What GPU do you need to run Gemma 4 26B A4B?
26B params 256K max context VisionTool useReasoning
The smallest GPU memory that runs Gemma 4 26B A4B well is 18 GB (Q4_K_M, e.g. M3 Pro). 47 of 132 GPUs run it well and 25 more are a tight fit. It's a mixture-of-experts model, so all weights must fit in memory but speed depends only on the active experts.
MMLU 82.6
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 256K |
|---|---|---|---|---|
| Q4_K_M | 15.9 GB | 16.2 GB | 16.5 GB | 20.7 GB |
| Q8_0 | 25.0 GB | 25.3 GB | 25.6 GB | 29.9 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run Gemma 4 26B A4B
Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.
As an Amazon Associate, I earn from qualifying purchases.
GPUs that can run Gemma 4 26B A4B
Best quantization that fits on each GPU, with the context and speed it gets. For chips with several memory sizes, the smallest that works is shown.
NVIDIA
- RTX 4060 Ti 16GB 16 GB Q4_K_M 6K context · ~84 tok/s tight fit
- RTX 4070 Ti SUPER 16 GB Q4_K_M 6K context · ~195 tok/s tight fit
- RTX 4080 16 GB Q4_K_M 6K context · ~208 tok/s tight fit
- RTX 4080 SUPER 16 GB Q4_K_M 6K context · ~214 tok/s tight fit
- RTX 5060 Ti 16GB 16 GB Q4_K_M 6K context · ~130 tok/s tight fit
- RTX 5070 Ti 16 GB Q4_K_M 6K context · ~260 tok/s tight fit
- RTX 5080 16 GB Q4_K_M 6K context · ~279 tok/s tight fit
- RTX 2000 Ada 16 GB Q4_K_M 6K context · ~65 tok/s tight fit
- RTX A4000 16 GB Q4_K_M 6K context · ~130 tok/s tight fit
- RTX PRO 2000 Blackwell 16 GB Q4_K_M 6K context · ~84 tok/s tight fit
- Tesla T4 16 GB Q4_K_M 6K context · ~93 tok/s tight fit
- Tesla V100 16GB 16 GB Q4_K_M 6K context · ~262 tok/s tight fit
- RTX 4000 Ada 20 GB Q4_K_M 217K context · ~105 tok/s
- RTX 3090 24 GB Q4_K_M 256K context · ~272 tok/s
- RTX 3090 Ti 24 GB Q4_K_M 256K context · ~293 tok/s
- RTX 4090 24 GB Q4_K_M 256K context · ~293 tok/s
- RTX 4500 Ada 24 GB Q4_K_M 256K context · ~126 tok/s
- RTX A5000 24 GB Q4_K_M 256K context · ~223 tok/s
- RTX PRO 4000 Blackwell 24 GB Q4_K_M 256K context · ~195 tok/s
- RTX Spark 24 GB Q4_K_M 256K context · ~87 tok/s
- Tesla P40 24 GB Q4_K_M 256K context · ~101 tok/s
- RTX 5090 32 GB Q8_0 256K context · ~369 tok/s
- RTX 5000 Ada 32 GB Q8_0 256K context · ~119 tok/s
- RTX PRO 4500 Blackwell 32 GB Q8_0 256K context · ~185 tok/s
- Tesla V100 32GB 32 GB Q8_0 256K context · ~186 tok/s
- L40S 48 GB Q8_0 256K context · ~178 tok/s
- RTX 6000 Ada 48 GB Q8_0 256K context · ~198 tok/s
- RTX A6000 48 GB Q8_0 256K context · ~158 tok/s
- RTX PRO 5000 Blackwell 48 GB Q8_0 256K context · ~277 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 256K context · ~277 tok/s
- RTX PRO 5500 Blackwell 84 GB Q8_0 256K context · ~288 tok/s
- RTX PRO 6000 Blackwell 96 GB Q8_0 256K context · ~369 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 256K context · ~56 tok/s
AMD
- AMD BC-250 16 GB Q4_K_M 6K context · ~130 tok/s tight fit
- Radeon RX 6800 16 GB Q4_K_M 6K context · ~149 tok/s tight fit
- Radeon RX 6800 XT 16 GB Q4_K_M 6K context · ~149 tok/s tight fit
- Radeon RX 6900 XT 16 GB Q4_K_M 6K context · ~149 tok/s tight fit
- Radeon RX 6950 XT 16 GB Q4_K_M 6K context · ~167 tok/s tight fit
- Radeon RX 7600 XT 16 GB Q4_K_M 6K context · ~84 tok/s tight fit
- Radeon RX 7800 XT 16 GB Q4_K_M 6K context · ~181 tok/s tight fit
- Radeon RX 7900 GRE 16 GB Q4_K_M 6K context · ~167 tok/s tight fit
- Radeon RX 9060 XT 16GB 16 GB Q4_K_M 6K context · ~93 tok/s tight fit
- Radeon RX 9070 16 GB Q4_K_M 6K context · ~186 tok/s tight fit
- Radeon RX 9070 XT 16 GB Q4_K_M 6K context · ~186 tok/s tight fit
- Radeon RX 7900 XT 20 GB Q4_K_M 217K context · ~233 tok/s
- Radeon RX 7900 XTX 24 GB Q4_K_M 256K context · ~279 tok/s
- Radeon AI PRO R9700 32 GB Q8_0 256K context · ~132 tok/s
- Ryzen AI Max (Strix Halo) 32 GB Q8_0 256K context · ~53 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 256K context · ~56 tok/s
Intel
- Arc A770 16GB 16 GB Q4_K_M 6K context · ~163 tok/s tight fit
- Arc Pro B65 32 GB Q8_0 256K context · ~125 tok/s
- Arc Pro B70 32 GB Q8_0 256K context · ~125 tok/s
Apple
- M1 16 GB Q4_K_M 6K context · ~20 tok/s tight fit
- M3 Pro 18 GB Q4_K_M 112K context · ~44 tok/s
- M2 24 GB Q4_K_M 256K context · ~29 tok/s
- M3 24 GB Q4_K_M 256K context · ~29 tok/s
- M4 24 GB Q4_K_M 256K context · ~35 tok/s
- M4 Pro 24 GB Q4_K_M 256K context · ~79 tok/s
- M5 24 GB Q4_K_M 256K context · ~44 tok/s
- M5 Pro 24 GB Q4_K_M 256K context · ~89 tok/s
- M6 24 GB Q4_K_M 256K context · ~49 tok/s
- M1 Max 32 GB Q8_0 256K context · ~82 tok/s
- M1 Pro 32 GB Q8_0 256K context · ~41 tok/s
- M2 Max 32 GB Q8_0 256K context · ~82 tok/s
- M2 Pro 32 GB Q8_0 256K context · ~41 tok/s
- M3 Max 36 GB Q8_0 256K context · ~62 tok/s
- M4 Max 36 GB Q8_0 256K context · ~85 tok/s
- M5 Max 36 GB Q8_0 256K context · ~95 tok/s
- M1 Ultra 64 GB Q8_0 256K context · ~165 tok/s
- M2 Ultra 64 GB Q8_0 256K context · ~165 tok/s
- M3 Ultra 96 GB Q8_0 256K context · ~169 tok/s
- M5 Ultra 96 GB Q8_0 256K context · ~247 tok/s