Skip to content

Gemma 3 12B IT VRAM requirements

How much VRAM Gemma 3 12B IT needs
4-bit weights: 6.9 GBWith 8K context: 8.8 GBWith 32K context: 10.3 GB8-bit, 8K context: 13.9 GB
Parameters
12.2B
Longest context
128K
Released
Mar 2025
License
Gemma
Access
Gated on the Hub
Strongest at
Text and images

Hugging Face asks you to accept its terms before it serves the weights. Its licence (Gemma) is not a standard open-source one: check the terms before commercial use.

How much VRAM Gemma 3 12B IT needs

Memory Gemma 3 12B IT needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit22.7 GB24.6 GB26.1 GB32.1 GB
8-bit12.1 GB13.9 GB15.4 GB21.4 GB
6-bit9.3 GB11.2 GB12.7 GB18.7 GB
5-bit8.1 GB10.0 GB11.5 GB17.5 GB
4-bit6.9 GB8.8 GB10.3 GB16.3 GB
3-bit5.5 GB7.4 GB8.9 GB14.9 GB
2-bit4.3 GB6.1 GB7.6 GB13.6 GB
1.6-bit2.3 GB4.2 GB5.7 GB11.7 GB

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

GPUs and Macs that run Gemma 3 12B IT

The best way Gemma 3 12B IT runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Gemma 3 12B IT on RTX 4060 report
RTX 5060Too bigPDF: Gemma 3 12B IT on RTX 5060 report
RTX 3070Too bigPDF: Gemma 3 12B IT on RTX 3070 report
RTX 3080 10 GB4-bit16K71 tok/s0.8 sPDF: Gemma 3 12B IT on RTX 3080 10 GB report
RTX 3060 12 GB5-bit28K30 tok/s1.8 sPDF: Gemma 3 12B IT on RTX 3060 12 GB report
Arc B5805-bit28K26 tok/s1.2 sPDF: Gemma 3 12B IT on Arc B580 report
RTX 40705-bit28K42 tok/s0.8 sPDF: Gemma 3 12B IT on RTX 4070 report
RTX 4070 Super5-bit28K42 tok/s0.6 sPDF: Gemma 3 12B IT on RTX 4070 Super report
RTX 50705-bit28K55 tok/s0.7 sPDF: Gemma 3 12B IT on RTX 5070 report
RTX 4060 Ti 16 GB8-bit28K17 tok/s1.0 sPDF: Gemma 3 12B IT on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit28K25 tok/s1.0 sPDF: Gemma 3 12B IT on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit28K22 tok/s1.1 sPDF: Gemma 3 12B IT on Arc A770 16 GB report
Radeon RX 9070 XT8-bit28K30 tok/s0.6 sPDF: Gemma 3 12B IT on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit28K37 tok/s0.5 sPDF: Gemma 3 12B IT on RTX 4070 Ti Super report
RTX 4080 Super8-bit28K41 tok/s0.4 sPDF: Gemma 3 12B IT on RTX 4080 Super report
RTX 5070 Ti8-bit28K49 tok/s0.5 sPDF: Gemma 3 12B IT on RTX 5070 Ti report
RTX 50808-bit28K52 tok/s0.4 sPDF: Gemma 3 12B IT on RTX 5080 report
Radeon RX 7900 XTX8-bit128K43 tok/s1.0 sPDF: Gemma 3 12B IT on Radeon RX 7900 XTX report
RTX 30908-bit128K51 tok/s0.6 sPDF: Gemma 3 12B IT on RTX 3090 report
RTX 40908-bit128K55 tok/s0.3 sPDF: Gemma 3 12B IT on RTX 4090 report
Radeon AI PRO R97008-bit128K30 tok/s0.6 sPDF: Gemma 3 12B IT on Radeon AI PRO R9700 report
RTX 50908-bit128K92 tok/s0.2 sPDF: Gemma 3 12B IT on RTX 5090 report
2× RTX 30908-bit128K51 tok/s0.4 sPDF: Gemma 3 12B IT on 2× RTX 3090 report
2× RTX 40908-bit128K55 tok/s0.2 sPDF: Gemma 3 12B IT on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K71 tok/s0.3 sPDF: Gemma 3 12B IT on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K52 tok/s0.2 sPDF: Gemma 3 12B IT on RTX 6000 Ada report
L40S8-bit128K47 tok/s0.4 sPDF: Gemma 3 12B IT on L40S report
2× RTX 50908-bit128K92 tok/s0.1 sPDF: Gemma 3 12B IT on 2× RTX 5090 report
4× RTX 30908-bit128K51 tok/s0.2 sPDF: Gemma 3 12B IT on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K92 tok/s0.2 sPDF: Gemma 3 12B IT on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB4-bit32K14 tok/s14 sPDF: Gemma 3 12B IT on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit28K8.0 tok/s14 sPDF: Gemma 3 12B IT on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit112K8.0 tok/s14 sPDF: Gemma 3 12B IT on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit28K18 tok/s7.2 sPDF: Gemma 3 12B IT on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K18 tok/s7.2 sPDF: Gemma 3 12B IT on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K18 tok/s7.2 sPDF: Gemma 3 12B IT on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K30 tok/s3.6 sPDF: Gemma 3 12B IT on M4 Max 128 GB report
M5 Max 128 GB8-bit128K34 tok/s1.0 sPDF: Gemma 3 12B IT on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K38 tok/s2.0 sPDF: Gemma 3 12B IT on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K38 tok/s2.0 sPDF: Gemma 3 12B IT on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K38 tok/s2.0 sPDF: Gemma 3 12B IT on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K52 tok/s0.6 sPDF: Gemma 3 12B IT on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K13 tok/s2.1 sPDF: Gemma 3 12B IT on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K13 tok/s2.1 sPDF: Gemma 3 12B IT on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K17 tok/s0.7 sPDF: Gemma 3 12B IT on RTX Spark report
DGX Spark8-bit128K16 tok/s0.7 sPDF: Gemma 3 12B IT on DGX Spark report
Dual DGX Spark8-bit128K16 tok/s0.4 sPDF: Gemma 3 12B IT on Dual DGX Spark report
DGX Station GB3008-bit128K261 tok/s0.1 sPDF: Gemma 3 12B IT on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K92 tok/s0.1 sPDF: Gemma 3 12B IT on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K92 tok/s0.1 sPDF: Gemma 3 12B IT on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K92 tok/s0.1 sPDF: Gemma 3 12B IT on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K102 tok/s0.5 sPDF: Gemma 3 12B IT on A100 80 GB SXM report
H100 SXM8-bit128K154 tok/s0.1 sPDF: Gemma 3 12B IT on H100 SXM report
4× H200 NVL8-bit128K201 tok/s0.1 sPDF: Gemma 3 12B IT on 4× H200 NVL report
8× H200 NVL8-bit128K201 tok/s0.1 sPDF: Gemma 3 12B IT 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

llama.cpp
llama-server -hf unsloth/gemma-3-12b-it-GGUF
vLLM
vllm serve google/gemma-3-12b-it

Questions

How much VRAM does Gemma 3 12B IT need?

Gemma 3 12B IT needs 6.9 GB for its weights at 4-bit, 8.8 GB with an 8K context and 10.3 GB with 32K. At 8-bit, 13.9 GB.

What is the minimum VRAM for Gemma 3 12B IT?

About 7.4 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 3080 10 GB.

How big is Gemma 3 12B IT?

12.2B parameters: 22.7 GB of weights at 16-bit, 6.9 GB at 4-bit.

Can Gemma 3 12B IT run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 55 tokens a second.

More models