Skip to content

Devstral Small 2 24B VRAM requirements

How much VRAM Devstral Small 2 24B needs
4-bit weights: 13.6 GBWith 8K context: 15.7 GBWith 32K context: 19.4 GB
Parameters
24B
Longest context
384K
Released
Dec 2025
License
Apache 2.0
Strongest at
Coding

How much VRAM Devstral Small 2 24B needs

Memory Devstral Small 2 24B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP8)24.0 GB26.1 GB29.9 GB44.9 GB
6-bit18.3 GB20.4 GB24.2 GB39.2 GB
5-bit15.9 GB18.0 GB21.8 GB36.8 GB
4-bit13.6 GB15.7 GB19.4 GB34.4 GB
3-bit10.9 GB13.0 GB16.8 GB31.8 GB
2-bit8.4 GB10.5 GB14.2 GB29.2 GB
1.6-bit4.6 GB6.7 GB10.5 GB25.5 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Devstral Small 2 24B

The best way Devstral Small 2 24B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Devstral Small 2 24B on RTX 4060 report
RTX 5060Too bigPDF: Devstral Small 2 24B on RTX 5060 report
RTX 3070Too bigPDF: Devstral Small 2 24B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Devstral Small 2 24B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Devstral Small 2 24B on RTX 3060 12 GB report
Arc B580Too bigPDF: Devstral Small 2 24B on Arc B580 report
RTX 4070Too bigPDF: Devstral Small 2 24B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Devstral Small 2 24B on RTX 4070 Super report
RTX 5070Too bigPDF: Devstral Small 2 24B on RTX 5070 report
RTX 4060 Ti 16 GB3-bit20K17 tok/s2.3 sPDF: Devstral Small 2 24B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB3-bit20K27 tok/s2.1 sPDF: Devstral Small 2 24B on RTX 5060 Ti 16 GB report
Arc A770 16 GB3-bit20K23 tok/s2.4 sPDF: Devstral Small 2 24B on Arc A770 16 GB report
Radeon RX 9070 XT3-bit20K32 tok/s1.4 sPDF: Devstral Small 2 24B on Radeon RX 9070 XT report
RTX 4070 Ti Super3-bit20K39 tok/s1.1 sPDF: Devstral Small 2 24B on RTX 4070 Ti Super report
RTX 4080 Super3-bit20K43 tok/s1.0 sPDF: Devstral Small 2 24B on RTX 4080 Super report
RTX 5070 Ti3-bit20K52 tok/s1.1 sPDF: Devstral Small 2 24B on RTX 5070 Ti report
RTX 50803-bit20K55 tok/s0.9 sPDF: Devstral Small 2 24B on RTX 5080 report
Radeon RX 7900 XTX6-bit20K28 tok/s2.2 sPDF: Devstral Small 2 24B on Radeon RX 7900 XTX report
RTX 30906-bit20K33 tok/s1.4 sPDF: Devstral Small 2 24B on RTX 3090 report
RTX 40906-bit20K35 tok/s0.6 sPDF: Devstral Small 2 24B on RTX 4090 report
Radeon AI PRO R9700FP832K15 tok/s1.4 sPDF: Devstral Small 2 24B on Radeon AI PRO R9700 report
RTX 5090FP832K47 tok/s0.5 sPDF: Devstral Small 2 24B on RTX 5090 report
2× RTX 3090FP8128K25 tok/s0.8 sPDF: Devstral Small 2 24B on 2× RTX 3090 report
2× RTX 4090FP8128K27 tok/s0.3 sPDF: Devstral Small 2 24B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBFP8128K36 tok/s0.7 sPDF: Devstral Small 2 24B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaFP8128K26 tok/s0.5 sPDF: Devstral Small 2 24B on RTX 6000 Ada report
L40SFP8128K23 tok/s0.9 sPDF: Devstral Small 2 24B on L40S report
2× RTX 5090FP8224K47 tok/s0.3 sPDF: Devstral Small 2 24B on 2× RTX 5090 report
4× RTX 3090FP8384K25 tok/s0.4 sPDF: Devstral Small 2 24B on 4× RTX 3090 report
