Can Gemma 4 26B A4B IT run on the Arc A770 16GB?
Yes. At 8K context the best fit is Q3_K_M, using about 13.5 GB of the 15.2 GB usable on the Arc A770 16GB. Estimated speed ceiling: ~277 tokens/sec.
Better quality with offload: 4-bit (Q4_K_M)
Runs with offload 4-bit (Q4_K_M) at 8K context: all attention layers plus the experts of 24 of 30 MoE layers on the Arc A770 16GB (about 13.6 GB), and the experts of the other 6 layers in system RAM (about 2.8 GB). Needs a PC with at least 32 GB of system RAM. Estimated speed: 66–92 tokens/sec (usable for chat), versus a bandwidth ceiling of ~212 tok/s if it fit fully in VRAM.
llama-server -hf bartowski/google_gemma-4-26B-A4B-it-GGUF:Q4_K_M -c 8192 -ngl 999 --n-cpu-moe 6 --no-mmproj
Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, raise --n-cpu-moe by 1-2; if VRAM is left over, lower it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.
+ 1.1 GB with image input. The figures on this page are for text only. With image input enabled, the vision projector (mmproj) file loads alongside the model: Q3_K_M at 8K context then needs about 14.6 GB (tight). llama.cpp's -hf downloads and loads it automatically unless you add --no-mmproj. Calculator with image input on.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | 11.5 GB Fits | 11.6 GB Fits | 12.5 GB Fits | 16.3 GB No | 330 |
| 3-bit (Q3_K_M) • | 13.4 GB Fits | 13.5 GB Fits | 14.5 GB Tight | 18.2 GB No | 277 |
| 4-bit (Q4_K_M) • | 17.1 GB No | 17.3 GB No | 18.2 GB No | 22.0 GB No | — |
| 5-bit (Q5_K_M) • | 19.2 GB No | 19.4 GB No | 20.3 GB No | 24.1 GB No | — |
| 6-bit (Q6_K) • | 22.5 GB No | 22.7 GB No | 23.6 GB No | 27.4 GB No | — |
| 8-bit (Q8_0) • | 26.3 GB No | 26.4 GB No | 27.4 GB No | 31.1 GB No | — |
| 16-bit (F16/BF16) • | 48.3 GB No | 48.4 GB No | 49.4 GB No | 53.1 GB No | — |
• real GGUF file size from bartowski/google_gemma-4-26B-A4B-it-GGUF. Speed is a bandwidth ceiling (560 GB/s), not a benchmark. Methodology.
What fits instead
- Cheapest listed GPU that runs Gemma 4 26B A4B IT: RX 7900 XTX (24 GB, launch MSRP $999) at Q5_K_M, fully in VRAM at 8K context.
- Largest popular model that fits the Arc A770 16GB: gpt-oss 20B (21B, MoE) at Q8_0, about 12.3 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
Gemma 4 26B A4B IT on every GPU and Mac · Every model that runs on the Arc A770 16GB · Other models that fit 16 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (google/gemma-4-26B-A4B-it), Intel specs. Model data last checked 2026-11-18; refreshed hourly.