Mistral AI Mistral Small

Mistral Small 3.2 24B hardware requirements

Fits on one Instinct MI355X OAM at BF16. For 1,000 tokens/s at 32k context you need 8 x H100 SXM5 80GB at $14.64/hr or 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $8.00/hr.

Parameters

24.0B

Active per token

All

dense model

KV cache per token

160 KB

16-bit cache

Native precision

BF16

Rent Model card

Top options for Mistral Small 3.2 24B

The strongest part each vendor offers for Mistral Small 3.2 24B at BF16 with 32,768 tokens per user, rentable parts first. Each card is one replica, the fewest accelerators that hold the model, and the line it draws on the graph below. Select a card to plot it.

Tokens per second as users grow

Each line is one replica, the fewest accelerators that hold the model, serving more users at once. It climbs while memory bandwidth is the limit, flattens once compute is, and stops where memory runs out at 32,768 tokens per user. Hover a line to bring it forward.

1 x GB300 NVL72 GPU 288GB 1 x Instinct MI355X OAM 1 x Ascend 910C 1 x Gaudi 2 96GB 1 x TPU v7 192GB 2 x Inferentia2

Memory fills at: 1 x GB300 NVL72 GPU 288GB 42 users, 1 x Instinct MI355X OAM 42 users, 1 x Ascend 910C 13 users, 1 x Gaudi 2 96GB 8 users, 1 x TPU v7 192GB 25 users, 2 x Inferentia2 2 users. Roofline estimates at BF16.

Bare minimum

Fewest GPUs

1 x Instinct MI355X OAM

BF16, 4,096 tokens, one stream

Tokens/s: 122
Per hour: $5.99
Per million tokens: $13.69
Memory used: 18%

Rent on Daytona

Cheapest per hour

1 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

BF16, 4,096 tokens, one stream, live price

Tokens/s: 27.1
Per hour: $0.50
Per million tokens: $5.12
Memory used: 53%

Rent on RunPod

Cheapest per million tokens

4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

BF16, 32,768 tokens, 32 streams

Tokens/s: 332
Per hour: $2.00
Per million tokens: $1.68
Memory used: 60%

Rent on RunPod

Configurations

Every datacenter GPU in the catalog, sized for these settings. Change precision, context and concurrency and the whole page follows. Each point is one replica: the smallest tensor-parallel group of that GPU that holds the model.

Precision Context Streams

Vendor

Priced only

Priced configurations only; 67 more fit without a live price and appear in the table. Marker size grows with the tensor-parallel width. Roofline estimates.

