AMD가 까다로운 AI 애플리케이션을 지원하는 오라클 클라우드 인프라에 AMD 인스팅트 MI300X 가속기가 탑재되며, 파이어워크 AI를 포함한 고객들은 새로운 OCI 컴퓨팅 인스턴스를 활용해 AI 추론 및 트레이닝 워크로드를 강화할 것으로 기대된다.
OCI Supercluster Supports Up to 16,384 AMD Instinct MI300X GPUs
AMD Instinct MI300X accelerators will be integrated into Oracle Cloud Infrastructure to support demanding AI applications, and customers, including Firework AI, are expected to leverage the new OCI compute instances to power their AI inference and training workloads.
AMD today announced that Oracle Cloud Infrastructure (OCI) has selected ROCm™ open software and the AMD Instinct™ MI300X accelerator to power its latest OCI Compute Supercluster instance, BM.GPU.MI300X.8.
The OCI Supercluster, powered by AMD MI300X, supports up to 16,384 GPUs in a single cluster, leveraging the same high-speed network fabric technology as OCI’s other accelerators for configurable AI models with hundreds of billions of parameters.
OCI Bare Metal instances, which offer industry-leading memory capacity and bandwidth and are designed to run demanding AI workloads, including large-scale language model (LLM) inference and training that require high throughput, have already been adopted by companies such as Fireworks AI.
“AMD Instinct MI is a powerful, high-performance computing solution that is designed to meet the needs of today’s demanding workloads,” said Andrew Dieckmann, corporate vice president and general manager, AMD Data Center GPU Business Unit.“The 300X and ROCm open software support critical areas such as OCI AI workloads, and through this, we continue to grow as a trusted solution,” he explained.
“As these solutions expand into growing AI-intensive markets, the combination will benefit OCI customers by improving performance, efficiency and system design flexibility,” it added.
“The inference capabilities of the AMD Instinct MI300X accelerator add to Oracle Cloud Infrastructure’s broad range of high-performance bare metal instances, eliminating the overhead of virtualized compute commonly used for AI infrastructure,” said Donald Lu, senior vice president, Software Development, Oracle Cloud Infrastructure. “We are excited to provide more choice at an affordable price point for customers looking to accelerate their AI workloads.”
OCI focuses on AI inference and training capabilities that can support latency-optimized use cases even at larger batch sizes, and performance suitable for the largest LLM models on a single node. AMD Instinct MI300X has undergone extensive testing against these OCI criteria and is attracting attention from AI model developers.
Firework AI provides a fast platform designed to build and deploy generative AI. With over 100 models, Firework AI is leveraging the performance and benefits of OCI powered by AMD Instinct MI300X.
“Firework AI helps enterprises build and deploy complex AI systems across a wide range of industries and use cases,” said Lin Qiao, CEO of Firework AI. “The memory capacity available with the AMD Instinct MI300X and ROCm open software helps our customers continuously grow their models and scale their services.”