 | 24 GB | Q8_0 | Q8_0 | 89K | 78 | no | Check |
| GeForce RTX 4080 Super | 16 GB | Q5_K_M | Q6_K | 38K | 57 | no | Check |
| GeForce RTX 4080 | 16 GB | Q5_K_M | Q6_K | 38K | 56 | no | Check |
| GeForce RTX 4070 Ti Super | 16 GB | Q5_K_M | Q6_K | 38K | 52 | no | Check |
| GeForce RTX 4070 Ti | 12 GB | Q3_K_M | Q4_K_M | 13K | 39 | no | Check |
| GeForce RTX 4070 Super | 12 GB | Q3_K_M | Q4_K_M | 13K | 39 | no | Check |
| GeForce RTX 4070 | 12 GB | Q3_K_M | Q4_K_M | 13K | 39 | no | Check |
| GeForce RTX 4060 Ti 16 GB | 16 GB | Q5_K_M | Q6_K | 38K | 22 | no | Check |
| GeForce RTX 4060 Ti | 8 GB | — | offload | — | ~16 | no | Check |
| GeForce RTX 4060 | 8 GB | — | offload | — | ~16 | no | Check |
| GeForce RTX 3090 Ti | 24 GB | Q8_0 | Q8_0 | 89K | 78 | no | Check |
| GeForce RTX 3090 | 24 GB | Q8_0 | Q8_0 | 89K | 72 | no | Check |
| GeForce RTX 3080 Ti | 12 GB | Q3_K_M | Q4_K_M | 13K | 71 | no | Check |
| GeForce RTX 3080 12 GB | 12 GB | Q3_K_M | Q4_K_M | 13K | 71 | no | Check |
| GeForce RTX 3080 | 10 GB | — | Q3_K_M | — | — | no | Check |
| GeForce RTX 3070 Ti | 8 GB | — | offload | — | ~22 | no | Check |
| GeForce RTX 3070 | 8 GB | — | offload | — | ~19 | no | Check |
| GeForce RTX 3060 Ti | 8 GB | — | offload | — | ~19 | no | Check |
| GeForce RTX 3060 12 GB | 12 GB | Q3_K_M | Q4_K_M | 13K | 28 | no | Check |
| GeForce RTX 2080 Ti | 11 GB | Q3_K_M | Q4_K_M | 8K | 48 | no | Check |
| GeForce RTX 2060 12 GB | 12 GB | Q3_K_M | Q4_K_M | 13K | 26 | no | Check |
| GeForce GTX 1080 Ti | 11 GB | Q3_K_M | Q4_K_M | 8K | 37 | no | Check |
NVIDIA · GeForce laptop
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| GeForce RTX 5090 Laptop | 24 GB | Q8_0 | Q8_0 | 89K | 69 | no | Check |
| GeForce RTX 4090 Laptop | 16 GB | Q5_K_M | Q6_K | 38K | 45 | no | Check |
| GeForce RTX 4080 Laptop | 12 GB | Q3_K_M | Q4_K_M | 13K | 33 | no | Check |
| GeForce RTX 4070 Laptop | 8 GB | — | offload | — | ~15 | no | Check |
NVIDIA · Workstation
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| RTX 6000 Ada | 48 GB | Q8_0 | Q8_0 | 243K | 74 | yes | Check |
| RTX 5000 Ada | 32 GB | Q8_0 | Q8_0 | 140K | 45 | tight | Check |
| RTX 4000 Ada | 20 GB | Q8_0 | Q8_0 | 63K | 28 | no | Check |
| RTX 2000 Ada | 16 GB | Q5_K_M | Q6_K | 38K | 17 | no | Check |
| RTX A6000 | 48 GB | Q8_0 | Q8_0 | 243K | 59 | yes | Check |
| RTX A5000 | 24 GB | Q8_0 | Q8_0 | 89K | 59 | no | Check |
| RTX A4000 | 16 GB | Q5_K_M | Q6_K | 38K | 35 | no | Check |
NVIDIA · Data centre
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| L40S | 48 GB | Q8_0 | Q8_0 | 243K | 67 | yes | Check |
| L4 | 24 GB | Q8_0 | Q8_0 | 89K | 23 | no | Check |
| A10 | 24 GB | Q8_0 | Q8_0 | 89K | 46 | no | Check |
| A100 40 GB | 40 GB | Q8_0 | Q8_0 | 191K | 120 | yes | Check |
| A100 80 GB | 80 GB | Q8_0 | Q8_0 | 256K | 158 | yes | Check |
| H100 PCIe | 80 GB | Q8_0 | Q8_0 | 256K | 155 | yes | Check |
| H100 SXM | 80 GB | Q8_0 | Q8_0 | 256K | 259 | yes | Check |
| H200 SXM | 141 GB | Q8_0 | Q8_0 | 256K | 372 | yes | Check |
| Tesla P40 | 24 GB | Q8_0 | Q8_0 | 89K | 27 | no | Check |
AMD · Radeon
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Radeon RX 9070 XT | 16 GB | Q5_K_M | Q6_K | 38K | 50 | no | Check |
| Radeon RX 9070 | 16 GB | Q5_K_M | Q6_K | 38K | 50 | no | Check |