#									Rent
1	Instinct MI455X	1	53 %	1.1k tok/s	79.5 tok/s	—	—	—	Specs →
2	Rubin SXM	1	79 %	955 tok/s	75.1 tok/s	—	—	—	Specs →
3	TPU 8t announced	2	53 %	~862 tok/s	42.3 tok/s	—	—	—	Specs →
4	TPU 8i announced	1	79 %	~485 tok/s	29.3 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	1	79 %	466 tok/s	27.3 tok/s	$5.99	$3.57	1.4 kW	Rent →
6	GB300 NVL72 GPU 288GB	1	79 %	464 tok/s	27.3 tok/s	$12.42	$7.43	—	Rent →
7	GB200 NVL72 GPU 186GB	2	62 %	882 tok/s	51.9 tok/s	$21.00	$6.61	—	Rent →
8	Instinct MI350X OAM	1	79 %	447 tok/s	27.3 tok/s	$6.16	$3.83	1.0 kW	Rent →
9	TPU v7 192GB	2	60 %	814 tok/s	47.8 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	1	53 %	~671 tok/s	79.5 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	1	87 %	441 tok/s	27.3 tok/s	$4.99	$3.14	1.4 kW	Rent →
12	B200 SXM 180GB	2	64 %	822 tok/s	49.9 tok/s	$8.18	$2.76	2.0 kW	Rent →
13	Ascend 970 announced	1	79 %	527 tok/s	49.1 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	2	60 %	735 tok/s	51.9 tok/s	—	—	1.4 kW	Specs →
15	Jalapeño announced	2	53 %	~914 tok/s	99.8 tok/s	—	—	1.4 kW	Specs →
16	Instinct MI325X OAM	1	89 %	289 tok/s	20.5 tok/s	$3.80	$3.65	1.0 kW	Rent →
17	Instinct MI300X 192GB	2	60 %	520 tok/s	34.4 tok/s	$4.78	$2.56	1.5 kW	Rent →
18	BR100	4	90 %	440 tok/s	19.7 tok/s	—	—	2.2 kW	Specs →
19	Ascend 960 announced	1	79 %	288 tok/s	32.8 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	2	81 %	427 tok/s	31.1 tok/s	$4.96	$3.23	1.4 kW	Rent →
21	H100 SXM5 80GB	4	72 %	682 tok/s	41.1 tok/s	$7.32	$2.98	2.8 kW	Rent →
22	TPU v6e 32GB	8	92 %	816 tok/s	38.0 tok/s	—	—	—	Specs →
23	Data Center GPU Max 1550 128GB	2	90 %	328 tok/s	21.2 tok/s	—	—	1.2 kW	Specs →
24	H200 NVL 141GB	2	81 %	~386 tok/s	31.1 tok/s	$7.23	$5.21	1.2 kW	Rent →
25	H100 NVL 94GB	4	61 %	~671 tok/s	47.9 tok/s	$12.76	$5.28	1.6 kW	Rent →
26	Ascend 910C	2	90 %	316 tok/s	20.7 tok/s	—	—	1.2 kW	Specs →
27	H100 PCIe 80GB	4	72 %	455 tok/s	24.6 tok/s	$10.00	$6.11	1.4 kW	Rent →
28	H800 PCIe 80GB	4	72 %	455 tok/s	24.6 tok/s	—	—	1.4 kW	Specs →
29	Ascend 950DT	2	80 %	275 tok/s	25.9 tok/s	—	—	—	Specs →
30	RTX PRO 6000 Blackwell Workstation Edition	4	60 %	356 tok/s	22.0 tok/s	$7.20	$5.62	2.4 kW	Rent →
31	Ascend 950PR	2	90 %	176 tok/s	10.4 tok/s	—	—	1.8 kW	Specs →
32	MTT S5000	4	72 %	~334 tok/s	19.7 tok/s	—	—	—	Specs →
33	RTX PRO 6000 Blackwell Server Edition	4	60 %	334 tok/s	19.6 tok/s	$2.76	$2.30	2.4 kW	Rent →
34	Gaudi 3 128GB	2	90 %	478 tok/s	24.0 tok/s	—	—	1.8 kW	Specs →
35	TPU v5p 95GB	4	61 %	409 tok/s	34.0 tok/s	—	—	1.8 kW	Specs →
