With Using New DPX Instructions NVIDIA H100 Hopper GPU, A New Architecture Can Accelerate Dynamic Programming Close To 40x
The algorithms of Dynamic programming are generally used in robotics, healthcare, data science, quantum computing, and more.
The architecture of Hopper GPU is unveiled by Nvidia today at GTC, and it will accelerate dynamic programming — a problem-solving methodology used in algorithms for quantum computing, genomics, route optimization, and more — by up to 40x using new DPX instructions.
A set of instructions is built into NVIDIA H100 GPUs and DPX will help developers for writing code to achieve speedups on dynamic programming algorithms in many industries, boosting workflows for quantum simulation, disease diagnosis, routing optimizations, and graph analytics.
What Is the Meaning of Dynamic Programming?
Dynamic programming was developed in the 1950s. This is a popular methodology for solving complex problems. Dynamic programming has two key techniques: memoization and recursion.
Recursion can break a complex problem down into some simpler sub-problems. For that reason, recursion saves computational effort and time. Memoization can provide the solutions to these sub-problems, which can be reused many times when solving the main problem — are stored. Memoization improves efficiency, so the simpler sub-problems do not need to be recomputed when needed later on in the main problem.
Dynamic programming algorithms can be accelerated by DPX instructions by close to 7x on an NVIDIA H100 GPU, compared to the NVIDIA GPUs with Ampere architecture. The acceleration can be boosted even more by a node with four NVIDIA H100 GPUs.
Use Cases Span Healthcare, Quantum Computing, Robotics, Data Science
Dynamic programming is generally used in many omics algorithms, optimizations, and data processing. Until now, most developers have run these kinds of algorithms on FPGAs or CPUs — but can unlock dramatic speedups with the help of DPX instructions on NVIDIA Hopper GPUs.
Omics
Omics covers a range of biological branches which include proteomics (focused on proteins), genomics (focused on DNA), and transcriptomics (focused on RNA). These branches, which inform the critical work of drug discovery and disease research, all rely on algorithmic analyses which can be sped up with the help of DPX instructions.
the Needleman-Wunsch and Smith-Waterman algorithms of dynamic programming are used for protein classification, protein folding, and DNA sequence alignment. A scoring method is used by both algorithms to measure how well genetic sequences from different samples align.
Smith-Waterman creates highly accurate results but takes more time and compute resources compared to other alignment methods. DPX instructions are used on a node with four NVIDIA H100 GPUs, so the scientists can speed this process up to 35x to achieve real-time processing, where the work of alignment and base calling takes place at the same rate as DNA sequencing.
This acceleration can help democratize genomic analysis in hospitals globally, bringing scientists closer to providing patients with personalized medicine.
Route Optimization
Finding the optimal route for multiple moving pieces is necessary for independent robots moving through a dynamic warehouse, or even a sender sending data to many receivers in a computer network.
To handle this optimization problem, developers rely on a dynamic programming algorithm named Floyd-Warshall. This algorithm is used to find the shortest distances between all pairs of destinations in a graph or map. A server consists of four H100 GPUs from NVIDIA, the acceleration known as Floyd-Warshall can be boosted up to 40x more than a traditional type dual-socket which is a CPU-only type server.
Put together with the NVIDIA cuOpt AI logistics software, this speedup in routing optimization can be used for real-time applications in autonomous vehicles, factories, or routing and mapping algorithms in abstract graphs.
Availability
NVIDIA DGX PODs, DGX H100 systems, and DGX SuperPODs will be available from NVIDIA’s global partners starting in the third quarter of 2022.
Customers also can select to deploy the system which is DGX at layout facilities that are operated by the partners of DGX-Ready type Data Center from NVIDIA including Digital Realty, Cyxtera, and Equinix IBX data centers.
In this content, we already shared some information about NVIDIA H100 GPU. We will publish more news on NVIDIA H100 GPU later. That is all for today. If you have any questions regarding H100 Hopper GPU, Dynamic Programming, DPX instructions, NVIDIA DGX PODs, etc. then leave your comments in the comment box below. We will answer your every question after some time.
