Can I run Gemma 3 27B locally?
Yes. At Q4 with 8K context, Gemma 3 27B needs about 18 GB, so it runs entirely on any GPU with 19 GB of VRAM or more, starting with the AMD Radeon RX 7900 XT.
Memory needed by compression and context length
| Compression | 4K context | 8K context | 32K context | 128K context |
|---|---|---|---|---|
Q4 The most common download. Small quality loss. | 17 GB | 18 GB | 19 GB | 27 GB |
Q5 Very close to full quality. | 20 GB | 20 GB | 22 GB | 30 GB |
Q8 Practically indistinguishable from FP16 at about half the size. | 29 GB | 29 GB | 31 GB | 39 GB |
FP16 Uncompressed. Reference quality, largest size. | 53 GB | 53 GB | 55 GB | 62 GB |
Total memory: model weights, context (KV cache) and about 1 GB of runtime overhead. To run entirely on the GPU, VRAM must be about 5% larger than this.
GPUs that run Gemma 3 27B entirely on the GPU
- Amazon
AMD Radeon RX 7900 XT
20 GB · ~32 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 3090 Ti
24 GB · ~40 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 3090
24 GB · ~37 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 4090
24 GB · ~40 tok/s at Q4
- Amazon
AMD Radeon RX 7900 XTX
24 GB · ~38 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 4090 D
24 GB · ~40 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 5090 D
32 GB · ~68 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 5090
32 GB · ~68 tok/s at Q4
- Amazon
NVIDIA GeForce RTX 5090 D v2
32 GB · — at Q4
Model facts
- Parameters
- 27.4B
- Max context
- 128K tokens
- License
- Gemma Terms of Use
- Released
- Mar 2025
Check it with your hardware
Change the compression, context length and system RAM in the calculator.
Open calculator