36	RTX PRO 6000 Blackwell Max-Q Workstation Edition	4	60 %	332 tok/s	22.0 tok/s	$2.00	$1.68	1.2 kW	Rent →
37	Cloud AI 100 Ultra	2	90 %	~84.4 tok/s	3.6 tok/s	—	—	0.3 kW	Specs →
38	Gaudi 2 96GB	4	60 %	376 tok/s	30.1 tok/s	$4.40	$3.25	2.4 kW	Rent →
39	Atlas 350	4	51 %	289 tok/s	17.2 tok/s	—	—	2.4 kW	Specs →
40	Data Center GPU Max 1100 48GB	8	61 %	505 tok/s	28.5 tok/s	—	—	2.4 kW	Specs →
41	GeForce RTX 5090 32GB	8	92 %	397 tok/s	41.6 tok/s	$9.52	$6.67	4.6 kW	Rent →
42	Ascend 910B	4	90 %	299 tok/s	19.7 tok/s	—	—	1.6 kW	Specs →
43	MI250X 128GB	2	90 %	203 tok/s	21.2 tok/s	—	—	1.1 kW	Specs →
44	RTX 6000 Ada 48GB	8	61 %	~413 tok/s	22.3 tok/s	$7.76	$5.22	2.4 kW	Rent →
45	Instinct MI250 128GB	2	90 %	195 tok/s	21.2 tok/s	—	—	1.0 kW	Specs →
46	L40S 48GB	8	61 %	387 tok/s	20.0 tok/s	$6.08	$4.36	2.8 kW	Rent →
47	A800 PCIe 80GB	4	72 %	281 tok/s	23.8 tok/s	—	—	1.2 kW	Specs →
48	A800 SXM4	4	72 %	286 tok/s	25.0 tok/s	—	—	1.6 kW	Specs →
49	A100 PCIe 80GB	4	72 %	281 tok/s	23.8 tok/s	$5.40	$5.34	1.2 kW	Rent →
50	A100 SXM4 80GB	4	72 %	286 tok/s	25.0 tok/s	$5.60	$5.43	1.6 kW	Rent →
51	A100 PCIe 40GB	8	73 %	487 tok/s	36.1 tok/s	—	—	2.0 kW	Specs →
52	A100 SXM4 40GB	8	73 %	487 tok/s	36.1 tok/s	$12.49	$7.13	3.2 kW	Rent →
53	GeForce RTX 5090 D V2 24GB 2 nodes	16	62 %	~784 tok/s	55.0 tok/s	—	—	9.2 kW	Specs →
54	RTX 5880 Ada 48GB	8	61 %	~365 tok/s	22.3 tok/s	—	—	2.3 kW	Specs →
55	TPU v4 32GB	8	92 %	405 tok/s	27.8 tok/s	—	—	—	Specs →
56	RTX 5000 Ada 32GB	8	92 %	~266 tok/s	13.4 tok/s	$6.64	$6.93	2.0 kW	Rent →
57	RTX PRO 5000 Blackwell 48GB	8	61 %	~410 tok/s	31.2 tok/s	$7.68	$5.20	2.4 kW	Rent →
58	RTX PRO 5000 Blackwell 72GB	4	80 %	~217 tok/s	16.5 tok/s	—	—	1.2 kW	Specs →
59	RTX PRO 4500 Blackwell Server Edition	8	92 %	285 tok/s	18.6 tok/s	—	—	1.3 kW	Specs →
60	RTX PRO 4500 Blackwell 32GB	8	92 %	300 tok/s	20.8 tok/s	$5.76	$5.33	1.6 kW	Rent →
61	Radeon AI PRO R9700 32GB	8	92 %	247 tok/s	14.8 tok/s	—	—	2.4 kW	Specs →
62	Inferentia2	8	92 %	278 tok/s	19.0 tok/s	—	—	—	Specs →
63	Trainium	8	92 %	278 tok/s	19.0 tok/s	—	—	—	Specs →
64	Instinct MI100 32GB	8	92 %	196 tok/s	28.5 tok/s	—	—	2.4 kW	Specs →
65	Arc Pro B70 32GB	8	92 %	~236 tok/s	14.1 tok/s	—	—	1.8 kW	Specs →
66	L40 48GB	8	61 %	277 tok/s	20.0 tok/s	$6.56	$6.57	2.4 kW	Rent →
67	Instinct MI210 PCIe	4	90 %	184 tok/s	20.1 tok/s	—	—	1.2 kW	Specs →
68	GeForce RTX 4090 24GB 2 nodes	16	62 %	493 tok/s	41.3 tok/s	$14.24	$8.03	7.2 kW	Rent →
69	A30 24GB 2 nodes	16	62 %	478 tok/s	38.2 tok/s	—	—	2.6 kW	Specs →
70	RTX 4500 Ada 24GB 2 nodes	16	62 %	~323 tok/s	17.7 tok/s	—	—	3.4 kW	Specs →
71	RTX A6000 48GB	8	61 %	241 tok/s	17.8 tok/s	$4.00	$4.61	2.4 kW	Rent →
