Nvidia is building the fastest computer!

Combines 16 graphics cards into one giant virtual GPU.

Nvidia just announced its HGX-2 cloud server platform, which the company says will be “the fastest computer humanity has ever created.”

The computer will combine 16 Tesla V100 graphics cards that work together to create a giant virtual GPU with half a terabyte of video memory and two petaflops of processing power. This is made possible by Nvidia’s NVSwitch technology, which links video cards together to work like a single GPU.

The announcement was made at the Nvidia GTC (GPU Technology Congerence) in Taiwan, a warm-up before the company kicks off in full force at Computex 2018 next week.

While the HGX-2 certainly has some mind-boggling specs, the platform is not planned for standard computers. Instead, it will be capable of high-precision computing using FP64 and FP32 for scientific computing and simulation, including allowing FP16 and Int8 for AI training.

| BEST TEACHER

Jen-Hsun Huang, Founder and CEO of Nvidia, announced to GTC that “the world of computing has changed […] Processor scaling has slowed down while the demand for computers has skyrocketed. NVIDIA HGX-2 with Tensor Core GPUs promises the industry a powerful, versatile computing platform that brings together HPC and AI to tackle the world’s big challenges. ”

According to Nvidia, the HGX-2 achieved record AI training speeds of 15,500 frames per second on the ResNet-50 test bench and is powerful enough to replace up to 300 servers.

At Computex 2017, Nvidia announced the popular HGX-1, which is used by companies relying on mass data centers like Facebook and Microsoft.

Nvidia has high hopes for the HGX-2, as well as some big companies including Lenovo, QCT, Supermicro, Foxconn, and Wiwynn, announcing plans to launch HGX-2 systems this year.

According to Paul Yu, Nvidia is raising the bar with an architecture capable of delivering two petaflops of performance for heavy AI and HPC workloads. “With the HGX-2 server architecture, we will be able to develop new systems that can meet the growing needs of our customers for maximum performance on a large scale.”

| STATISTICS GPU

The HGX-2 is powered by Nvidia Tesla V100 GPUs that feature 32GB of high bandwidth memory to deliver 125 teraflops of deep learning. Combining 16 such GPUs together promises monstrous results.

“Each GPU can interact with each GPU at the same time with a bandwidth of 300 Gb / s, 10x the speed of PCI Express,” Huang said, “so that everyone can communicate simultaneously.”

NVIDIA DGX-2

Nvidia also showcased the Nvidia DGX-2, the first system built using the HGX-2 server platform, the system comes with 2 petaflops of processing power and 512GB of HBM2 memory.

According to Huang, “This is the fastest computer that mankind has ever created.” Pretty interesting.

Nvidia Turing: Release date, news and rumors.