WHY THIS MATTERS IN BRIEF
- Many of todays technology advances, such as AI and materials science, have been made possible by the dramatic increase in computing power that’s occurred over the last decade. Governments, such as China, Japan, the UK and the US are locked in a supercomputer arms race as they see these systems as a way to boost innovation, jobs growth and GDP
When technologists think of “supercomputer” they usually imagine government-owned computers like China’s TaihuLight or public-private partnerships like the Department of Energy’s planned Summit system. Private in house supercomputers are much rarer and that makes Nvidia’s new supercomputer, the DGX SaturnV, even more surprising.
Not only has Nvidia revealed its own supercomputer, but the machine cracked TOP500’s list of the 500 most powerful computers and took the number one spot as the world’s most energy-efficient supercomputer.
The DGX SaturnV has 63,488GB of RAM and 60,512 Intel Xeon E5-2698v4 CPU cores and relies on numerous Nvidia’s 125 DGX-1s, the “AI supercomputer in a box” units built for deep learning that Nvidia’s CEO presented to Elon Musk last year to help support his OpenAI project. It also uses the company’s Tesla P100 GPUs.
The gallery was not found!
The machine delivers 9.46 gigaflops/watt, making it 42 percent more energy efficient than the most efficient supercomputer on TOP500’s June list and it’s a clear sign from Nvidia that it’s intent on making sure that its new DGX-1 system is taken seriously as the deep learning platform of choice for enterprises around the world.
In the meantime, not just content with their lot Nvidia is busy training its own neural networks on the intricacies of chipset designs so they can help it build even more powerful and more efficient GPUs in the future.