72	A40 48GB	8	61 %	227 tok/s	16.1 tok/s	$4.72	$5.78	2.4 kW	Rent →
73	H20 141GB HBM3e	2	81 %	98.3 tok/s	25.9 tok/s	—	—	0.8 kW	Specs →
74	H20 96GB	4	60 %	186 tok/s	49.1 tok/s	—	—	1.6 kW	Specs →
75	RTX PRO 4000 Blackwell 24GB 2 nodes	16	62 %	~390 tok/s	27.5 tok/s	$9.12	$6.49	2.3 kW	Rent →
76	BI-V100 Tiangai 100	8	92 %	274 tok/s	27.8 tok/s	—	—	2.0 kW	Specs →
77	Atlas 300I Duo 48GB	8	61 %	168 tok/s	9.5 tok/s	—	—	1.2 kW	Specs →
78	Atlas 300I Duo 96GB	4	60 %	89.0 tok/s	5.0 tok/s	—	—	0.6 kW	Specs →
79	RTX A5500 24GB 2 nodes	16	62 %	~395 tok/s	31.4 tok/s	—	—	3.7 kW	Specs →
80	Wormhole n300s 2 nodes	16	62 %	341 tok/s	23.6 tok/s	—	—	4.8 kW	Specs →
81	V100S PCIe 32GB	8	92 %	248 tok/s	26.3 tok/s	$7.04	$7.90	2.0 kW	Rent →
82	ATOM-Max	4	90 %	126 tok/s	12.6 tok/s	—	—	1.4 kW	Specs →
83	A10 24GB 2 nodes	16	62 %	339 tok/s	24.6 tok/s	$17.63	$14.46	2.4 kW	Rent →
84	Radeon PRO W7900 Dual Slot 48GB	8	61 %	219 tok/s	20.0 tok/s	—	—	2.4 kW	Specs →
85	Radeon PRO W7900 48GB	8	61 %	219 tok/s	20.0 tok/s	—	—	2.4 kW	Specs →
86	Radeon RX 7900 XTX 24GB 2 nodes	16	62 %	399 tok/s	39.3 tok/s	—	—	5.7 kW	Specs →
87	L4 24GB 2 nodes	16	62 %	234 tok/s	12.3 tok/s	$9.44	$11.22	1.2 kW	Rent →
88	L20 48GB	8	61 %	215 tok/s	20.0 tok/s	—	—	2.2 kW	Specs →
89	Radeon AI PRO R9600D 32GB	8	92 %	171 tok/s	14.8 tok/s	—	—	1.2 kW	Specs →
90	Arc Pro B65 32GB	8	92 %	~167 tok/s	14.1 tok/s	—	—	1.6 kW	Specs →
91	Arc Pro B60 24GB 2 nodes	16	62 %	~263 tok/s	18.7 tok/s	—	—	3.2 kW	Specs →
92	L2 24GB 2 nodes	16	62 %	212 tok/s	12.3 tok/s	—	—	—	Specs →
93	RTX PRO 4000 Blackwell SFF 24GB 2 nodes	16	62 %	~253 tok/s	17.7 tok/s	—	—	1.1 kW	Specs →
94	MLU370-X8	8	61 %	165 tok/s	14.3 tok/s	—	—	2.0 kW	Specs →
95	Radeon PRO W7800 48GB	8	61 %	177 tok/s	20.0 tok/s	—	—	2.1 kW	Specs →
96	Radeon PRO W7800 32GB	8	92 %	155 tok/s	13.4 tok/s	—	—	2.1 kW	Specs →
97	GeForce RTX 3090 24GB 2 nodes	16	62 %	265 tok/s	38.3 tok/s	$8.00	$8.39	5.6 kW	Rent →
98	A10G 24GB 2 nodes	16	62 %	234 tok/s	24.6 tok/s	$16.10	$19.11	4.8 kW	Rent →
99	Atlas 300V Pro	8	61 %	84.2 tok/s	4.8 tok/s	—	—	—	Specs →
100	Atlas 300I Pro 2 nodes	16	62 %	149 tok/s	8.4 tok/s	—	—	—	Specs →
101	Atlas 300V 2 nodes	16	62 %	126 tok/s	8.4 tok/s	—	—	1.2 kW	Specs →
102	GeForce RTX 3090 Ti 24GB 2 nodes	16	62 %	166 tok/s	41.3 tok/s	$7.36	$12.29	7.2 kW	Rent →
103	Radeon PRO W6800 32GB	8	92 %	76.8 tok/s	11.9 tok/s	—	—	2.0 kW	Specs →
104	RTX A5000 24GB 2 nodes	16	62 %	117 tok/s	31.4 tok/s	$4.32	$10.26	3.7 kW	Rent →
105	Radeon PRO V710 28GB 2 nodes	16	53 %	108 tok/s	18.3 tok/s	—	—	2.5 kW	Specs →

