In the ever-changing world of AI, companies compete fiercely to tailor accelerators for evolving workloads. Amid this race, Nvidia’s CUDA platform emerges as the unrivaled leader, sparking a surge in demand for Nvidia-based AI infrastructure. Recognizing this trend, Amazon Web Services (AWS) and Nvidia forge a strategic partnership to deliver cutting-edge Nvidia-based infrastructure for generative AI – a synergy that propels AWS to the forefront of GPU operations.
AWS CEO Adam Selipsky highlights the significance, stating, “Today, we proudly offer a comprehensive array of Nvidia GPU solutions. Extending beyond graphics and gaming, we enter the pioneering frontier of generative AI.” This transformative partnership leverages the cutting-edge Nvidia Grace Hopper Superchips, seamlessly integrated into AWS’s robust networking capabilities, elevated by hyper-scale clustering and state-of-the-art virtualization. Together, they forge a powerful collaboration, propelling us to the forefront of advancing AI technology.
A cornerstone initiative, Project Ceiba, aims to create the world’s fastest GPU-powered AI supercomputer exclusively on AWS for Nvidia’s generative AI R&D. And also boasting 16,384 Nvidia GH200 Superchips, this supercomputer promises a colossal 65 ‘AI ExaFLOPS,’ propelling generative AI projects into new realms of innovation.
The partnership unveils the Nvidia DGX Cloud on AWS, a major component providing AI-training-as-a-service. By integrating the GH200 NVL32 machine, it revolutionizes training for advanced generative AI and large language models. And it is providing the largest shared memory in a single instance.
AWS takes the lead, offering a cloud-based AI supercomputer based on Nvidia’s GH200 Grace Hopper Superchips. There fore this unique configuration connects 32 Superchips per instance, scaling up to thousands, supported by Amazon’s EFA networking, AWS Nitro System, and Amazon EC2 UltraClusters. Those are pushing the boundaries of AI processing capabilities.
Introducing new Nvidia-powered Amazon EC2 instances, featuring H200 Tensor Core GPUs with up to 141 GB of HBM3e memory, showcasing the versatility of Nvidia’s technology. Additionaly G6 and G6e instances, equipped with NvidiaL4 and L40S GPUs, cater to a wide array of applications, from AI fine-tuning to 3D workflow development.
The collaboration extends beyond hardware, introducing Nvidia’s advanced software to expedite generative AI development on AWS. And aslo this includes the NeMo LLM framework, NeMo Retriever, and BioNeMo, collectively accelerating various facets of generative AI development and application.
Jensen Huang, Nvidia’s founder, and CEO highlights the transformative role of generative AI on cloud workloads, stating as “Generative AI is reshaping the cloud landscape, establishing accelerated computing as the cornerstone for diverse content generation”. Additionaly united by a common goal to provide cost-effective, cutting-edge generative AI, Nvidia and AWS join forces to shape the trajectory of AI’s future. additionally their collaborative efforts span the entire computing spectrum, from foundational infrastructure and acceleration libraries to pioneering models and innovative generative AI services. Together, they chart a course towards a future where AI seamlessly integrates into every facet of computing.
