AMD demonstrated the latest Instinct MI300X GPU at the data center and AI technology premiere held on Tuesday. AMD did not disclose too many details in the keynote speech, and Hoang Anh Phu later found that the MI300X (192 GB HBM3, OAM module) TBP is 750 W, while the previous generation MI250X TBP is only 500-560 W.
It was previously reported that the MI300X is a pure GPU version that uses AMD CDNA 3 technology and uses up to 192 GB of HBM3 high-bandwidth memory to accelerate large language models and generative AI calculations.
Designed for large language models and other advanced AI models, the MI300X and its CDNA architecture pack together 12 5nm chiplets for a total of 153 billion transistors.
This new AI chip ditches the APU’s 24 Zen cores and I/O chips in favor of more CDNA 3 GPUs and a larger 192GB HBM3, delivering 5.2 TB/s of memory bandwidth and 896GB/s of unlimited bandwidth .