RTX PRO 6000 BlackwellFP8384K47 tok/s0.4 sPDF: Devstral Small 2 24B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Devstral Small 2 24B on Mac mini M4 16 GB report
Mac mini M4 24 GB3-bit20K8.4 tok/s32 sPDF: Devstral Small 2 24B on Mac mini M4 24 GB report
Mac mini M4 32 GB5-bit20K5.8 tok/s32 sPDF: Devstral Small 2 24B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB3-bit20K18 tok/s16 sPDF: Devstral Small 2 24B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBFP860K8.7 tok/s16 sPDF: Devstral Small 2 24B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBFP8128K8.7 tok/s16 sPDF: Devstral Small 2 24B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBFP8384K16 tok/s7.9 sPDF: Devstral Small 2 24B on M4 Max 128 GB report
M5 Max 128 GBFP8384K17 tok/s2.3 sPDF: Devstral Small 2 24B on M5 Max 128 GB report
M3 Ultra 96 GBFP8256K20 tok/s4.4 sPDF: Devstral Small 2 24B on M3 Ultra 96 GB report
M3 Ultra 256 GBFP8384K20 tok/s4.4 sPDF: Devstral Small 2 24B on M3 Ultra 256 GB report
M3 Ultra 512 GBFP8384K20 tok/s4.4 sPDF: Devstral Small 2 24B on M3 Ultra 512 GB report
M5 Ultra 512 GBFP8384K28 tok/s1.4 sPDF: Devstral Small 2 24B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBFP8128K6.1 tok/s4.5 sPDF: Devstral Small 2 24B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBFP8384K6.1 tok/s4.5 sPDF: Devstral Small 2 24B on Ryzen AI Max+ 395 128 GB report
RTX SparkFP8384K8.3 tok/s1.6 sPDF: Devstral Small 2 24B on RTX Spark report
DGX SparkFP8384K7.6 tok/s1.6 sPDF: Devstral Small 2 24B on DGX Spark report
Dual DGX SparkFP8384K7.6 tok/s0.9 sPDF: Devstral Small 2 24B on Dual DGX Spark report
DGX Station GB300FP8384K160 tok/s0.1 sPDF: Devstral Small 2 24B on DGX Station GB300 report
2× RTX PRO 6000 BlackwellFP8384K47 tok/s0.2 sPDF: Devstral Small 2 24B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 BlackwellFP8384K47 tok/s0.1 sPDF: Devstral Small 2 24B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 BlackwellFP8384K47 tok/s0.1 sPDF: Devstral Small 2 24B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMFP8320K53 tok/s1.0 sPDF: Devstral Small 2 24B on A100 80 GB SXM report
H100 SXMFP8320K84 tok/s0.3 sPDF: Devstral Small 2 24B on H100 SXM report
4× H200 NVLFP8384K115 tok/s0.1 sPDF: Devstral Small 2 24B on 4× H200 NVL report
8× H200 NVLFP8384K115 tok/s0.1 sPDF: Devstral Small 2 24B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

Ollama
ollama run devstral-small-2:24b
llama.cpp
llama-server -hf unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF
vLLM
vllm serve mistralai/Devstral-Small-2-24B-Instruct-2512
MLX (Mac)
mlx_lm.chat --model mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit

Questions

How much VRAM does Devstral Small 2 24B need?

Devstral Small 2 24B needs 13.6 GB for its weights at 4-bit, 15.7 GB with an 8K context and 19.4 GB with 32K.

What is the minimum VRAM for Devstral Small 2 24B?

About 13.0 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060 Ti 16 GB.

How big is Devstral Small 2 24B?

24B parameters: 24.0 GB of weights as released (FP8), 13.6 GB at 4-bit.

Can Devstral Small 2 24B run on an RTX 4090?

Yes. At 6-bit it holds 20K of context and writes about 35 tokens a second.

More models