NVIDIA chips power world’s most powerful supercomputer

Last week the US Department of Energy debuted a new supercomputer with the name Summit, built by the Oak Ridge National Laboratory in Tennessee, that is eight times faster than the previous top US-based supercomputer (named Titan) that was launched just five years earlier. For some AI applications and workloads Summit will be more than 100x faster than Titan. This supercomputer recaptures the top performing spot for the United States for the first time since 2013 and unseats the Chinese-based Sunway TaihuLight with 60% better performance.

Summit is comprised of 4,608 servers with 9,216 IBM Power9 processors but at the core of the massive level of computational capability is the 27,648 flagship NVIDIA Volta graphics chips that are capable of drastically accelerated AI and machine learning performance.

Though built by the Department of Energy and Oak Ridge, NVIDIA is the company that benefits the most from its launch. As the market continues to adopt GPU technology in high performance computing systems, NVIDIA’s leadership position with these platforms provides long-term leadership stability for the company.

The impact for science

The DOE designed Summit to tackle real problems, especially world-changing problems. It will accelerate the top scientists’ work in physics, material creation, healthcare, and more. The DOE said at its launch event that the supercomputer already has a full schedule working on complex tasks for scientific advancement.


Cancer research is starting with a process that analyzes health data to find relationships between disease factors like genetics, biological markers, and the environment. Summit will be modeling the capability for a fusion reactor, the energy source that powers the sun, to try to find a way to commercialize this clean energy source. Researchers are using AI to identify function and evolution patters of proteins and cellular systems to understand Alzheimer’s, heart disease, and drug discovery.

Summit is capable of 200 petaflops, or 200 billion billion floating point operations per second, and this is what gives the supercomputer planet-changing capability. An early example of system utilization solved a genomics problem in about an hour that would take a single high-performance PC more than 30 years. Though access to the powerful hardware will be limited to a small section of the scientific computing, as hardware pricing continues to fall, more researchers will have access to supercomputer-class capability.

The impact for NVIDIA

With 95% of the total compute power of Summit provided by the 27,000+ NVIDIA graphics chips, it isn’t a stretch to see the important of the company’s technology to supercomputing and scientific discovery. At it’s heart, this is an NVIDIA system, even though the US Department of Energy and Oak Ridge National Labs built the beast.

This is yet another indicator that NVIDIA owns the supercomputing market, with leadership performance and partnerships. The company’s investments in developer communities and software integration for its graphics chips over the last decade has become one of the most impressive long-term investments from a corporate strategy perspective.

In the same way that Cray became known as the supercomputer company in the 1970s and 1980s, NVIDIA is generating the same brand power and awareness with its consistent integration with supercomputer designs around the globe. Cray itself used NVIDIA graphics chips starting in 2011.

Architectural shifts in the supercomputing market take a lot time to develop, and that extended investment by NVIDIA has cemented its position for today’s designs. As with all computing leaders, there is a risk of changing workloads and movement to other platforms, but NVIDIA has shown the capability to change and adapt. The company knows the value that the market has placed on NVIDIA’s leadership in supercomputing, as well as machine learning and AI, and that it will have to maintain revenue and mindshare growth as this market expands.