How much memory does Gemma 4 31B need at Q4?

The Q4 figure used here is 19.7 GB. GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.

What is the cheapest catalog device that runs Gemma 4 31B at Q4?

Radeon RX 7900 XTX, list price $999.

How fast is Gemma 4 31B on the GeForce RTX 3090, as a bandwidth estimate?

About 51.1 tok/s at Q4. That is device bandwidth divided by active weight bytes, not a measured run.

Yes. Gemma 4 31B fits on the GeForce RTX 3090 at Q4, with 3.1 GB of headroom. Usable memory is 22.8 GB.

Can the GeForce RTX 3090 run Gemma 4 31B?

Data updated 16 October 2026. Last reviewed the same day, when this catalog snapshot was written.

Usable memory on this 24 GB VRAM config is 22.8 GB (95% of the sticker figure). A result under 2 GB of headroom is labeled tight.

Quant ladder

Estimated memory for Gemma 4 31B on GeForce RTX 3090
Quant	Estimate	Headroom	Est. tok/s	Verdict
Q4	19.7 GB	3.1 GB	51.1 tok/s	Yes
Q8	34.0 GB	-11.2 GB	28.7 tok/s	No
FP16	75.1 GB	-52.3 GB	15.0 tok/s	No

GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.
Formula estimate: parameter count × bytes per weight × 1.20. No GGUF file size is stored for this quant.
Tokens/sec is a bandwidth roofline: GB/s divided by active weight GB. It is not a measured generation speed.

KV cache at 8K and 32K

FP16 KV-cache estimate: 1.3 GB at 8,192 tokens and 5.4 GB at 32,768 tokens. GQA counts two tensors (K and V) for the stored KV heads and head dimension. config.json also sets a sliding window of 1,024 tokens. The figures above assume every counted layer stores the full context, so they can overstate cache on sliding layers.

Offload

The Q4 weight estimate fits in usable VRAM, so this page does not assume a system-RAM offload.

Ollama

This command pulls the Q4_K_M file from unsloth/gemma-4-31B-it-GGUF, the Hub repo whose file size is used above. It is not an official Ollama library tag.

ollama run hf.co/unsloth/gemma-4-31B-it-GGUF:Q4_K_M

Other picks

Cheapest catalog device at Q4: Radeon RX 7900 XTX
Closest catalog model that still fits at Q4: Qwen3.6 35B-A3B.

All hardware for Gemma 4 31B · All models for the GeForce RTX 3090

Questions

How much memory does Gemma 4 31B need at Q4?: The Q4 figure used here is 19.7 GB. GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.
Can the GeForce RTX 3090 run Gemma 4 31B?: Yes. Gemma 4 31B fits on the GeForce RTX 3090 at Q4, with 3.1 GB of headroom. Usable memory is 22.8 GB.
What is the cheapest catalog device that runs Gemma 4 31B at Q4?: Radeon RX 7900 XTX, list price $999.
How fast is Gemma 4 31B on the GeForce RTX 3090, as a bandwidth estimate?: About 51.1 tok/s at Q4. That is device bandwidth divided by active weight bytes, not a measured run.