NVDA Unveils the World’s First Deep-Learning Supercomputer, DGX-1

NVIDIA moves to supercomputers

In the previous part of the series, we saw that NVIDIA (NVDA) unveiled its much-awaited Pascal GPU (graphics processing unit), aimed at accelerating deep learning, at the GTC 2016 conference. The company stated that it has begun volume production of its first Pascal-based Tesla P100 GPU.

Nvidia’s Tesla GPUs are already used in most of the world’s supercomputers. The company seeks to achieve the full potential of its fifth-generation Pascal structure by exploring the use of its Tesla P100 GPU technology in supercomputers. NVIDIA expects the hyper-scale market to reach $500 billion by 2025.

NVDA Unveils the World’s First Deep-Learning Supercomputer, DGX-1

DGX-1 at a glance

NVIDIA has unveiled the world’s first deep-learning supercomputer, DGX-1, which was made using eight Tesla P100 16GB (gigabyte) GPUs. The DGX-1 has CPU (central processing unit) and GPU computing power of 170 teraflops and SSD (solid state drive) capacity of 7 TB (terabytes). It houses two Intel (INTC) Xeons E5 V4 processors, Quad Infiniband, and Dual 10 GE (gigabit Ethernet).

What does this mean in terms of performance? DGX-1 is equivalent to 250 servers, according to NVIDIA. DGX-1 has been received well by consumers, as all systems have already been booked. The company aims to launch the product in 1Q17.

Competition

2017 will witness the launch of some strong high-end GPU architectures. Advanced Micro Devices (AMD) plans to launch its Vega architecture in 2017. In 2018, NVIDIA and AMD plan to launch their next-generation GPU architectures Volta and Navi, respectively.

Other Tesla upgrades

While DGX-1 is a year away, NVIDIA has launched upgrades of its Maxwell-based Tesla cards, the M40 and M4 deep-learning processors. IBM (IBM) and Qualcomm (QCOM) are using Xilinx’s (XLNX) FPGAs (field-programmable gate array) to boost server speed to cater to HPC (high-performance computing) requirements. With the M40 and M4, NVIDIA aims to replace FPGA with AI (artificial intelligence) GPUs.

The iShares Russell 1000 Value ETF (IWD) has exposure to large-capitalization stocks across various sectors, including technology. It has 0.19% exposure to NVDA, 1.4% to INTC, 0.65% to QCOM, and 0.62% to IBM.