AMD Reveals AI Performance of RX 9070 XT vs RX 7900 on Windows 11

AMD officially announced the new generation of RX 9070 series desktop graphics cards based on RDNA 4 architecture today. In addition to improvements related to gaming performance, “Team Red” also did not forget to emphasize major advances in AI in this important product launch.

AMD says it has added new dedicated math pipelines to accelerate machine learning (ML), optimized for better performance with narrower data types. The new GPUs also support FP8 (8-bit floating point), including BF8 (Brain Float 8)/FP8 E4M3 and HF8 (half precision float)/FP8 E5M2, to improve inferencing performance.

To take advantage of these improvements, the WMMA (Wave Matrix Multiply Accumulate) instructions have been upgraded on the 9070 series cards. If you didn't know, AMD added WMMA support for the first time with the RDNA 3 architecture.

AMD promises 2x FP16 performance on RDNA 4 compared to RDNA 3 (RX 7000), and 4x in the case of INT8; and doubles these numbers when using Sparsity.

For AI visualization using Stable Diffusion, the new RDNA 4 graphics cards offer AMD-optimized models that promise superior performance thanks to improved memory utilization. The company also expects to support FLUX AI and large language models (LLMs) in the future.

For real-world productivity applications like Adobe Lightroom and Da Vinci Resolve, AMD promises 34% better performance from the RX 9070 XT than the RX 7900 GRE. Meanwhile, for generative AI, the company expects the RX 9070 XT to be 70% faster than the RX 7900 GRE. Note that all of these tests were performed on Windows 11.

“With great power comes great responsibility,” and AMD seems to be living up to that promise. Using all the AI ​​power in its arsenal, the company is trying to further improve display driver stability with the new AMD AI Inspector feature. Additionally, AMD is finally bringing ML-based upscaling with FSR 4 and other frame generation technologies.