Oracle Teams Up with AMD for AI Advancements
Oracle is thrilled to announce a groundbreaking collaboration with AMD, becoming one of the first hyperscalers to deploy an AI supercomputer powered by the latest AMD Instinct MI355X GPUs. This innovative partnership aims to enhance performance for customers engaged in large-scale AI and agentic workloads, empowering them to utilize cutting-edge technology at an unprecedented scale.
Introducing OCI's Zettascale AI Cluster
Oracle Cloud Infrastructure (OCI) is set to launch an impressive zettascale AI cluster that incorporates up to 131,072 MI355X GPUs. This robust deployment not only provides customers with enhanced performance options but significantly improves price-performance ratios—doubling the advantages over previous GPU generations. With this advanced architecture, users can effortlessly build, train, and implement AI at scale.
Enhanced Performance Driven by Innovation
Mahesh Thiagarajan, executive vice president of Oracle Cloud Infrastructure, expressed the commitment to facilitating the most demanding AI workloads in the cloud. According to him, combining the power of AMD Instinct GPUs with OCI’s superior performance and security will meet growing customer needs for training and inference related to AI workloads and emergent agentic applications.
Why Choose the OCI Supercluster?
As modern AI applications evolve, they increasingly rely on larger and more intricate datasets. The OCI Supercluster powered by AMD Instinct MI355X GPUs is designed for this very purpose. It offers a high-throughput, ultra-low latency RDMA network architecture, dramatically improving the ease and speed at which customers can access and utilize AI technology.
Key Features of AMD Instinct MI355X on OCI
The AMD Instinct MI355X offers exceptional value, flexibility, and open-source compatibility, making it the ideal solution for customers managing the largest language models and AI workloads. Here’s what the new platform offers:
- Remarkable Performance Boost: Expect an impressive increase in AI deployment performance, achieving up to 2.8X higher throughput, which translates to faster results and enhanced operational efficiency.
- Larger Memory Capacity: The MI355X boasts 288 gigabytes of high-bandwidth memory (HBM3) with a memory bandwidth of up to eight terabytes per second, allowing customers to run large models entirely in memory—crucial for enhancing inference and training speeds.
- Support for New FP4 Standard: This latest iteration allows for the efficient deployment of large language and generative AI models through the new 4-bit floating point compute (FP4) standard, enabling swift and high-speed inference capabilities.
- Innovative Design: The dense, liquid-cooled configuration optimizes performance density at 125 kilowatts per rack, which means faster training times combined with higher throughput and reduced latency for demanding workloads.
- Production-Scale Training: This architecture is crafted to expedite deploying new applications, ensuring an impressive time-to-first token (TTFT) and optimizing performance for both training and inference workloads.
- Powerful Head Node: This enables customers to orchestrate jobs and process data efficiently using an AMD Turin high-frequency CPU with up to three terabytes of system memory, enhancing overall GPU performance.
- Open-Source Compatibility: The platform supports AMD ROCm, an open software stack that facilitates flexible architectures, enabling seamless migration of existing code without vendor lock-in.
- Advanced Network Designs: Oracle is pioneering the use of AMD Pollara AI NICs to create innovative network fabric designs, introducing advanced RoCE functions for high-performance, low-latency networking.
About Oracle
Oracle provides a complete suite of applications alongside secure, autonomous infrastructure within the Oracle Cloud. As a leader in cloud computing innovation, Oracle aims to help businesses achieve their goals by delivering reliable and efficient technology solutions.
Frequently Asked Questions
What is the main focus of Oracle and AMD's collaboration?
The collaboration aims to enhance customer performance for large-scale AI workloads through the deployment of advanced AMD Instinct MI355X GPUs.
What are the specifications of the OCI AI cluster?
The OCI AI cluster supports up to 131,072 MI355X GPUs, offering superior performance and price-performance ratios compared to previous generations.
How does the AMD Instinct MI355X improve AI workloads?
It provides up to 2.8X higher throughput, larger memory capacity, and efficient support for new computing standards, enhancing both training and inference speeds.
What kind of applications can benefit from this technology?
Applications requiring expansive datasets for AI training and inference, such as large language and generative AI models, will see notable benefits.
What is Oracle's commitment with this new technology?
Oracle is dedicated to delivering the broadest AI infrastructure offerings to support and meet the demanding needs of AI workloads and new applications.
