AMD has announced that Oracle Cloud Infrastructure (OCI) has selected its AMD Instinct MI300X accelerators, powered by the ROCm open software platform, to support OCI’s latest high-performance compute supercluster instances—BM.GPU.MI300X.8. Designed to handle demanding AI workloads, including massive large language model (LLM) inference and training, the MI300X-powered OCI supercluster can scale up to an impressive 16,384 GPUs within a single cluster, leveraging OCI’s high-speed networking infrastructure.
These bare-metal instances cater to AI workloads that require top-tier throughput, memory capacity, and bandwidth. Companies like Fireworks AI are already adopting this solution to enhance performance and scalability. Andrew Dieckmann, Corporate Vice President of Data Center GPU at AMD, highlighted that the MI300X and ROCm software offer high performance and flexibility, making them ideal for OCI’s growing AI-centric market.
Donald Lu, Senior VP of Software Development at OCI, emphasized that AMD’s MI300X brings enhanced inference capabilities without virtualization overhead, offering cost-effective AI acceleration options for OCI customers.
The AMD Instinct MI300X has been rigorously tested and validated by OCI, demonstrating its ability to handle even the most latency-sensitive applications and accommodate large-scale AI models. Companies like Fireworks AI are benefiting from this power, using the MI300X to build and deploy generative AI systems across various industries.
All Comments (0)