Diversify your compute stack to meet all AI and HPC workload requirements

Introduction of GPUs in data centers has been a game changer for HPC and AI workloads. Parallel processing capabilities of GPUs unlocked unprecedented computational power, and by offloading parallel processing tasks from CPUs, workloads experienced significant performance gains. However, GPUs are not universally suitable for all types of computational tasks. Their parallel processing architecture is less efficient for tasks requiring complex logic and sequential execution. This limitation led to certain workloads remaining reliant on CPUs, resulting in suboptimal resource utilization. Enter Data Processing Units (DPUs), specialized hardware accelerators designed to address specific data-centric tasks such as data movement, manipulation, and acceleration. DPUs offer an opportunity to further diversify high-performance workloads, increasing efficiency by complementing the capabilities of GPUs and CPUs. By offloading data-centric tasks from CPUs and GPUs to DPUs, data centers can achieve better resource and performance optimization across a wider range of workloads. DPUs excel in tasks like data protection, reduction, networking, and compute offloading, allowing GPUs and CPUs to focus on their respective strengths. For AI workloads, for example, DPUs enhance inference by efficiently offloading and accelerating specific tasks, such as image processing and data preprocessing, resulting in faster and more power-efficient AI model execution. In this session, discover the benefits of completing your compute stack with DPUs: • Accelerate your computational AI workloads: DPUs enable faster AI model execution. • Decrease your data center TCO: DPUs boost performance per Watt and per dollar. • Optimize your data center efficiency: DPU programmability facilitates adaptation to varying workloads.