AMD · RDNA 3 · 2022
AMD Radeon RX 7900 XTX for local AI
24GB at 960 GB/s with solid ROCm and Vulkan support, though used prices now rival a 3090.
Updated · Speeds marked ~ are estimates; seemethodology
- VRAM
- 24GB
- Bandwidth
- 960 GB/s
- Power
- 355W
- Runs well
- 8 models
RX 7900 XTX
Typically about $1,050 used
Can the RX 7900 XTX run LLMs?
The RX 7900 XTX has 24GB of GDDR6 and 960 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 35B parameters: the largest in our catalog is Qwen3.6 35B-A3B at Q3_K_M (19GB with 8K context, ~125 tokens/sec). Our top all-round pick for it is Qwen3.8 27B (Q4_K_M, ~36 tokens/sec). Llama 3.3 70B Instruct only runs with partial CPU offload, at about 1.8 tokens/sec with CPU offload.
Recommendations
Best models for the RX 7900 XTX
Best overall · also coding, reasoning, vision
Qwen3.8 27BQ4_K_M · 18GB · ~36 tokens/sec
Runs wellRunner-up
Gemma 4 31BQ4_K_M · 22GB · ~30 tokens/sec
Fits, tightFastest good model · also largest that fits
Qwen3.6 35B-A3BQ3_K_M · 19GB · ~125 tokens/sec
Runs well
Compatibility
What runs on the RX 7900 XTX
Every current model in our catalog, at the best quantization that fits in 24GB with 8K tokens of context. "Partial offload" means part of the model runs from system RAM (we assume 64GB), which is much slower.
| Model | Parameters | Best quant | Memory | Speed | Verdict |
|---|---|---|---|---|---|
| Gemma 4 E4B | 8B | Q8_0 | 10 GB | ~60 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q8_0 | 12 GB | ~53 tok/s | Runs well |
| Phi-4 | 14.7B | Q8_0 | 18 GB | ~38 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15 GB | ~99 tok/s | Runs well |
| Gemma 4 26B A4B | 26B (3.8B active) | Q5_K_M | 20 GB | ~91 tok/s | Runs well |
| Qwen3.8 27B | 27B | Q4_K_M | 18 GB | ~36 tok/s | Runs well |
| Gemma 4 31B | 31B | Q4_K_M | 22 GB | ~30 tok/s | Fits, tight |
| Qwen3.6 35B-A3B | 35B (3B active) | Q3_K_M | 19 GB | ~125 tok/s | Runs well |
| GLM-4.5-Air | 106B (12B active) | Q4_K_M | 68 GB | ~8.2 tok/s | Partial offload |
| Llama 4 Scout | 109B (17B active) | Q4_K_M | 70 GB | ~5.8 tok/s | Partial offload |
| gpt-oss-120b | 117B (5.1B active) | MXFP4 | 68 GB | ~20 tok/s | Partial offload |
| Qwen3.5 122B-A10B | 122B (10B active) | — | — | — | Won't fitWhat it needs |
| DeepSeek V4 Flash | 284B (13B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3-Flash | 320B (18B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3 | 744B (40B active) | — | — | — | Won't fitWhat it needs |
Performance
RX 7900 XTX vs. alternatives on Qwen3.8 27B
Generation speed in tokens per second for a single user. Token generation is limited mostly by memory bandwidth, which is why bandwidth predicts these results so well.