By vendor

At BF16 the NVIDIA pick is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 332 tokens/s for $2.00/hr; the AMD pick is 2 x Instinct MI300X 192GB at about 520 tokens/s for $4.78/hr. Per rental dollar NVIDIA delivers 1.53x the tokens of AMD here.

NVIDIA 51 parts fit

Best value: 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition, 332 tok/s for $2.00/hr

Fewest GPUs: 1 x Rubin SXM, 955 tok/s

AMD 18 parts fit

Best value: 2 x Instinct MI300X 192GB, 520 tok/s for $4.78/hr

Fewest GPUs: 1 x Instinct MI455X, 1.1k tok/s

Huawei 10 parts fit

Fewest GPUs: 2 x Ascend 910C, 316 tok/s

Intel 7 parts fit

Best value: 4 x Gaudi 2 96GB, 376 tok/s for $4.40/hr

Fewest GPUs: 2 x Gaudi 3 128GB, 478 tok/s

Google 4 parts fit

Fewest GPUs: 2 x TPU v7 192GB, 814 tok/s

AWS 2 parts fit

Fewest GPUs: 8 x Inferentia2, 278 tok/s

Other 7 parts fit

Fewest GPUs: 2 x Cloud AI 100 Ultra, 84.4 tok/s

Fleet what-if

Compare whole fleets at these settings: a thousand of one part against a hundred of another. Each fleet splits into replicas of its tensor-parallel width; GPUs left over sit idle.

A GPUs Part

B GPUs Part

Fleet letter	Fleet	Replicas	Idle	Tokens/s	Per hour	Power	Per M tokens
A	1,000 x H200 SXM 141GB	500 x TP2	0	213k tok/s	$2,480	700 kW	$3.23
B	100 x GB200 NVL72 GPU 186GB	50 x TP2	0	44k tok/s	$1,050	—	$6.61

Size Mistral Small 3.2 24B for a tokens per second target

The calculator starts from your traffic instead of a fleet: set a target rate and read the replica count for every GPU.

Open in the LLM GPU Calculator

Frequently asked questions

How much GPU memory does Mistral Small 3.2 24B need?

The weights take 48.0 GB at the native BF16 precision (48.0 GB at BF16, 24.0 GB at FP8, 13.5 GB at INT4). Each concurrent stream adds 160 KB of KV cache per token: 0.7 GB at 4,096 tokens and 5.4 GB at 32,768 tokens.

What is the bare minimum to run Mistral Small 3.2 24B?

1 x Instinct MI355X OAM at BF16 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 1 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $0.50 per hour.

How many H100s do you need to run Mistral Small 3.2 24B?

one H100 at BF16 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is 4 H100s, and 1,000 tokens per second takes 8 H100s across 2 replicas.

How much does Mistral Small 3.2 24B cost per million tokens on an H100?

About $2.98 per million output tokens at BF16, 32,768 tokens of context and 32 streams, using the cheapest live on-demand price of $1.83 per GPU-hour. The cheapest part per token is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $1.68.

What precision does Mistral Small 3.2 24B ship in?

The published checkpoint is BF16. FP8 and INT4 figures describe post-training quantisations that halve and quarter the footprint at a small accuracy cost.

All figures are roofline estimates from the model's config.json and each GPU's published memory, bandwidth and tensor-core peak, with fixed efficiency factors. Prices are the cheapest live on-demand listing per GPU when the site has one. They are a planning floor; a tuned serving stack can do better.