Nvidia takes the wraps off the DGX-1, touting it as the "world's first deep learning supercomputer" that can deliver the computing needs of virtual reality developers and artificial intelligence researchers.

Considering how the hardware maker already developed specialized GPU units for deep learning software, it shouldn't come as much of a surprise that it's going further on that front.

"The DGX-1 is easy to deploy and was created for one purpose: to unlock the powers of superhuman capabilities and apply them to problems that were once unsolvable," Jen-Hsun Huang, cofounder and CEO of Nvidia, says, noting that researchers will no longer have to rely solely on home-built computing solutions.

As for what the DGX-1 exactly is, it's essentially a box that packs in eight Tesla P100 units with 16 GB worth of memory per card, producing up to 170 teraflops. On top of that, it houses a 7 TB SSD along with what Nvidia calls the NVLink Hybrid Cube Mesh.

While researchers usually have software of their own that they prefer to use, the DGX-1 is packaged with the Nvidia Deep Learning GPU Training System plus the Nvidia CUDA Deep Neural Network library version 5, including deep learning frameworks such as Caffe, Theano and Torch.

It's also worth mentioning that Google's Deep Dream is a notable example of a neural network that could take advantage of hardware like this.

Nvidia is expected to roll out the DGX-1 in the United States sometime in June, but it'll sport quite a hefty price tag of $129,000. That's more or less expected, as it's arguably among the hardware that'll advance the technological world in the foreseeable future.

At any rate, having Nvidia to power the computing department of researchers will definitely benefit them, as the company's support and updates for the hardware will make things a lot easier on their end.

Watch the video below to find out what Nvidia has in mind, pointing out why deep learning is one of the most popular technologies to work on today.

ⓒ 2021 TECHTIMES.com All rights reserved. Do not reproduce without permission.