Can I run Gemma 4 31B on an RTX 5090?
Yes: Gemma 4 31B needs about 23.9 GB at Q4_K_M with 32K tokens of context, which fits the 32 GB RTX 5090 with 8.08 GB to spare. At 32K the RTX 5090 holds up to Q6_K (30.8 GB), and Q4_K_M runs up to 120K tokens.
Yes Q4_K_M with 32K tokens of context
- Q4_K_M, 32K
- 23.9 GB
- RTX 5090
- 32 GB, 1,792 GB/s
- To spare
- 8.08 GB
- Tokens/s
- 40–57 tokens/s
At Q4_K_M with 32K tokens of context it writes about 40–57 tokens/s for one request on an RTX 5090.
Best precision for Gemma 4 31B on an RTX 5090
The most precise setting that leaves at least 0.5 GB free; one that fits with less is marked tight.
| Context | Best fit | Memory | Free | Tokens/s |
|---|---|---|---|---|
| 8K | Q6_K | 28.7 GB | 3.25 GB | 34–48 |
| 32K | Q6_K | 30.8 GB | 1.19 GB | 32–44 |
| 128K | Q3_K_M | 28.4 GB | 3.55 GB | 34–48 |
| 256K (full) | Nothing fits | — | — | — |
Gemma 4 31B on the RTX 5090 as the context fills
One request, FP16 KV cache, 0.5 GB plus 10% overhead; a minus sign is memory missing, and tight is less than 0.5 GB free.
| Context | Q4_K_M | Free | Tokens/s | Q8_0 | Free | Tokens/s |
|---|---|---|---|---|---|---|
| 4K | 21.5 GB | 10.5 GB | 45–63 | 36.2 GB | −4.17 GB | — |
| 8K | 21.9 GB | 10.1 GB | 44–62 | 36.5 GB | −4.52 GB | — |
| 16K | 22.5 GB | 9.45 GB | 43–61 | 37.2 GB | −5.20 GB | — |
| 32K | 23.9 GB | 8.08 GB | 40–57 | 38.6 GB | −6.58 GB | — |
| 64K | 26.7 GB | 5.33 GB | 36–51 | 41.3 GB | −9.33 GB | — |
| 128K | 32.2 GB | −176 MB | — | 46.8 GB | −14.8 GB | — |
| 256K (full) | 43.2 GB | −11.2 GB | — | 57.8 GB | −25.8 GB | — |
Gemma 4 31B on more than one RTX 5090
| Cards | Best at 32K | Tokens/s | Q4_K_M longest | Q4_K_M tokens/s | Rent per hour |
|---|---|---|---|---|---|
| 2× (64 GB) | Q8_0 | 45–66 | 256K (full) | 66–102 | $1.38 |
| 4× (128 GB) | BF16 | 49–73 | 256K (full) | 108–180 | $2.76 |
Tensor parallel with the combined bandwidth, as in vLLM; llama.cpp splits layers by default and runs at about one card's speed.
Other options
- The next smaller setting, Q3_K_M, takes 20.2 GB at 32K, 11.8 GB under the RTX 5090; it fits with 0.5 GB to spare up to 163K tokens.
- Renting an RTX 5090 costs about $0.69 an hour (median on getdeploying.com, 2026-09-29): $3.4–4.7 per million tokens at 40–57 tokens/s.
- Smallest setup for Q4_K_M at 32K: RTX 3090 (24 GB).
Run Gemma 4 31B on the RTX 5090 with llama-server
llama-server -hf unsloth/gemma-4-31B-it-GGUF:Q4_K_M -c 122880 -ngl 99 -np 1gemma-4-31B-it-Q4_K_M.gguf, 18.3 GB, from unsloth/
Questions
Can I run Gemma 4 31B on an RTX 5090?
Yes: Gemma 4 31B needs about 23.9 GB at Q4_K_M with 32K tokens of context, which fits the 32 GB RTX 5090 with 8.08 GB to spare. At 32K the RTX 5090 holds up to Q6_K (30.8 GB), and Q4_K_M runs up to 120K tokens.
How fast is Gemma 4 31B on an RTX 5090?
At Q4_K_M with 32K tokens of context it writes about 40–57 tokens/s for one request on an RTX 5090.
What does a second RTX 5090 change for Gemma 4 31B?
Two RTX 5090 cards (64 GB in one tensor-parallel group) hold Gemma 4 31B at Q8_0 with 32K, and Q4_K_M up to 256K (full) tokens, at about 66–102 tokens/s.
Try other settings in the VRAM calculator or the speed calculator. See also Gemma 4 31B VRAM requirements, what LLMs an RTX 5090 can run and every pair, or detect your own GPU. Model data checked .