Hardware requirements

What hardware do I need to run Mistral Small 3.2 24B?

23.6B 128K context Apache 2.0

At Q4_K_M the weights are 13.3 GB; with an 8K KV cache and runtime overhead that is 15.1 GB, so it wants a 20 GB card to run with headroom — in practice 20 GB VRAM: Radeon RX 7900 XT and 1 other. 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 16 GB VRAM

GeForce RTX 5080 — Q3_K_M, 12.6 GB of 14.4 GB, ~54 tok/s, up to 19K context.

Also: GeForce RTX 5070 Ti, GeForce RTX 4080 Super, GeForce RTX 4080, GeForce RTX 4070 Ti Super, Radeon RX 9070 XT +12 more
the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running
Recommended 20 GB VRAM

Radeon RX 7900 XT — Q4_K_M, 15.1 GB of 18.4 GB, ~36 tok/s, up to 28K context.

Also: RTX 4000 Ada
runs the recommended quantisation with headroom at 8K context
High quality 32 GB VRAM

GeForce RTX 5090 — Q8_0, 25.2 GB of 30.4 GB, ~46 tok/s, up to 41K context.

Also: RTX 5000 Ada, Radeon Pro W7800
near-lossless Q8_0 with headroom
Long context 48 GB VRAM

RTX 6000 Ada — Q4_K_M, 33.9 GB of 46.4 GB, ~44 tok/s at 128K.

Also: L40S, Radeon Pro W7900, RTX A6000
128K tokens (or the model’s maximum) at the recommended quantisation
Apple Silicon 24 GB unified

M2 · 24 GB — Q4_K_M, 15.1 GB of 18.0 GB, ~4.2 tok/s, up to 26K context.

Also: M3 · 24 GB
smallest Mac that runs the recommended quantisation with headroom
CPU only 32 GB RAM

CPU only · DDR4 dual-channel — Q8_0, 25.2 GB of 25.6 GB, ~0.9 tok/s, up to 10K context.

no GPU — weights stream from system RAM

How much VRAM at each quantisation

QuantWeights+KV 8K+KV 32K+KV 128KTotal @ 8KNeedsQuality
Q8_023.4 GB1.25 GB5.00 GB20.0 GB 25.2 GB32 GB card−0.1% ppl
Q6_K18.0 GB1.25 GB5.00 GB20.0 GB 19.9 GB32 GB card−0.4% ppl
Q5_K_M15.6 GB1.25 GB5.00 GB20.0 GB 17.4 GB24 GB card−0.8% ppl
Q4_K_M13.3 GB1.25 GB5.00 GB20.0 GB 15.1 GB20 GB card−1.9% ppl
Q3_K_M10.7 GB1.25 GB5.00 GB20.0 GB 12.6 GB20 GB card−5.4% ppl
Q2_K9.2 GB1.25 GB5.00 GB20.0 GB 11.1 GB16 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.

Every device, checked

System RAM for offload:
163264128256

NVIDIA · GeForce

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
GeForce RTX 5090 32 GB Q8_0 Q8_0 105K 82 no Check
GeForce RTX 5080 16 GB — Q3_K_M — — no Check
GeForce RTX 5070 Ti 16 GB — Q3_K_M — — no Check
GeForce RTX 5070 12 GB — offload — ~13 no Check
GeForce RTX 5060 Ti 16 GB 16 GB — Q3_K_M — — no Check
GeForce RTX 5060 8 GB — offload — ~5.8 no Check
GeForce RTX 4090 24 GB Q5_K_M Q6_K 54K 46 no Check
GeForce RTX 4080 Super 16 GB — Q3_K_M — — no Check
GeForce RTX 4080 16 GB — Q3_K_M — — no Check
GeForce RTX 4070 Ti Super 16 GB — Q3_K_M — — no Check
GeForce RTX 4070 Ti 12 GB — offload — ~12 no Check
GeForce RTX 4070 Super 12 GB — offload — ~12 no Check
GeForce RTX 4070 12 GB — offload — ~12 no Check
GeForce RTX 4060 Ti 16 GB 16 GB — Q3_K_M — — no Check
GeForce RTX 4060 Ti 8 GB — offload — ~5.4 no Check
GeForce RTX 4060 8 GB — offload — ~5.4 no Check
GeForce RTX 3090 Ti 24 GB Q5_K_M Q6_K 54K 46 no Check
GeForce RTX 3090 24 GB Q5_K_M Q6_K 54K 43 no Check
GeForce RTX 3080 Ti 12 GB — offload — ~14 no Check
GeForce RTX 3080 12 GB 12 GB — offload — ~14 no Check
GeForce RTX 3080 10 GB — offload — ~8.4 no Check
GeForce RTX 3070 Ti 8 GB — offload — ~6.0 no Check
GeForce RTX 3070 8 GB — offload — ~5.8 no Check
GeForce RTX 3060 Ti 8 GB — offload — ~5.8 no Check
GeForce RTX 3060 12 GB 12 GB — offload — ~11 no Check
GeForce RTX 2080 Ti 11 GB — offload — ~9.9 no Check
GeForce RTX 2060 12 GB 12 GB — offload — ~10 no Check
GeForce GTX 1080 Ti 11 GB — offload — ~9.4 no Check

NVIDIA · GeForce laptop

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
GeForce RTX 5090 Laptop 24 GB Q5_K_M Q6_K 54K 41 no Check
GeForce RTX 4090 Laptop 16 GB — Q3_K_M — — no Check
GeForce RTX 4080 Laptop 12 GB — offload — ~11 no Check
GeForce RTX 4070 Laptop 8 GB — offload — ~5.3 no Check

