# NVIDIA GeForce RTX 5070 for local LLMs

> Fast memory, but 12GB caps it at 14B-class models; the RTX 5060 Ti 16GB holds larger ones.

The RTX 5070 has 12GB of GDDR7 and 672 GB/s of memory bandwidth, which decides how big a model fits and how fast it generates text. It comfortably runs models up to about 15B parameters: the largest in our catalog is Phi-4 at Q3_K_M (9.8GB with 8K context, ~69 tokens/sec). Our top all-round pick for it is Qwen3.5 9B (Q6_K, ~60 tokens/sec).

## Key specs

- Memory: 12GB GDDR7
- Memory bandwidth: 672 GB/s
- Architecture: Blackwell (2025)
- Power: 250W
- Typical price: ~$800 (checked 2026-10)
- Software: CUDA, Vulkan

## Best models for this GPU

- **Best overall (also coding, reasoning, vision, fastest):** Qwen3.5 9B at Q6_K, ~9.5GB with 8K context, ~60 tokens/sec
- **Largest model that fits:** Phi-4 at Q3_K_M, ~9.8GB with 8K context, ~69 tokens/sec
- **Stretch pick (CPU offload):** Qwen3.6 35B-A3B at Q4_K_M, ~23.7GB with 8K context, about 35 tokens/sec with CPU offload

## Compatibility (best quantization that fits, 8K context)

| Model | Params | Quant | Memory | Speed | Verdict |
| --- | --- | --- | --- | --- | --- |
| Gemma 4 E4B | 8B | Q8_0 | 10.5GB | ~54 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q6_K | 9.5GB | ~60 tok/s | Runs well |
| Phi-4 | 14.7B | Q3_K_M | 9.8GB | ~69 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15.0GB | ~47 tok/s | Partial offload |
| Gemma 4 26B A4B | 26B (3.8B active) | Q4_K_M | 17.3GB | ~39 tok/s | Partial offload |
| Qwen3.8 27B | 27B | Q4_K_M | 18.1GB | ~6.0 tok/s | Partial offload |
| Gemma 4 31B | 31B | Q4_K_M | 22.5GB | ~3.8 tok/s | Partial offload |
| Qwen3.6 35B-A3B | 35B (3B active) | Q4_K_M | 23.7GB | ~35 tok/s | Partial offload |
| GLM-4.5-Air | 106B (12B active) | - | - | - | Won't fit |
| Llama 4 Scout | 109B (17B active) | - | - | - | Won't fit |
| gpt-oss-120b | 117B (5.1B active) | - | - | - | Won't fit |
| Qwen3.5 122B-A10B | 122B (10B active) | - | - | - | Won't fit |
| DeepSeek V4 Flash | 284B (13B active) | - | - | - | Won't fit |
| GLM-5.3-Flash | 320B (18B active) | - | - | - | Won't fit |
| GLM-5.3 | 744B (40B active) | - | - | - | Won't fit |

## Pros
- 672 GB/s makes small models very fast
- Efficient 250W board power
- Full CUDA support

## Cons
- 12GB limits model size and context
- Costs more than the RTX 5060 Ti 16GB, which holds larger models

Full page: https://llamabuilds.ai/gpus/rtx-5070 · Best models: https://llamabuilds.ai/gpus/rtx-5070/best-models

Sources: [NVIDIA GeForce RTX 5070 family product page](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5070-family/)

---
Source: https://llamabuilds.ai/gpus/rtx-5070 · Updated October 14, 2026 · Llama Builds. Speeds marked ~ are estimates from our bandwidth model (https://llamabuilds.ai/methodology); measured results are labeled as such.