Gemma 4 31B VRAM requirements

Can I run Gemma 4 31B on…

• RTX 4090: Q4_K_M, 26K context
• RTX 3090: Q4_K_M, 26K context
• RTX 5090: Yes, up to Q6_K
• RTX 3060 12GB: No, 4 cards
• RTX 4060 Ti 16GB: No, 2 cards
• RTX 5060 Ti 16GB: No, 2 cards
• RTX 4070 12GB: No, 4 cards
• RTX 5080 16GB: No, 2 cards
• RTX 4080 Super 16GB: No, 2 cards
• RX 7900 XTX: Q4_K_M, 26K context
• H100 SXM: Yes, up to BF16

How much VRAM does Gemma 4 31B need?

Gemma 4 31B needs 21.9 GB at Q4_K_M, 36.5 GB at Q8_0 and 66.6 GB at FP16 with 8,192 tokens of context and one request. Each total below is the weights plus the FP16 KV cache and the runtime overhead (0.5 GB plus 10%), and each row opens the calculator with that setting. What the GGUF names mean.

Fine-tuning Gemma 4 31B? Gemma 4 31B VRAM for LoRA, QLoRA and full training.

Gemma 4 31B GGUF files on Hugging Face

The 8 GGUF files llama.cpp picks for Gemma 4 31B with -hf <repo>:<QUANT>, with the bytes the Hugging Face file list reported on 2026-09-29. Each total is that file plus the FP16 KV cache and 0.5 GB + 10%, one request. "vs estimate" compares the file with the bits-per-weight size the table above uses.

GB here is GiB (1024³ bytes), as everywhere on this site; Hugging Face shows the Q4_K_M file as 18.3 GB, in 10⁹ bytes.

Which Gemma 4 31B GGUF fits a 16, 24, 32, 48 or 80 GB GPU or a Mac

From the file sizes above: the largest file that fits each machine with an 8K context and 0.5 GB left free, the longest context it then has room for, and the longest context for Q4_K_M (17.1 GB). One request, FP16 KV cache, -np 1. The Q4_K_M file is 569 MB under its bits-per-weight estimate; the estimate tables and the llama-server command on this page count the estimate, so they show less context.

Run Gemma 4 31B with llama.cpp or Ollama

llama-server -hf unsloth/gemma-4-31B-it-GGUF:Q4_K_M -c 26624 -np 1

gemma-4-31B-it-Q4_K_M.gguf, 18.3 GB, from unsloth/gemma-4-31B-it-GGUF (checked 2026-09-29). On a 24 GB RTX 3090 / 4090, Q4_K_M with -c 26624 takes 23.4 GB and leaves 608 MB. -np 1: one slot, so one 1,024-token window.

ollama run gemma4:31b

Ollama library tag gemma4:31b; its quant is Ollama's choice. Same context: /set parameter num_ctx 26624.

Gemma 4 31B at 8K, 32K, 128K and 256K (full) tokens of context

10 of its 60 layers use full attention and 50 keep a sliding window of 1,024 tokens. Each extra token of context adds 80 KB of FP16 cache per request once the sliding windows are full. The last column is the smallest setup here, one card or a group, that holds it at Q4_K_M. How the KV cache works.

Speculative decoding with Gemma 4 31B

Extra memory a draft model or MTP head adds to the Unsloth Q4_K_M GGUF (21.7 GB alone with 32K tokens), from the measured files. Open the speculative decoding calculator.

How fast Gemma 4 31B writes

Tokens per second for one request with 8,192 tokens of context, estimated from memory bandwidth. A dash means it does not fit on one card. Try other settings in the speed calculator.

Longest context on one GPU

How many tokens of context Gemma 4 31B fits on a single card with one request, an FP16 KV cache and 0.5 GB left free. "Full" means the model's whole context window fits.

Model details

Parameters

31.3B (31,273,088,876)

Layers

10 of its 60 layers use full attention and 50 keep a sliding window of 1,024 tokens

Attention cache

4 KV heads × 512 (values come from the key projection but are cached separately)

Sliding-window layers

16 KV heads × 256

Context length

262,144 tokens

Published weights

58.3 GB (BF16)

On Hugging Face

google/gemma-4-31B-it

Reported benchmarks

GPQA Diamond 84.3 · MMLU-Pro 85.2 · LiveCodeBench v6 80.0. Vendor-reported: from the maker’s own model card, not rerun here; read 2026-09-30.

Gemma 4 31B next to similar-sized models

The three models here closest to it in parameter count, at Q4_K_M with 8,192 tokens of context.