Obsessed with technology?
Subscribe to the latest tech news as well as exciting promotions from us and our partners!
By subscribing, you indicate that you have read & understood the SPH's Privacy Policy and PDPA Statement.
News
News Categories

NVIDIA's DIGITS DevBox is a US$15,000 deskside deep learning machine

By Vijay Anand - on 18 Mar 2015, 8:07pm

NVIDIA's DIGITS DevBox is a US$15,000 deskside deep learning machine

This year at NVIDIA's GPU Tech Conference 2015, the focus was all out on Deep Learning and deep neural networks (DNN). We've a dedicated feature that discusses NVIDIA's new focus and what's in it for them and the industry as a whole.

To complement their new initiative, GTC 2015 also saw the launch of the GeForce GTX Titan X graphics card. With 8 billion transistors, 3,072 CUDA cores, 192 TMUs, 384-bit memory bus and a whopping 12GB of 7GHz GDDR5 memory, it's got a lot of bandwidth to make short work of almost anything you throw at it. Double-precision workloads still reign supreme on the Titan Z, but that's a different class graphics and price point. For a single GPU card with a 250W TDP, the Titan X is a tough act to beat.

While you could go about building your own deep learning capable machine with your own preferred specs, or appoint a suitable OEM to follow-up on your needs, the NVIDIA DevBox takes one step ahead to put together an awesome hardware configuration and all the software, tools and framework you'll need to dabble with DNNs.

While NVIDIA has no intention of upsetting the OEM market and supply chain, they've been through the processes of setting up systems for their internal researchers and found that it requires a number of software components in addition to having adequately capable hardware to have a platform ready for use. With feedback from their prospective end-users, NVIDIA decided to offer a kickass system that has everything ready to go - from taking out the guesswork on what components are required, to ensuring all the relevant software framework is in place for researchers to start work on deep learning research as soon as possible.

The DevBox comes in only one configuration:-

  • ASUS X99 (8 PCIe slots) + Core i7 processor
  • 64GB DDR4
  • 2 x 48-port gen3 PCIe (PEX8748) + CPU for PCIe connectivity
  • 4 x NVIDIA GeForce GTX Titan X graphics cards
  • Up to 3 x 3TB RAID 5 + M.2 SATA + SSD
  • 1500W PSU
  • Ubuntu 14.04
  • NVIDIA-qualified driver
  • NVIDIA CUDA Toolkit 7.0
  • RAID setup with SSD/M.2 caching
  • Caffe, Theano,Torch, BIDMach
  • NVIDIA DIGITS software providing powerful design, training, and visualization of deep neural networks for image classification

With a rather powerful hardware specs in a deskside enclosure (which is in fact a Corsair Carbide Series Air 540 chassis), NVIDIA claims that the DevBox is the most powerful machine to be powered from a single wall plug. It's also why it will set you back US$15,000.

Even if you have the cash, NVIDIA isn't going to ship you one unless you communicate with them on why you intend to get one and what would it be used for. They will then assess your needs and prioritize your order respectively. These checks are in place to ensure the NVIDIA team is able to assist only those who have a strong or urgent need to get in to deep learning research (and not just any brat who has the monetary sum to splurge on a machine to show off). However, if you plan on setting up a DIY DevBox but not necessarily of the same specs, NVIDIA has an appropriate support page to capture your requirements and send you the necessary instructions and tools.

So what are you waiting for? It's time to start your research on deep learning and dabble in DNNs!

The NVIDIA DevBox will be available from May 2015.

Source: NVIDIA