Can the GeForce RTX 5070 run Gemma 4 26B-A4B IT?
Only with CPU offload at Q4_K_M.
Gemma 4 26B-A4B IT is 25.8B parameters (MoE; all experts resident). At Q4_K_M the weights are about 14.4 GiB; with the 1.5 GiB runtime reserve that is 15.9 GiB against the card's 12GB, about 3.9 GiB more than the card has.
All three quants
12GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 14.4 GiB | 15.9 GiB | Only with CPU offload | short 3.9 GiB |
| Q8_0near-lossless | 25.5 GiB | 27.0 GiB | Does not fit | short 15.0 GiB |
| FP16full weights | 48.1 GiB | 49.6 GiB | Does not fit | short 37.6 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
Not this card. These fit it comfortably at Q4.
- GeForce RTX 5090 32GB · headroom 16.1 GiB
- GeForce RTX 3090 (used) 24GB · headroom 8.1 GiB
- GeForce RTX 4090 (used) 24GB · headroom 8.1 GiB
- Radeon RX 7900 XTX 24GB · headroom 8.1 GiB
What the GeForce RTX 5070 does run well → · Gemma 4 26B-A4B IT on every card →
About the model
Gemma 4 26B-A4B IT by Google: 262,144-token native context, text + vision, released 2026 under the Apache-2.0. Model card → · Hosted pricing on PicksByModel →