AMD attacks the rack with Helios systems that rival Nvidia's (www.theregister.com)

AMD has unveiled Helios, its first dedicated rack-scale AI platform, which claims to rival Nvidia's dominance in the data center market. The system boasts a formidable 72-GPU configuration and is engineered to deliver superior performance metrics, including 50% more HBM4 memory and 15-25% higher performance for AI training compared to competing solutions. The Helios architecture leverages AMD's innovative CDNA 5 compute architecture with the new MI455X GPUs, built using advanced 2nm and 3nm process technologies, allowing for significant improvements in floating-point performance and memory bandwidth. This launch signifies a pivotal moment for the AI/ML community, as Helios could challenge Nvidia’s established status quo. The transition to a single 96-core CPU per blade and the adoption of Ultra Accelerator Link over Ethernet (UALoE) for networking present a promising approach that could enhance scalability and efficiency. Moreover, AMD's focus on real-world performance metrics rather than just peak FLOPS indicates a strategic shift aimed at addressing the practical needs of AI workloads, making Helios an attractive option for major players including Meta and Microsoft. As the competition heats up, AMD aims to carve out a substantial share of the AI accelerator market with its cutting-edge technology.