NVIDIA · Workstation

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
RTX 6000 Ada 48 GB Q8_0 Q8_0 128K 44 yes Check
RTX 5000 Ada 32 GB Q8_0 Q8_0 105K 26 no Check
RTX 4000 Ada 20 GB Q4_K_M Q5_K_M 28K 16 no Check
RTX 2000 Ada 16 GB — Q3_K_M — — no Check
RTX A6000 48 GB Q8_0 Q8_0 128K 35 yes Check
RTX A5000 24 GB Q5_K_M Q6_K 54K 35 no Check
RTX A4000 16 GB — Q3_K_M — — no Check

NVIDIA · Data centre

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
L40S 48 GB Q8_0 Q8_0 128K 39 yes Check
L4 24 GB Q5_K_M Q6_K 54K 14 no Check
A10 24 GB Q5_K_M Q6_K 54K 27 no Check
A100 40 GB 40 GB Q8_0 Q8_0 128K 71 tight Check
A100 80 GB 80 GB Q8_0 Q8_0 128K 93 yes Check
H100 PCIe 80 GB Q8_0 Q8_0 128K 91 yes Check
H100 SXM 80 GB Q8_0 Q8_0 128K 153 yes Check
H200 SXM 141 GB Q8_0 Q8_0 128K 219 yes Check
Tesla P40 24 GB Q5_K_M Q6_K 54K 16 no Check

AMD · Radeon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Radeon RX 9070 XT 16 GB — Q3_K_M — — no Check
Radeon RX 9070 16 GB — Q3_K_M — — no Check
Radeon RX 7900 XTX 24 GB Q5_K_M Q6_K 54K 44 no Check
Radeon RX 7900 XT 20 GB Q4_K_M Q5_K_M 28K 36 no Check
Radeon RX 7900 GRE 16 GB — Q3_K_M — — no Check
Radeon RX 7800 XT 16 GB — Q3_K_M — — no Check
Radeon RX 7700 XT 12 GB — offload — ~11 no Check
Radeon RX 7600 XT 16 GB — Q3_K_M — — no Check
Radeon RX 6900 XT 16 GB — Q3_K_M — — no Check
Radeon RX 6800 XT 16 GB — Q3_K_M — — no Check
Radeon RX 6700 XT 12 GB — offload — ~11 no Check

AMD · Radeon Pro

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Radeon Pro W7900 48 GB Q8_0 Q8_0 128K 39 yes Check
Radeon Pro W7800 32 GB Q8_0 Q8_0 105K 26 no Check

AMD · Instinct

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Instinct MI210 64 GB Q8_0 Q8_0 128K 75 yes Check
Instinct MI300X 192 GB Q8_0 Q8_0 128K 242 yes Check

AMD · Ryzen AI Max

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Ryzen AI Max+ 395 · 128 GB 128 GB Q8_0 Q8_0 128K 11 yes Check
Ryzen AI Max+ 395 · 64 GB 64 GB Q8_0 Q8_0 128K 11 yes Check

Apple · Apple Silicon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
M1 · 16 GB 16 GB — no — — no Check
M1 Pro · 32 GB 32 GB Q6_K Q6_K 64K 8.4 no Check
M1 Max · 64 GB 64 GB Q8_0 Q8_0 128K 17 yes Check
M1 Ultra · 128 GB 128 GB Q8_0 Q8_0 128K 34 yes Check
M2 · 24 GB 24 GB Q4_K_M Q5_K_M 26K 4.2 no Check
M2 Pro · 32 GB 32 GB Q6_K Q6_K 64K 8.4 no Check
M2 Max · 96 GB 96 GB Q8_0 Q8_0 128K 17 yes Check
M2 Ultra · 192 GB 192 GB Q8_0 Q8_0 128K 34 yes Check
M3 · 24 GB 24 GB Q4_K_M Q5_K_M 26K 4.2 no Check
M3 Pro · 36 GB 36 GB Q6_K Q8_0 84K 6.3 no Check
M3 Max · 64 GB 64 GB Q8_0 Q8_0 128K 17 yes Check
M3 Max · 128 GB 128 GB Q8_0 Q8_0 128K 17 yes Check
M3 Ultra · 256 GB 256 GB Q8_0 Q8_0 128K 34 yes Check
M3 Ultra · 512 GB 512 GB Q8_0 Q8_0 128K 34 yes Check
M4 · 32 GB 32 GB Q6_K Q6_K 64K 5.1 no Check
M4 Pro · 48 GB 48 GB Q8_0 Q8_0 128K 11 tight Check
M4 Pro · 64 GB 64 GB Q8_0 Q8_0 128K 11 yes Check
M4 Max · 64 GB 64 GB Q8_0 Q8_0 128K 17 yes Check
M4 Max · 128 GB 128 GB Q8_0 Q8_0 128K 23 yes Check

Intel · Arc

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Arc B580 12 GB — offload — ~12 no Check
Arc B570 10 GB — offload — ~7.4 no Check
Arc A770 16 GB 16 GB — Q3_K_M — — no Check
Arc A750 8 GB — offload — ~5.9 no Check

CPU · CPU + system RAM

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
CPU only · DDR4 dual-channel 32 GB Q6_K Q8_0 75K 1.6 no Check
CPU only · DDR5 dual-channel 64 GB Q8_0 Q8_0 128K 2.8 yes Check
CPU only · DDR5 8-channel server 256 GB Q8_0 Q8_0 128K 9.7 yes Check