SoftBank Corp. and AMD to
Jointly Validate Utilization of AMD Instinct™ GPU
for Next Generation AI Infrastructure

Background

As the use of generative AI continues to grow, the demand for applications powered by AI models such as large language models (LLMs) continues to grow. In particular, LLMs require computing resources that vary greatly depending on the number of parameters and concurrent execution of the model, so allocating resources uniformly can lead to GPU shortage or surplus, resulting in inefficient operation.

Against this backdrop, SoftBank has been developing Orchestrator, which manages computing resources and optimally allocates AI applications, with the aim of realizing a next-generation AI infrastructure that can flexibly control computing resources according to the model size and execution requirements of AI applications.

Overview of the joint validation

In collaboration with AMD, SoftBank has developed an enhanced Orchestrator feature that leverages the GPU partitioning capabilities of AMD Instinct™ GPUs, which allow a single GPU to be used as multiple logical devices. This feature allows for the flexible and optimal allocation of GPU resources based on the requirements of the AI application, such as model size and concurrency. This makes it possible to efficiently run multiple AI applications on a single GPU, and it has been confirmed that the enhanced Orchestrator is capable of operating with minimal strain on GPU resources and the occurrence of surplus.

A demonstration concerning the joint verification is planned to be showcased at the AMD booth during MWC Barcelona 2026. Additionally, technical details such as the architecture and Orchestrator management methods are available at the SoftBank Research Institute of Advanced Technology's blog.

SoftBank and AMD will continue joint technical evaluation in advancing the potential utilization of AMD Instinct™ GPUs for next-generation AI infrastructure.

Ryuji Wakikawa, Vice President and Head of the Research Institute of Advanced Technology at SoftBank Corp., said:
"Through joint validation with AMD, we implemented orchestration logic to control AMD Instinct™ GPUs. This enables more efficient operation of multiple AI applications on a single GPU. SoftBank will continue to improve the efficiency of computing resource utilization."

Kumaran Siva, Corporate Vice President, Strategic Market Development, AMD, said:
"AMD and SoftBank share a focus on delivering high-performance, efficient infrastructure that helps bring AI inference into real-world deployments. Through our collaboration, SoftBank is right-sizing GPU resource allocation to match model requirements and help enable flexible inference platforms that support a wide range of real-world AI services. We are excited to continue working with SoftBank to advance enterprise technologies AI inference for next generation AI infrastructure."