Nvidia A100 GPUs and Fifty New Servers of Computer Makers

Computer makers have unveiled a total of 50 servers with Nvidia’s A100 GPUs (graphics processing units) to artificial power intelligence, a scientific computing application, and data science endeavors. The A100 is the first graphics processing unit based on the architecture of Nvidia Ampere. Moreover, the A100 is the most significant leap the company has made in GPU performance to date. It has features such as the ability for one GPU to be partitioned into sever separate GPUs if needed—which is what Nvidia said. The company made the announcement before the ISC High-Performance online event. This online event is dedicated to high-performance computing.

The A100 GPU has 54 billion transistors. These transistors are the on-off switches the building blocks of all things electronic. Moreover, they can execute five petaflops of performance. This is 20 times higher than the previous-generation chip Volta. That means a CPU (central processing unit) server, costing $20 million, and taking up 22 racks, can be replaced by new servers that cost $3 million and take up just four GPU-based server racks. This is what Paresh Kharya said in a press briefing. Paresh Kharya is the director of product marketing for accelerating computing at Nvidia.

The systems will be in use by several computer makers, which include Supermicro, Quanta/QCT, One Stop Systems, Lenovo, Inspur, Hewlett Packard Enterprise, Gigabyte, Fujitsu, Dell, Cisco, Atos, and Asus. Server availability is varying, with 30 systems expected this summer. Moreover, Kharya said that over 20 more systems will add at the end of the year.

Nvidia A100

The new machines include new interconnected technology from InfiniBand from Mellanox. Nvidia acquired it in 2019, for which it paid $7 billion. Moreover, Nvidia integrated Mellanox technology with the A100 to create Selene. Nvidia bills Selene as a top 10 supercomputer and the most energy-efficient computer in the world. Selene provides over one exaflop of Artificial Intelligence Processing power. It took less than a month to design Selene, overall. Kharya said that supercomputers like Selene help Nvidia penetrate further into the top supercomputers of the world.

Nvidia’s GPUs were part of 125 of the top 500 supercomputers of the world last year. That is, according to ISC. If someone were to count supercomputers with this Mellanox InfiniBand technology, the number will be more than 300. The list will most probably grow even more significantly in 2020.

Kharya said that, if you were to check the top 500 list, the reason why Nvidia is so successful in supercomputing is that scientific computing has changed. Kharya adds that we have entered a new era. That era has expanded beyond traditional simulation and modeling workloads to include artificial intelligence, big data visualization, edge screening, and data analytics.

Kharya said Mellanox interconnect chips power the leading weather forecast supercomputers in the world. Climate and weather models are both data and computationally intensive. Forecast quality highly depends on the complexity of the model and the resolution level. The performance also depends on interconnect technology. That is, to move data quickly around different computers.

Gilad Shainer is the senior vice president at Nvidia. He said, in a press briefing, that it is exciting to have the best computing on one side. Moreover, it is exciting to have the best network on the other side.