Can I run Gemma 3 12B on the GeForce RTX 3090?
Runs greatQ4_K_M fits in VRAM with 15.0 GB to spare: Runs great, est. 83.6 tok/s (73.6–93.6, calibrated estimate ±12%).
24 GB VRAM · 936.0 GB/s · FP16 35.6 TFLOPS. Assumes 32 GB of DDR5-5600 system RAM, Windows with this GPU driving the display, 8k context and an f16 KV cache.
Every quant of Gemma 3 12B on the GeForce RTX 3090
| Quant | File size | Verdict | Speed | Memory | Runs as | Notes |
|---|---|---|---|---|---|---|
| Q2_K | 4.77 GB | Runs great | est. 123.5 tok/s108.6–138.3calibrated estimate ±12% | 6.5 / 24.0 GB | Full GPU | — |
| IQ4_XS | 6.55 GB | Runs great | est. 92.5 tok/s81.4–103.5calibrated estimate ±12% | 8.3 / 24.0 GB | Full GPU | — |
| Q4_K_Mbaseline | 7.30 GB | Runs great | est. 83.6 tok/s73.6–93.6calibrated estimate ±12% | 9.0 / 24.0 GB | Full GPU | — |
| Q8_0 | 12.51 GB | Runs great | est. 50.2 tok/s44.2–56.2calibrated estimate ±12% | 14.2 / 24.0 GB | Full GPU | — |
Where the memory goes at Q4_K_M
- Weights
- 7.3 GB
- KV cache
- 0.5 GB
- Compute buffer
- 0.6 GB
- OS reserve
- 0.6 GB
- Free
- 15.0 GB
Context length vs. KV cache
| Context | KV f16 | KV q8_0 | KV q4_0 |
|---|---|---|---|
| 4k | 8.1 GBRuns great est. 86.6 tok/s76.2–97.0calibrated estimate ±12% | 8.0 GBRuns great est. 88.0 tok/s77.5–98.6calibrated estimate ±12% | 7.9 GBRuns great est. 88.8 tok/s78.2–99.5calibrated estimate ±12% |
| 8k | 8.4 GBRuns great est. 83.6 tok/s73.6–93.6calibrated estimate ±12% | 8.2 GBRuns great est. 86.4 tok/s76.0–96.7calibrated estimate ±12% | 8.0 GBRuns great est. 87.9 tok/s77.4–98.5calibrated estimate ±12% |
| 16k | 9.0 GBRuns great est. 78.2 tok/s68.9–87.6calibrated estimate ±12% | 8.5 GBRuns great est. 83.2 tok/s73.2–93.2calibrated estimate ±12% | 8.3 GBRuns great est. 86.1 tok/s75.8–96.5calibrated estimate ±12% |
| 32k | 10.2 GBRuns great est. 69.4 tok/s61.0–77.7calibrated estimate ±12% | 9.2 GBRuns great est. 77.5 tok/s68.2–86.9calibrated estimate ±12% | 8.7 GBRuns great est. 82.8 tok/s72.9–92.7calibrated estimate ±12% |
| 64k | 12.7 GBRuns great est. 56.5 tok/s49.7–63.3calibrated estimate ±12% | 10.7 GBRuns great est. 68.3 tok/s60.1–76.5calibrated estimate ±12% | 9.6 GBRuns great est. 76.9 tok/s67.6–86.1calibrated estimate ±12% |
| 128k | 17.6 GBRuns great est. 41.2 tok/s36.3–46.2calibrated estimate ±12% | 13.6 GBRuns great est. 55.1 tok/s48.5–61.7calibrated estimate ±12% | 11.4 GBRuns great est. 67.2 tok/s59.1–75.3calibrated estimate ±12% |
S = Runs great · A = Runs well · B = Runs slowly · F = Won't run
At Q4_K_M the verdict stays Runs great up to 128k context with an f16 KV cache, and up to 128k with q8_0.
Estimated speed
- Generation
- est. 83.6 tok/s (73.6–93.6, calibrated estimate ±12%)
- Prompt processing
- about 4,984 tok/s (2,990–6,978, rough estimate ±40%)
Measured on this exact combination
No public measurement for this exact combination yet; the numbers above are estimates.
If this is not enough
A larger model that also runs well on the GeForce RTX 3090
Runs greatQwen3.5 35B-A3B: Runs great, est. 204.8 tok/s (163.9–245.8, calibrated estimate ±20%)
A larger quant that still runs great
Runs greatQ8_0 (12.51 GB): Runs great, est. 50.2 tok/s (44.2–56.2, calibrated estimate ±12%)
How to run it
Commands for the Q4_K_M file. Both tools download from Hugging Face on first run.
File: google_gemma-3-12b-it-Q4_K_M.gguf (7.30 GB) from bartowski/google_gemma-3-12b-it-GGUF.
llama-server -hf bartowski/google_gemma-3-12b-it-GGUF:Q4_K_M -c 8192 -ngl all -fa on- -ngl all loads every layer on the GPU.
- -fa on enables flash attention (needed for KV cache quantization).
- --cache-type-k q8_0 --cache-type-v q8_0 shrinks the KV cache to 0.29 GB at 8k context.
PowerShell (quit the Ollama tray app first):
$env:OLLAMA_CONTEXT_LENGTH="8192"; ollama serve
ollama run hf.co/bartowski/google_gemma-3-12b-it-GGUF:Q4_K_M- Ollama defaults to 4k context on GPUs under 24 GB. Set OLLAMA_CONTEXT_LENGTH when starting the server, or /set parameter num_ctx inside the chat.
/set parameter num_ctx 8192
Related pages
Guide: How much VRAM do local LLMs need? 8–32 GB tiers (2026)
Other models on the GeForce RTX 3090
Gemma 3 12B on other hardware
Frequently asked questions
Can I run Gemma 3 12B on the GeForce RTX 3090?
Q4_K_M fits in VRAM with 15.0 GB to spare: Runs great, est. 83.6 tok/s (73.6–93.6, calibrated estimate ±12%). Q4_K_M needs 8.4 GB at 8k context; this setup has 23.4 GB of usable memory and 28.0 GB of free system RAM.
How much context can Gemma 3 12B use on the GeForce RTX 3090?
At Q4_K_M the verdict stays Runs great up to 128k context with an f16 KV cache (8.6 GB of KV) and up to 128k with a q8_0 KV cache (4.6 GB). The model supports up to 128k.
Which quant should I use, and how fast is it?
Q4_K_M (7.30 GB) is the recommended balance: Runs great, est. 83.6 tok/s (73.6–93.6, calibrated estimate ±12%). The largest tracked file that still gets Runs great is Q8_0 (12.51 GB), est. 50.2 tok/s (44.2–56.2, calibrated estimate ±12%).
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.