AMD is about to make a significant splash in the AI hardware scene with the upcoming launch of its Instinct MI350 series, scheduled for this week. The MI350 aims to revolutionize the energy efficiency of AI accelerators, with AMD promising up to a 30x improvement by 2025.
This is a huge leap for the company, which has been focused on building a software ecosystem to rival NVIDIA’s CUDA but hasn’t released AI hardware as frequently as its competitor.
At the ISC25 keynote, AMD’s CTO, Mark Papermaster, gave a preview of what’s to come. The MI350 series will feature cutting-edge technology, including a 3nm process node, up to 288 GB of HBM3E memory, and a next-gen “CDNA 4” architecture. These improvements make the MI350 a formidable competitor to NVIDIA’s Blackwell lineup, with the MI350 sitting somewhere between Blackwell and Blackwell Ultra in terms of performance.
The new architecture will likely be accompanied by AMD’s growing ROCm software platform, which has seen significant improvements and is now expected to gain broader adoption by partners. AMD’s efforts in scaling up performance per watt are also notable, with the MI350 expected to bring a massive leap in energy efficiency. With these advancements, AMD hopes to stay ahead of its rivals and deliver on its promises for the future of AI hardware.
Despite the strong competition, AMD seems poised for success in the AI space, and the MI350 release could provide valuable insights into the company’s direction for the coming years. As more details emerge, it will be fascinating to see how the industry reacts to this new chapter in AI hardware development.