SoftBank Corp. and AMD have begun a joint validation to utilize AMD Instinct GPUs in enhancing next-generation AI infrastructure computing resources. This collaboration focuses on developing SoftBank’s Orchestrator, which efficiently partitions a single AMD Instinct GPU into multiple logical devices, allowing for optimal GPU resource allocation based on the varying demands of AI applications, particularly large language models. The enhanced GPU partitioning technology can split a physical MI300X-class GPU into various isolated instances, enabling better management of computing resources and minimizing latency issues. A demonstration of their advancements is scheduled to take place at the AMD booth during MWC Barcelona 2026.
SoftBank, AMD validate GPU partitioning for AI infrastructure at MWC Barcelona 2026
