# Radeon RX 9070 XT

> OnPrem World · https://onpremworld.com/en/hardware/amd-rx-9070-xt · Last updated: 2026-10-14 · CC BY 4.0

16 GB of VRAM at 640 GB/s for $599 at launch, with FP8 support in RDNA 4; models up to about 24B fit at 4-bit.

## Specifications

|  |  |
|---|---|
| Memory | 16 GB GDDR6 |
| Memory bandwidth | 640 GB/s |
| Power | 304 W |
| Launch | 2025-03 |
| Launch price | US$599 |
| Architecture | RDNA 4 |
| Software | ROCm / Vulkan |

## What it can run (16 GB, 8K)

Estimates, not benchmarks. See the methodology.

| Model | Parameters | Best quant | Total | tok/s (est.) |
|---|---|---|---|---|
| Gemma 4 26B A4B IT | 25.2B | Q4_K_M | 15.7 GiB | 194 |
| Mistral Small 3.2 24B Instruct (2506) | 24B | Q4_K_M | 16.0 GiB | 31 |
| Magistral Small 1.2 (2509) | 24B | Q4_K_M | 16.0 GiB | 31 |
| Devstral Small 2 24B Instruct (2512) | 24B | Q4_K_M | 16.0 GiB | 31 |
| gpt-oss-20b | 21B | Q5_K_M | 15.3 GiB | 175 |
| Phi-4 (14B) | 14B | Q6_K | 13.3 GiB | 39 |
| Gemma 3 12B IT | 12B | Q8_0 | 13.8 GiB | 35 |
| Gemma 4 12B Unified IT | 11.95B | Q8_0 | 13.3 GiB | 35 |
| Qwen3.5-9B | 9B | Q8_0 | 10.1 GiB | 47 |
| Qwen3-8B | 8.2B | Q8_0 | 10.1 GiB | 51 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | Q8_0 | 10.1 GiB | 51 |
| Llama 3.1 8B Instruct | 8B | Q8_0 | 9.8 GiB | 53 |
| Gemma 4 E4B IT | 8B | Q8_0 | 9.1 GiB | 53 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 70 |
| Kimi K2.6 | 1000B | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B | Doesn't fit | – | – |
| DeepSeek-V4-Flash | 284B | Doesn't fit | – | – |
| Qwen3.8-Flash-Next | 180B | Doesn't fit | – | – |
| Qwen3.5-122B-A10B | 122B | Doesn't fit | – | – |
| Mistral Small 4 119B (2603) | 119B | Doesn't fit | – | – |
| gpt-oss-120b | 117B | Doesn't fit | – | – |
| Llama 4 Scout 17B-16E Instruct | 109B | Doesn't fit | – | – |
| GLM-4.5-Air | 106B | Doesn't fit | – | – |
| Qwen3-Coder-Next (80B-A3B) | 80B | Doesn't fit | – | – |
| Llama 3.3 70B Instruct | 70B | Doesn't fit | – | – |
| Qwen3.5-35B-A3B | 35B | Doesn't fit | – | – |
| Qwen3.6-35B-A3B | 35B | Doesn't fit | – | – |
| Qwen3-32B | 32.8B | Doesn't fit | – | – |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | Doesn't fit | – | – |
| Gemma 4 31B IT | 30.7B | Doesn't fit | – | – |
| Qwen3-30B-A3B (2507) | 30.5B | Doesn't fit | – | – |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B | Doesn't fit | – | – |
| GLM-4.7-Flash (30B-A3B) | 30B | Doesn't fit | – | – |
| Qwen3.5-27B | 27B | Doesn't fit | – | – |
| Qwen3.6-27B | 27B | Doesn't fit | – | – |
| Qwen3.8-27B | 27B | Doesn't fit | – | – |
| Gemma 3 27B IT | 27B | Doesn't fit | – | – |

## Prices

Latest price at each shop, as listed on the day we checked.

| Region | Shop | Listing | Price | Per GB | Checked |
|---|---|---|---|---|---|
| Taiwan | 原價屋 CoolPC | ｛華擎 RX9070XT Challenger 16GB｝2970MHz/29cm/三風扇/註五年保 (Line on the shop's public price list (no per-product page); list stamped 2026/10/14 15:47) | [NT$29,890](https://www.coolpc.com.tw/evaluate.php) | NT$1,868/GB | 2026-10-14 |
| Taiwan | 欣亞數位 Sinya | 撼訊 Reaper RX 9070 XT 16GB GDDR6 遊蕩者 (std:2970HMz/三風扇/三年保/MIT台灣製/長28.9cm) | [NT$30,990](https://www.sinya.com.tw/prod/209242) | NT$1,937/GB | 2026-10-14 |
| US | Newegg | ASRock Challenger Radeon RX 9070 XT 16GB GDDR6 PCI Express 5.0 x16 Graphics Card RX9070XT CL 16G | [US$879.99](https://www.newegg.com/asrock-challenger-rx9070xt-cl-16g-radeon-rx-9070-xt-16gb-graphics-card-triple-fans/p/N82E16814930145) | US$55/GB | 2026-10-14 |

Taiwan value: about 92.4 tok/s on Llama 3.1 8B Instruct (Q4_K_M, 8K), or 3.1 tok/s per NT$1,000.

US value: about 92.4 tok/s on Llama 3.1 8B Instruct (Q4_K_M, 8K), or 10.5 tok/s per US$100.

All prices: https://onpremworld.com/en/prices.md

## Frequently asked questions

### Which LLMs can the Radeon RX 9070 XT run?

With 16 GB and an 8K context, the largest models that fit include: Gemma 4 26B A4B IT, Mistral Small 3.2 24B Instruct (2506), Magistral Small 1.2 (2509), Devstral Small 2 24B Instruct (2512), gpt-oss-20b. See the full table below.

### How fast is the Radeon RX 9070 XT for local LLMs?

Generation speed is mostly set by memory bandwidth (640 GB/s). For Llama 3.1 8B Instruct at Q8_0, expect roughly 53 tokens/s; bigger models are slower.

### Can the Radeon RX 9070 XT run 70B models?

Llama 3.3 70B Instruct does not fit on the Radeon RX 9070 XT (16 GB), even at Q4_K_M with an 8K context.

## Sources

- https://www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070xt.html
- https://www.amd.com/en/newsroom/press-releases/2025-2-28-amd-unveils-next-generation-amd-rdna-4-architectu.html
- https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html