| Radeon RX 7900 XTX | 24 GB | Q8_0 | Q8_0 | 89K | 74 | no | Check |
| Radeon RX 7900 XT | 20 GB | Q8_0 | Q8_0 | 63K | 62 | no | Check |
| Radeon RX 7900 GRE | 16 GB | Q5_K_M | Q6_K | 38K | 45 | no | Check |
| Radeon RX 7800 XT | 16 GB | Q5_K_M | Q6_K | 38K | 48 | no | Check |
| Radeon RX 7700 XT | 12 GB | Q3_K_M | Q4_K_M | 13K | 33 | no | Check |
| Radeon RX 7600 XT | 16 GB | Q5_K_M | Q6_K | 38K | 22 | no | Check |
| Radeon RX 6900 XT | 16 GB | Q5_K_M | Q6_K | 38K | 40 | no | Check |
| Radeon RX 6800 XT | 16 GB | Q5_K_M | Q6_K | 38K | 40 | no | Check |
| Radeon RX 6700 XT | 12 GB | Q3_K_M | Q4_K_M | 13K | 30 | no | Check |
AMD · Radeon Pro
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Radeon Pro W7900 | 48 GB | Q8_0 | Q8_0 | 243K | 67 | yes | Check |
| Radeon Pro W7800 | 32 GB | Q8_0 | Q8_0 | 140K | 45 | tight | Check |
AMD · Instinct
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Instinct MI210 | 64 GB | Q8_0 | Q8_0 | 256K | 127 | yes | Check |
| Instinct MI300X | 192 GB | Q8_0 | Q8_0 | 256K | 411 | yes | Check |
AMD · Ryzen AI Max
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Ryzen AI Max+ 395 · 128 GB | 128 GB | Q8_0 | Q8_0 | 256K | 18 | yes | Check |
| Ryzen AI Max+ 395 · 64 GB | 64 GB | Q8_0 | Q8_0 | 253K | 18 | yes | Check |
Apple · Apple Silicon
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| M1 · 16 GB | 16 GB | Q3_K_M | Q4_K_M | 14K | 4.9 | no | Check |
| M1 Pro · 32 GB | 32 GB | Q8_0 | Q8_0 | 99K | 14 | no | Check |
| M1 Max · 64 GB | 64 GB | Q8_0 | Q8_0 | 253K | 29 | yes | Check |
| M1 Ultra · 128 GB | 128 GB | Q8_0 | Q8_0 | 256K | 57 | yes | Check |
| M2 · 24 GB | 24 GB | Q6_K | Q8_0 | 61K | 7.1 | no | Check |
| M2 Pro · 32 GB | 32 GB | Q8_0 | Q8_0 | 99K | 14 | no | Check |
| M2 Max · 96 GB | 96 GB | Q8_0 | Q8_0 | 256K | 29 | yes | Check |
| M2 Ultra · 192 GB | 192 GB | Q8_0 | Q8_0 | 256K | 57 | yes | Check |
| M3 · 24 GB | 24 GB | Q6_K | Q8_0 | 61K | 7.1 | no | Check |
| M3 Pro · 36 GB | 36 GB | Q8_0 | Q8_0 | 118K | 11 | no | Check |
| M3 Max · 64 GB | 64 GB | Q8_0 | Q8_0 | 253K | 29 | yes | Check |
| M3 Max · 128 GB | 128 GB | Q8_0 | Q8_0 | 256K | 29 | yes | Check |
| M3 Ultra · 256 GB | 256 GB | Q8_0 | Q8_0 | 256K | 57 | yes | Check |
| M3 Ultra · 512 GB | 512 GB | Q8_0 | Q8_0 | 256K | 57 | yes | Check |
| M4 · 32 GB | 32 GB | Q8_0 | Q8_0 | 99K | 8.6 | no | Check |
| M4 Pro · 48 GB | 48 GB | Q8_0 | Q8_0 | 176K | 20 | yes | Check |
| M4 Pro · 64 GB | 64 GB | Q8_0 | Q8_0 | 253K | 20 | yes | Check |
| M4 Max · 64 GB | 64 GB | Q8_0 | Q8_0 | 253K | 29 | yes | Check |
| M4 Max · 128 GB | 128 GB | Q8_0 | Q8_0 | 256K | 39 | yes | Check |
Intel · Arc
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Arc B580 | 12 GB | Q3_K_M | Q4_K_M | 13K | 35 | no | Check |
| Arc B570 | 10 GB | — | Q3_K_M | — | — | no | Check |
| Arc A770 16 GB | 16 GB | Q5_K_M | Q6_K | 38K | 43 | no | Check |
| Arc A750 | 8 GB | — | offload | — | ~20 | no | Check |
CPU · CPU + system RAM
| Device | Memory | Best quant with headroom | Fits at all | Max context @ Q4_K_M | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| CPU only · DDR4 dual-channel | 32 GB | Q8_0 | Q8_0 | 109K | 2.7 | no | Check |
| CPU only · DDR5 dual-channel | 64 GB | Q8_0 | Q8_0 | 256K | 4.8 | yes | Check |
| CPU only · DDR5 8-channel server | 256 GB | Q8_0 | Q8_0 | 256K | 16 | yes | Check |