# AMD Radeon RX 9070 XT for local LLMs

> A solid 16GB RDNA 4 card with improving ROCm support, trailing the RTX 5070 Ti on bandwidth.

The RX 9070 XT has 16GB of GDDR6 and 640 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 Q5_K_M (13GB with 8K context, ~37 tokens/sec). Our top all-round pick for it is Qwen3.8 27B (Q3_K_M, ~30 tokens/sec).

## Key specs

- Memory: 16GB GDDR6
- Memory bandwidth: 640 GB/s
- Architecture: RDNA 4 (2025)
- Power: 304W
- Typical price: ~$800 (checked 2026-10)
- Software: ROCm, Vulkan

## Best models for this GPU

- **Best overall (also coding, reasoning, vision, largest that fits):** Qwen3.8 27B at Q3_K_M, ~14.8GB with 8K context, ~30 tokens/sec
- **Runner-up:** Qwen3.5 9B at Q8_0, ~12.0GB with 8K context, ~36 tokens/sec
- **Fastest good model:** Gemma 4 26B A4B at Q3_K_M, ~14.1GB with 8K context, ~90 tokens/sec

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

| Model | Params | Quant | Memory | Speed | Verdict |
| --- | --- | --- | --- | --- | --- |
| Gemma 4 E4B | 8B | Q8_0 | 10.5GB | ~41 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q8_0 | 12.0GB | ~36 tok/s | Runs well |
| Phi-4 | 14.7B | Q5_K_M | 13.2GB | ~37 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15.0GB | ~75 tok/s | Fits, tight |
| Gemma 4 26B A4B | 26B (3.8B active) | Q3_K_M | 14.1GB | ~90 tok/s | Fits, tight |
| Qwen3.8 27B | 27B | Q3_K_M | 14.8GB | ~30 tok/s | Fits, tight |
| Gemma 4 31B | 31B | Q4_K_M | 22.5GB | ~5.4 tok/s | Partial offload |
| Qwen3.6 35B-A3B | 35B (3B active) | Q4_K_M | 23.7GB | ~41 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
- Cheaper than NVIDIA's comparable 16GB cards
- Much faster AI hardware than RDNA 3
- Good Vulkan and ROCm support in llama.cpp

## Cons
- 640 GB/s trails the RTX 5070 Ti's 896 GB/s
- Some tools remain CUDA-only
- 16GB limits model size

Full page: https://llamabuilds.ai/gpus/rx-9070-xt · Best models: https://llamabuilds.ai/gpus/rx-9070-xt/best-models

Sources: [Wikipedia, Radeon RX 9000 series](https://en.wikipedia.org/wiki/Radeon_RX_9000_series)

---
Source: https://llamabuilds.ai/gpus/rx-9070-xt · Updated October 14, 2026 · Llama Builds. Speeds marked ~ are estimates from our bandwidth model (https://llamabuilds.ai/methodology); measured results are labeled as such.