Skip to content

Ministral 3 8B VRAM requirements

How much VRAM Ministral 3 8B needs
4-bit weights: 5.0 GBWith 8K context: 6.9 GBWith 32K context: 10.1 GB8-bit, 8K context: 10.7 GB
Parameters
8.9B
Longest context
256K
Released
Dec 2025
License
Apache 2.0
Strongest at
Docs

How much VRAM Ministral 3 8B needs

Memory Ministral 3 8B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP8)9.7 GB11.6 GB14.8 GB27.5 GB
8-bit8.8 GB10.7 GB13.9 GB26.7 GB
6-bit6.8 GB8.7 GB11.9 GB24.6 GB
5-bit5.9 GB7.8 GB11.0 GB23.7 GB
4-bit5.0 GB6.9 GB10.1 GB22.9 GB
3-bit4.1 GB5.9 GB9.1 GB21.9 GB
2-bit3.1 GB5.0 GB8.2 GB20.9 GB
1.6-bit1.7 GB3.6 GB6.8 GB19.5 GB

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

GPUs and Macs that run Ministral 3 8B

The best way Ministral 3 8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit20K44 tok/s1.2 sPDF: Ministral 3 8B on RTX 4060 report
RTX 50603-bit20K71 tok/s0.9 sPDF: Ministral 3 8B on RTX 5060 report
RTX 30703-bit20K71 tok/s0.9 sPDF: Ministral 3 8B on RTX 3070 report
RTX 3080 10 GB5-bit20K82 tok/s0.6 sPDF: Ministral 3 8B on RTX 3080 10 GB report
RTX 3060 12 GB6-bit28K35 tok/s1.4 sPDF: Ministral 3 8B on RTX 3060 12 GB report
Arc B5806-bit28K31 tok/s1.0 sPDF: Ministral 3 8B on Arc B580 report
RTX 40706-bit28K49 tok/s0.6 sPDF: Ministral 3 8B on RTX 4070 report
RTX 4070 Super6-bit28K49 tok/s0.5 sPDF: Ministral 3 8B on RTX 4070 Super report
RTX 50706-bit28K64 tok/s0.6 sPDF: Ministral 3 8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit40K22 tok/s0.8 sPDF: Ministral 3 8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit40K34 tok/s0.7 sPDF: Ministral 3 8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit40K30 tok/s0.8 sPDF: Ministral 3 8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit40K40 tok/s0.5 sPDF: Ministral 3 8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit40K50 tok/s0.4 sPDF: Ministral 3 8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit40K55 tok/s0.3 sPDF: Ministral 3 8B on RTX 4080 Super report
RTX 5070 Ti8-bit40K66 tok/s0.4 sPDF: Ministral 3 8B on RTX 5070 Ti report
RTX 50808-bit40K70 tok/s0.3 sPDF: Ministral 3 8B on RTX 5080 report
Radeon RX 7900 XTX8-bit96K58 tok/s0.8 sPDF: Ministral 3 8B on Radeon RX 7900 XTX report
RTX 30908-bit96K68 tok/s0.5 sPDF: Ministral 3 8B on RTX 3090 report
RTX 40908-bit96K73 tok/s0.2 sPDF: Ministral 3 8B on RTX 4090 report
Radeon AI PRO R97008-bit152K40 tok/s0.5 sPDF: Ministral 3 8B on Radeon AI PRO R9700 report
RTX 50908-bit152K123 tok/s0.2 sPDF: Ministral 3 8B on RTX 5090 report
2× RTX 30908-bit256K68 tok/s0.3 sPDF: Ministral 3 8B on 2× RTX 3090 report
2× RTX 40908-bit256K73 tok/s0.1 sPDF: Ministral 3 8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K95 tok/s0.3 sPDF: Ministral 3 8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K70 tok/s0.2 sPDF: Ministral 3 8B on RTX 6000 Ada report
L40S8-bit256K64 tok/s0.3 sPDF: Ministral 3 8B on L40S report
2× RTX 50908-bit256K123 tok/s0.1 sPDF: Ministral 3 8B on 2× RTX 5090 report
4× RTX 30908-bit256K68 tok/s0.2 sPDF: Ministral 3 8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K123 tok/s0.1 sPDF: Ministral 3 8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit20K14 tok/s11 sPDF: Ministral 3 8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit40K11 tok/s11 sPDF: Ministral 3 8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit80K11 tok/s11 sPDF: Ministral 3 8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit40K24 tok/s5.6 sPDF: Ministral 3 8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit184K24 tok/s5.6 sPDF: Ministral 3 8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K24 tok/s5.6 sPDF: Ministral 3 8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K41 tok/s2.8 sPDF: Ministral 3 8B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K45 tok/s0.8 sPDF: Ministral 3 8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K51 tok/s1.6 sPDF: Ministral 3 8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K51 tok/s1.6 sPDF: Ministral 3 8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K51 tok/s1.6 sPDF: Ministral 3 8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K71 tok/s0.5 sPDF: Ministral 3 8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K17 tok/s1.6 sPDF: Ministral 3 8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K17 tok/s1.6 sPDF: Ministral 3 8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K23 tok/s0.6 sPDF: Ministral 3 8B on RTX Spark report
DGX Spark8-bit256K21 tok/s0.6 sPDF: Ministral 3 8B on DGX Spark report
Dual DGX Spark8-bit256K21 tok/s0.3 sPDF: Ministral 3 8B on Dual DGX Spark report
DGX Station GB3008-bit256K356 tok/s0.1 sPDF: Ministral 3 8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K123 tok/s0.1 sPDF: Ministral 3 8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K123 tok/s0.1 sPDF: Ministral 3 8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K123 tok/s0.1 sPDF: Ministral 3 8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K138 tok/s0.4 sPDF: Ministral 3 8B on A100 80 GB SXM report
H100 SXM8-bit256K208 tok/s0.1 sPDF: Ministral 3 8B on H100 SXM report
4× H200 NVL8-bit256K272 tok/s0.1 sPDF: Ministral 3 8B on 4× H200 NVL report
8× H200 NVL8-bit256K272 tok/s0.1 sPDF: Ministral 3 8B 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 ministral-3:8b
llama.cpp
llama-server -hf unsloth/Ministral-3-8B-Instruct-2512-GGUF
vLLM
vllm serve mistralai/Ministral-3-8B-Instruct-2512
MLX (Mac)
mlx_lm.chat --model mlx-community/Ministral-3-8B-Instruct-2512-4bit

Questions

How much VRAM does Ministral 3 8B need?

Ministral 3 8B needs 5.0 GB for its weights at 4-bit, 6.9 GB with an 8K context and 10.1 GB with 32K. At 8-bit, 10.7 GB.

What is the minimum VRAM for Ministral 3 8B?

About 5.9 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.

How big is Ministral 3 8B?

8.9B parameters: 9.7 GB of weights as released (FP8), 5.0 GB at 4-bit.

Can Ministral 3 8B run on an RTX 4090?

Yes. At 8-bit it holds 96K of context and writes about 73 tokens a second.

More models