Redefining Accelerated Computing with NVIDIA Hopper™ Architecture
In the ever-evolving landscape of artificial intelligence (AI) and high-performance computing (HPC), one name consistently appears at the forefront: NVIDIA. With their latest brainchild, the Hopper architecture, NVIDIA seems ready to revolutionize the industry once more. Aimed at seamlessly scaling a plethora of workloads across data centers of all sizes, the Hopper accelerates operations at an unprecedented pace, making it a giant leap forward in the field of accelerated computing.
The Evolution of Enterprise AI
The mainstream adoption of AI across enterprises necessitates an infrastructure that can support the fast pace of evolution. NVIDIA’s Hopper architecture, an embodiment of this necessity, advances Tensor Core technology with the Transformer Engine, designed to hasten the training of AI models.
The architecture is equipped with Hopper Tensor Cores that apply mixed FP8 and FP16 precisions to expedite AI calculations significantly. As an added boon, Hopper triples the floating-point operations per second (FLOPS) for several precisions compared to its predecessor. Coupled with the Transformer Engine and fourth-generation NVIDIA® NVLink®, Hopper Tensor Cores enable a dramatic speedup on HPC and AI workloads.
High-Throughput AI Workloads
The Hopper architecture’s innovative features pave the way for high-throughput AI workloads. Fourth-generation NVLink can enhance multi-GPU IO with NVIDIA DGX™ and HGX™ servers at a rate of 900 gigabytes per second (GB/s) bidirectional per GPU, offering over 7X the bandwidth of PCIe Gen5.
The third-generation NVIDIA NVSwitch™, meanwhile, supports the Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)™ in-network computing, offering a 3X increase in all-reduce throughput within eight H100 GPU servers compared to the previous-generation A100 Tensor Core GPU systems.
Security, Integrity, and Innovation
NVIDIA doesn’t compromise on security either. The Hopper ensures that unauthorized entities can neither view nor modify the application code and data when it’s in use. This promises confidentiality and integrity of data and applications while accessing the unparalleled acceleration of H100 GPUs for AI training, AI inference, and HPC workloads.
The Hopper architecture advances the Multi-Instance GPU (MIG), allowing a GPU to be divided into smaller, isolated instances with their own memory, cache, and compute cores. The Hopper architecture enhances MIG by supporting multi-tenant, multi-user configurations in virtualized environments across up to seven GPU instances, securely isolating each instance with confidential computing at the hardware and hypervisor level.
Unprecedented Performance with Dynamic Programming
One of the standout features of the Hopper is its dynamic programming, an algorithmic technique for solving complex recursive problems by breaking them down into simpler subproblems. Hopper’s DPX instructions expedite dynamic programming algorithms by 40X compared to traditional dual-socket CPU-only servers and by 7X compared to NVIDIA Ampere architecture GPUs. The dramatic increase in speed results in faster disease diagnosis, routing optimizations, and graph analytics.
Looking Ahead
In conclusion, the NVIDIA Hopper architecture has set a new standard for accelerated computing. It promises to redefine the performance, scalability, and security for every workload, giving us a glimpse into the future of the AI landscape. Through continued innovations, NVIDIA is set to securely accelerate all workloads from enterprise to exascale, shaping the next generation of enterprise AI and high-performance computing. As we look forward to the release of the NVIDIA Hopper, there’s no doubt that this technology will usher in a new era for both AI and HPC.