NVIDIA Tensor Core GPUs power the fastest AI supercomputers

Source: NVIDIA. The Volta Tensor Core GPU.

NVIDIA has announced that the world’s top artificial intelligence (AI) supercomputers are all powered by NVIDIA Tensor Core GPUs. Designed specifically for deep learning, Tensor Cores deliver up to 12X higher peak teraflops (TFLOPS) for training and 6X higher peak TFLOPS for inference over the previous generation of NVIDIA technology.

The latest TOP500 list of the world’s 500 fastest systems shows that several AI supercomputers in the lead. AI supercomputers can process both traditional high performance computing (HPC) simulations and new AI workloads. The world’s fastest system and third-fastest systems, Summit and Sierra, are both AI supercomputers running on Tensor Core GPUs. The GPUs also power the fastest supercomputer in Japan, ABCI.

According to NVIDIA, the new systems reflect the broader shift to accelerators in the TOP500 list: GPUs now power five out of the world’s seven fastest systems, as well as 17 of the 20 most energy-efficient systems on the new GREEN500 list. In addition, the majority of compute performance added to the list comes from NVIDIA GPUs.

“The new TOP500 list clearly shows that GPUs are the path forward for supercomputing in an era when Moore’s Law has ended,” said Ian Buck, VP and GM, accelerated computing at NVIDIA. “With the invention of our Volta Tensor Core GPU, we can now combine simulation with the power of AI to advance science, find cures for disease and develop new forms of energy. These new AI supercomputers will redefine the future of computing.”

Jack Dongarra, professor at the University of Tennessee and Oak Ridge National Laboratory in the US, publishes the TOP500 list. He said: “This year’s TOP500 list represents a clear shift toward systems that support both HPC and AI computing. Accelerators, such as GPUs, are critical to deliver this capability at the performance and efficiency targets demanded by the supercomputing community.” 

The No. 1 system on the TOP500 is Oak Ridge National Laboratory’s Summit, which went online earlier this month. It was clocked at 122 petaflops (million billion floating point calculations) of processing power in double-precision performance, surpassing the No. 2 system, China’s Sunway TaihuLight, which was recorded at 93 petaflops. Summit draws 95% of its performance from its 27,648 Tensor Core GPUs, the most GPUs of any supercomputer. Summit is also capable of over 3 exaops of AI performance, or 3 billion billion multi-precision AI calculations per second.

The no. 3 top system powered by NVIDIA GPUs is Sierra, based at the Lawrence Livermore National Laboratory in the US. It delivers 71 petaflops of performance using 17,280 GPUs. The top Asian supercomputer ranks No. 5. The AI Bridging Cloud Infrastructure (ABCI), Japan’s fastest system, delivers 19.6 petaflops of performance using 4,352 GPUs.

Explore:

Read the TechTrade Asia blog post about the Summit supercomputer