Number of systems using NVIDIA GPU accelerators,
Up 48% in one year, more than three times higher than five years ago
NVIDIA T4 GPU for Google Cloud Unveiled for the First Time
NVIDIA has once again solidified its position in the global supercomputing field by setting a new record on the recently announced list of the '500 Fastest Supercomputers in the World'.

The number of systems using NVIDIA GPU accelerators among the top 500 supercomputers announced at Supercomputing 2018, an HPC conference held in Dallas, Texas, from the 11th to the 16th (local time), increased by 48% over the past year. This rises from 86 a year ago to 127 this year, a figure more than three times higher than five years ago.

NVIDIA GPUs are also installed in 'Summit' and 'Sierra,' the world's two fastest supercomputers deployed by the U.S. Department of Energy at Oak Ridge National Laboratory and Lawrence Livemore National Laboratory, respectively. Utilizing these two systems equipped with over 40,000 NVIDIA V100 Tensor Core GPUs, the world's leading researchers were recognized for innovative research in five of the six categories of this year's Gordon Bell Awards, and the awards ceremony will be held at Supercomputing 2018. The fastest supercomputers in Europe and Japan are also being accelerated by NVIDIA GPUs.
In addition, on the 'GREEN500' list, which indicates the energy efficiency of supercomputing systems, 22 of the top 25 most eco-friendly supercomputers were powered by NVIDIA technology.

"With the end of Moore's Law, a new HPC market has emerged centered on AI and machine learning workloads," said Jensen Huang, founder and CEO of NVIDIA. "This market relies more than ever on high-performance and high-efficiency GPU platforms to provide the performance needed to solve challenging problems in science and society."

While only 33 of the supercomputers newly added to the top 500 list a year ago used GPU acceleration systems, this year 52 of the 153 new supercomputers, or about one-third, were found to be GPU-accelerated.

Interest in NVIDIA GPUs was high not only in supercomputers but also in other fields.

NVIDIA announced that the new NVIDIA T4 GPU is showing the fastest adoption rate among server GPUs.
The NVIDIA T4 GPU, launched last September, has been integrated into 57 server designs provided by computer manufacturers worldwide. With the T4 GPU now available in the cloud, the T4 for Google Cloud Platform customers has also been unveiled for the first time.

“Never before has a data center processor seen such a rapid adoption rate,” said Ian Buck, Vice President and General Manager of Accelerated Computing at NVIDIA. “The T4 became available in the cloud just 60 days after launch and is supported by a network of server manufacturers worldwide.”

He also added, "T4 provides the performance and efficiency required for today's public and private clouds to handle compute-intensive workloads."

T4 accelerates various cloud workloads such as HPC, deep learning training and inference, machine learning, data analysis, and graphics. Based on the new NVIDIA Turing architecture, this product features multi-precision Turing Tensor Cores and new RT Cores, and provides higher performance than before when combined with an accelerated containerized software stack.
“Lower latency is required to deliver real-time visualization and online inference workloads to end users,” said Damion Heredia, Senior Director of Product Management at Google Cloud. “NVIDIA T4 GPUs for Google Cloud provide our machine learning and visualization customers with a low-latency platform that is highly scalable and cost-effective.”

He also explained, "Customers can innovate in new ways by combining Google Cloud's network capabilities with T4's features, increasing application speed while lowering costs."