AI at Scale
Supermicro COMPUTEX CEO Keynote
Accelerate Everything AI
Edge AI – Applications Anywhere
Harness the full potential of AI with Supermicro’s cutting-edge liquid-cooled infrastructure solutions. At VB Transform 2024, Supermicro CEO Charles Liang – joined by VentureBeat Founder Matt Marshall on stage – discussed the challenges and considerations in scaling AI infrastructure for modern applications, and shared insights on Supermicro’s optimized direct liquid cooling (DLC) solutions, engineered specifically for generative AI and inference workloads.
“With liquid cooling, our data center solutions consume 30 to 40% less energy, so that not just saves customers energy cost, but also enables customers to deploy 30% more computing power with the same power budget”
Large Scale AI Training
Large Language Models, Generative AI Training, Autonomous Driving, Robotics
Large-Scale AI training demands cutting-edge technologies to maximize parallel computing power of GPUs to handle billions if not trillions of AI model parameters to be trained with massive datasets that are exponentially growing. Leveraging NVIDIA’s HGX H100/H200 SXM 8-GPU/4-GPU and the fastest NVlink® & NVSwitch® GPU-GPU interconnects with up to 900GB/s bandwidth, and fastest 1:1 networking to each GPU for node clustering, these systems are optimized to train large language models from scratch in the shortest amount of time. Completing the stack with all-flash NVMe for a faster AI data pipeline, we provide fully-integrated racks with liquid cooling options to ensure fast deployment and a smooth AI training experience.
Workload Sizes
- Extra Large
- Large
- Medium
- Storage
Resources
HPC/AI
Engineering Simulation, Scientific Research, Genomic Sequencing, Drug Discovery
Accelerating time to discovery for scientists, researchers and engineers, more and more HPC workloads are augmenting machine learning algorithms and GPU-accelerated parallel computing to achieve faster results. Many of the world’s fastest supercomputing clusters are now taking advantage of GPUs and the power of AI.
HPC workloads typically require data-intensive simulations and analytics with massive datasets and precision requirements. GPUs such as NVIDIA’s H100/H200 provide unprecedented double-precision performance, delivering 60 teraflops per GPU, and Supermicro’s highly flexible HPC platforms allow high GPU counts and CPU counts in a variety of dense form factors with rack scale integration and liquid cooling.
Workload Sizes
- Large
- Medium
Resources
Enterprise AI Inference & Training
Generative AI Inference, AI-enabled Services/Applications, Chatbots, Recommender System, Business Automation
The rise of generative AI has been recognized as the next frontier for various industries, from tech to banking and media. The race to adopt AI has begun as a source to breed innovation, significantly boost productivity, streamline operations, make data-driven decisions, and improve customer experience.
Whether it is AI-assisted applications and business models, intelligent human-like chatbots for customer service, or AI to co-pilot code generation and content creation, enterprises can leverage open frameworks, libraries, pre-trained AI models, and fine-tune them for unique use cases with their own dataset. As the enterprise adopts AI infrastructure, Supermicro’s variety of GPU-optimized systems provide open modular architecture, vendor flexibility, and easy deployment and upgrade paths for rapidly-evolving technologies.
Workload Sizes
- Extra Large
- Large
- Medium
Resources
Visualization & Design
Real-Time Collaboration, 3D Design, Game Development
Increased fidelity of 3D graphics and AI-enabled applications by modern GPUs is accelerating industrial digitization, transforming product development and design processes, manufacturing, and content creation with true-to-reality 3D simulations to achieve new heights of quality, infinite iterations at no opportunity costs, and faster time-to-market.
Build virtual production infrastructure at scale to accelerate industrial digitalization through Supermicro’s fully-integrated solutions, including the 4U/5U 8-10 GPU systems, an NVIDIA OVX™ reference architecture, optimized for NVIDIA Omniverse Enterprise with Universal Scene Description (USD) connectors, and NVIDIA-certified rackmount servers and multi-GPU workstations.
Workload Sizes
- Large
- Medium
Resources
Content Delivery & Virtualization
Content Delivery Networks (CDNs), Transcoding, Compression, Cloud Gaming/Streaming
Video delivery workloads continue to make up a significant portion of current Internet traffic today. As streaming service providers increasingly offer content in 4K and even 8K, or cloud gaming in a higher refresh rate, GPU acceleration with media engines is a must to enable multi-fold throughput performance for streaming pipelines while reducing the amount of data required with better visual fidelity, thanks to the latest technologies such as AV1 encoding and decoding.
Supermicro’s multi-node and multi-GPU systems, such as the 2U 4-Node BigTwin® system meet the stringent requirements of modern video delivery, each node supporting the NVIDIA L4 GPU with the ability to feature plenty of PCIe Gen5 storage and networking speed to drive the demanding data pipeline for content delivery networks.
Workload Sizes
- Large
- Medium
- Small
Resources
Edge AI
Edge Video Transcoding, Edge Inference, Edge Training
Across industries, businesses whose employees and customers engage at edge locations – in cities, factories, retail stores, hospitals, and many more – are increasingly investing in deploying AI at the edge. By processing data and utilizing AI and ML algorithms at the edge, businesses overcome bandwidth and latency limitations, enabling real-time analytics for timely decision making, predictive care and personalized services, and streamlined business operations.
Purpose-built, environment-optimized Supermicro Edge AI servers with various compact form factors deliver the performance needed for low-latency, open architecture with pre-integrated components, diverse hardware and software stack compatibility, and privacy and security featuresets required for complex edge deployments out of the box.
Workload Sizes
- Extra Large
- Large
- Medium
- Small
Resources
Broadest Portfolio of AI-Ready Systems
Deploy NVIDIA Omniverse™ at Scale
COMPUTEX 2024 CEO Keynote