RTX 3090
Q4_K_M · 24GB
~39tok/s
RX 7900 XTX
Q4_K_M · 24GB
~36tok/s
RX 9070 XT
Q3_K_M · 16GB
~30tok/s
R9700
Q6_K · 32GB
~18tok/s
Measured
RX 7900 XTX benchmark results
| Model | Quant | Runtime | Generation | Prompt | Source |
|---|---|---|---|---|---|
| Llama 2 7B Chat | Q4_0 | llama.cpp (Vulkan) | 185 tok/s | 3725 tok/s | llama.cpp discussion #10879: Performance of llama.cpp with Vulkan · October 2026 |
| Llama 2 7B Chat | Q4_0 | llama.cpp (ROCm (HIP)) | 165 tok/s | 3550 tok/s | llama.cpp discussion #15021: Performance of llama.cpp on AMD ROCm (HIP) · October 2026 |
| Llama 2 7B Chat | Q4_0 | llama.cpp (ROCm (HIP)) | 155 tok/s | 4145 tok/s | llama.cpp discussion #15021: Performance of llama.cpp on AMD ROCm (HIP) · October 2026 |
| 2× · Gemma 4 31B | Q8_0 | llama.cpp (ROCm (HIP)) | 32 tok/s | 1115 tok/s | llama.cpp discussion #15021: Performance of llama.cpp on AMD ROCm (HIP) · October 2026 |
| 2× · Gemma 4 31B | Q8_0 | llama.cpp (Vulkan) | 24 tok/s | 850 tok/s | llama.cpp discussion #10879: Performance of llama.cpp with Vulkan · October 2026 |
| 2× · Gemma 4 31B | Q8_0 | llama.cpp (ROCm (HIP)) | 23 tok/s | 920 tok/s | llama.cpp discussion #15021: Performance of llama.cpp on AMD ROCm (HIP) · October 2026 |
Our take
Should you buy the RX 7900 XTX for AI?
The Radeon RX 7900 XTX has 24GB of GDDR6 on a 384-bit bus for 960 GB/s, which puts its raw bandwidth just ahead of the RTX 3090. ROCm officially supports it, and llama.cpp's Vulkan backend runs well without extra setup. Prompt processing and some tools still lag CUDA, so check that your software supports AMD before buying. Used cards sold for about $1,050 in October 2026, above the $999 launch price and similar to a used RTX 3090.
What we like
- 24GB with 960 GB/s bandwidth
- Official ROCm support and strong Vulkan performance
- Open-source Linux drivers
Watch out for
- Slower prompt processing than comparable NVIDIA cards
- Some tools and quantization formats are CUDA-only
- Used prices above launch MSRP
Best for: Linux users comfortable with ROCm or Vulkan · 24GB builds for 32B models
Specifications
RX 7900 XTX specs that matter for AI
- Memory
- 24GB GDDR6
- Memory bandwidth
- 960 GB/s
- Memory bus
- 384-bit
- Architecture
- RDNA 3
- Power
- 355W
- Interface
- PCIe 4.0 x16
- Released
- 2022
- Launch price
- $999
- Street price (used)
- ~$1,050 (2026-10)
- Software
- ROCm, Vulkan
Sources: Wikipedia, Radeon RX 7000 series, CPUTronic RX 7900 XTX specifications
Cross-shop
Alternatives to the RX 7900 XTX
Frequently asked questions
How much VRAM does the RX 7900 XTX have?
The AMD Radeon RX 7900 XTX has 24GB of GDDR6 with 960 GB/s of memory bandwidth on a 384-bit bus.
What is the largest LLM the RX 7900 XTX can run?
Fully in GPU memory, the largest model in our catalog that runs comfortably is Qwen3.6 35B-A3B (35B (3B active)) at Q3_K_M, using about 19GB with 8K context at ~125 tokens/sec.
Can the RX 7900 XTX run Llama 3.3 70B?
Only with partial CPU offload. At Q4_K_M the model needs about 47GB with 8K context, so part of it spills into system RAM and speed drops to about 1.8 tokens/sec with CPU offload. For usable 70B speeds you want 48GB of VRAM or more (for example two 24GB cards).
What is the best LLM to run on the RX 7900 XTX?
Our top pick is Qwen3.8 27B at Q4_K_M, which fits in 24GB with room for 8K of context and runs at ~36 tokens/sec. See our full list of the best models for the RX 7900 XTX for coding, reasoning, and vision picks.
Is the RX 7900 XTX good for local AI in 2026?
24GB at 960 GB/s with solid ROCm and Vulkan support, though used prices now rival a 3090.