Skip to content

Gemma 4 31B VRAM requirements

How much VRAM Gemma 4 31B needs
4-bit weights: 17.7 GBWith 8K context: 19.9 GBWith 32K context: 20.8 GB8-bit, 8K context: 33.2 GB
Parameters
31.3B
Longest context
256K
Released
Mar 2026
License
Apache 2.0
Strongest at
Chat, Coding, Agents, Docs

How much VRAM Gemma 4 31B needs

Memory Gemma 4 31B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit58.2 GB60.5 GB61.4 GB65.1 GB
8-bit30.9 GB33.2 GB34.1 GB37.8 GB
6-bit23.9 GB26.1 GB27.0 GB30.8 GB
5-bit20.7 GB22.9 GB23.9 GB27.6 GB
4-bit17.7 GB19.9 GB20.8 GB24.6 GB
3-bit14.2 GB16.4 GB17.4 GB21.1 GB
2-bit10.9 GB13.1 GB14.1 GB17.8 GB
1.6-bit6.0 GB8.2 GB9.2 GB12.9 GB

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

GPUs and Macs that run Gemma 4 31B

The best way Gemma 4 31B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Gemma 4 31B on RTX 4060 report
RTX 5060Too bigPDF: Gemma 4 31B on RTX 5060 report
RTX 3070Too bigPDF: Gemma 4 31B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Gemma 4 31B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Gemma 4 31B on RTX 3060 12 GB report
Arc B580Too bigPDF: Gemma 4 31B on Arc B580 report
RTX 4070Too bigPDF: Gemma 4 31B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Gemma 4 31B on RTX 4070 Super report
RTX 5070Too bigPDF: Gemma 4 31B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Gemma 4 31B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Gemma 4 31B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Gemma 4 31B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Gemma 4 31B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Gemma 4 31B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Gemma 4 31B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Gemma 4 31B on RTX 5070 Ti report
RTX 5080Too bigPDF: Gemma 4 31B on RTX 5080 report
Radeon RX 7900 XTX4-bit80K28 tok/s2.9 sPDF: Gemma 4 31B on Radeon RX 7900 XTX report
RTX 30904-bit80K33 tok/s1.9 sPDF: Gemma 4 31B on RTX 3090 report
RTX 40904-bit80K35 tok/s0.8 sPDF: Gemma 4 31B on RTX 4090 report
Radeon AI PRO R97006-bit112K14 tok/s1.9 sPDF: Gemma 4 31B on Radeon AI PRO R9700 report
RTX 50906-bit112K45 tok/s0.6 sPDF: Gemma 4 31B on RTX 5090 report
2× RTX 30908-bit256K19 tok/s1.1 sPDF: Gemma 4 31B on 2× RTX 3090 report
2× RTX 40908-bit256K20 tok/s0.5 sPDF: Gemma 4 31B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K27 tok/s1.0 sPDF: Gemma 4 31B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K20 tok/s0.7 sPDF: Gemma 4 31B on RTX 6000 Ada report
L40S8-bit256K18 tok/s1.2 sPDF: Gemma 4 31B on L40S report
2× RTX 50908-bit256K35 tok/s0.4 sPDF: Gemma 4 31B on 2× RTX 5090 report
4× RTX 30908-bit256K19 tok/s0.6 sPDF: Gemma 4 31B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K35 tok/s0.5 sPDF: Gemma 4 31B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Gemma 4 31B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Gemma 4 31B on Mac mini M4 24 GB report
Mac mini M4 32 GB4-bit20K5.1 tok/s42 sPDF: Gemma 4 31B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Gemma 4 31B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit36K6.5 tok/s21 sPDF: Gemma 4 31B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K6.5 tok/s21 sPDF: Gemma 4 31B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K12 tok/s10 sPDF: Gemma 4 31B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K13 tok/s3.0 sPDF: Gemma 4 31B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K15 tok/s5.9 sPDF: Gemma 4 31B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K15 tok/s5.9 sPDF: Gemma 4 31B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K15 tok/s5.9 sPDF: Gemma 4 31B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K21 tok/s1.8 sPDF: Gemma 4 31B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K4.6 tok/s6.0 sPDF: Gemma 4 31B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K4.6 tok/s6.0 sPDF: Gemma 4 31B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K6.2 tok/s2.1 sPDF: Gemma 4 31B on RTX Spark report
DGX Spark8-bit256K5.7 tok/s2.1 sPDF: Gemma 4 31B on DGX Spark report
Dual DGX Spark8-bit256K5.7 tok/s1.2 sPDF: Gemma 4 31B on Dual DGX Spark report
DGX Station GB3008-bit256K118 tok/s0.2 sPDF: Gemma 4 31B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K35 tok/s0.3 sPDF: Gemma 4 31B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K35 tok/s0.2 sPDF: Gemma 4 31B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K35 tok/s0.1 sPDF: Gemma 4 31B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K40 tok/s1.3 sPDF: Gemma 4 31B on A100 80 GB SXM report
H100 SXM8-bit256K63 tok/s0.4 sPDF: Gemma 4 31B on H100 SXM report
4× H200 NVL8-bit256K86 tok/s0.2 sPDF: Gemma 4 31B on 4× H200 NVL report
8× H200 NVL8-bit256K86 tok/s0.1 sPDF: Gemma 4 31B 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 gemma4:31b
llama.cpp
llama-server -hf unsloth/gemma-4-31B-it-GGUF
vLLM
vllm serve google/gemma-4-31B-it
MLX (Mac)
mlx_lm.chat --model mlx-community/gemma-4-31b-it-4bit

Questions

How much VRAM does Gemma 4 31B need?

Gemma 4 31B needs 17.7 GB for its weights at 4-bit, 19.9 GB with an 8K context and 20.8 GB with 32K. At 8-bit, 33.2 GB.

What is the minimum VRAM for Gemma 4 31B?

About 16.4 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 Radeon RX 7900 XTX.

How big is Gemma 4 31B?

31.3B parameters: 58.2 GB of weights at 16-bit, 17.7 GB at 4-bit.

Can Gemma 4 31B run on an RTX 4090?

Yes. At 4-bit it holds 80K of context and writes about 35 tokens a second.

More models