The Asus Turbo Radeon AI Pro R9700, a 32GB GPU, demonstrates effective performance for local large language model (LLM) and image diffusion tasks, even when installed in older PC hardware. In testing, the card achieved 42.5 tokens per second for text generation using the Qwen 3.8 27B model with MTP (Multi-Token Prediction) enabled. When processing a 69-page FCC filing, the card maintained a speed of 55 tokens per second. For image generation, the card produced a 1024x1024 image in approximately 90 seconds, with subsequent iterations showing improved speed as the model remained loaded in VRAM. The GPU also performed well in gaming, maintaining 70-75 frames per second at 4K resolution on ultra settings in No Man's Sky. While the card is more expensive than some other 32GB options, it offers a more accessible entry point than high-end alternatives. The card requires a 750W power supply and uses a single power connector, necessitating an adapter cable for some older systems.
The Asus Turbo Radeon AI Pro R9700 features 32GB of onboard memory suitable for local AI tasks. The GPU achieved 42.5 tokens per second for text generation with MTP enabled on the Qwen 3.8 27B model.
The card successfully processed a 69-page document at 55 tokens per second during analysis. Image generation tasks were completed in approximately 90 seconds for a 1024x1024 resolution image.
Gaming performance in No Man's Sky at 4K ultra settings reached 70-75 frames per second. The card requires a 750W power supply and may need an adapter cable for older systems.
Chapter guide
Worth noting
- The GPU was provided on loan from Asus for the purpose of this review.
- The reviewer notes that the opinions expressed are their own and the manufacturer did not review or approve the content before publication.