Can an RTX 3070 run Gemma 4 26B A4B Instruct?
No
Only barely. It loads at IQ2_XXS, which compresses the weights far enough to visibly degrade the model, so the honest answer is that this pairing does not work.
The numbers
Gemma 4 26B A4B Instruct has 25.81 billion parameters across 30 layers. The GeForce RTX 3070 has 8 GB of VRAM at 448 GB/s, of which about 7.2 GB is left for a model once the operating system has taken its share.
| Quantisation | Total needed | Fits | Speed |
|---|---|---|---|
| Q8_0 | 26.55 GB | no | — |
| Q6_K | 20.72 GB | no | — |
| Q5_K_M | 18.1 GB | no | — |
| Q4_K_M | 15.52 GB | no | — |
| IQ4_XS | 13.78 GB | no | — |
| Q3_K_M | 12.76 GB | no | — |
| IQ3_XXS degraded | 10.2 GB | no | — |
| Q2_K degraded | 11.07 GB | no | — |
| IQ2_XXS degraded | 7.2 GB | yes | 57.2 t/s |
1 of the 9 levels fit. Levels marked degraded are listed for completeness, not as advice.
What to do instead
The straightforward answer is a smaller model from the same family. These do fit on a GeForce RTX 3070, at a quantisation worth using:
- Gemma 3 12B Instruct — 12.19B at Q3_K_M, 7.2 GB, about 57.8 tokens/s
- Gemma 4 12B Instruct — 11.96B at IQ4_XS, 7.13 GB, about 58.4 tokens/s
- Gemma 2 9B Instruct — 9.24B at IQ4_XS, 6.7 GB, about 62.4 tokens/s
Shortening the context window will not rescue this one: at 4k it still wants 15.43 GB against 15.52 GB at 8k, because the weights rather than the cache are what fill the card. The gap here is too large for settings to close.
Or the hardware that does run it
The cheapest device we track that runs Gemma 4 26B A4B Instruct at a quantisation worth using is the Arc A770 16GB, around $280 used , running it at IQ4_XS at about 28 tokens per second. Check the current price (affiliate link; indicative price reviewed 2026-09-01). Compare every option.
See how fast it feels
GeForce RTX 3070 running Gemma 4 26B A4B Instruct at IQ2_XXS
YouWhy does my model use more memory when the conversation gets longer?
Because of the KV cache. Every token you send leaves behind a key and a value vector in each layer of the model, and those stay in memory for as long as the conversation lasts.
The weights are a fixed cost: load a 4-bit 8B model and that is about 4.8 GB, whether you write one word or ten thousand. The cache is the part that grows, and it grows in a straight line with the number of tokens in the window.
How steeply depends on the model's attention design. With grouped-query attention, several query heads share one key-value pair, which cuts the cache by that ratio. Without it, every head keeps its own, and a long context can cost more memory than the weights themselves.
If you are short on memory, the first thing to try is lowering the context window in your runtime, not the quantisation.
Reading your question
Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU. The same answer on RTX 4090 at Q6_K would take about 4.4 s.
Keep going
Everything the GeForce RTX 3070 runs, the full breakdown for Gemma 4 26B A4B Instruct, or check a different pairing in the calculator.