Nvidia DGX-1 Signed, Sealed, Delivered by Jen-Hsun Huang

1 post in this topic

Nvidia DGX-1 Signed, Sealed, Delivered by Jen-Hsun Huang

Nvidia's new deep learning supercomputer, the Nvidia DGX-1 was delivered by Jen-Hsun Huang last week ready to undertake it's most challenging task to date. Taking delivery was Greg Brockman, CTO and Co-Founder at Open AI. After 3 years development, the system wasn't a small task by any stretch of the imagination as he joked "So if this is the only one ever shipped, this project would cost $2 billion."

Jen-Hsun Huang taking a peak at an equivalent of 250 x86 servers

Originally unveiled via paper launch in April 2016, the Open AI project will help to progress AI in everything we use, they mentioned understanding human speech and how it is interpreted e.g. Microsoft's Tay failure will be testament to such a difficult task. Not a feat you could set any system up for, the outlay of $129,000 see's a 'convenient' 3U form-factor with P100 providing a total of 169.6 TFLOPs (FP16) via eight Tesla P100 or 21.2 TFLOPS per P100 of compute.

DGX-1 comprises two Intel Xeon E5-2698 V3 as support is limited to two CPUs though not that it matters as a total of 32 cores and 64 threads is enough to feed Pascal's 16nm P100. Each of the eight P100's has 3584 CUDA's and 16GB HBM2 topping out at 720GBps, NVLink interconnect at 160GBps minimises bottlenecks while PCI-E 3.0 x16 tops out at only 32GBps - P100 supports both HBM2 and PCI-E. System RAM is the standard affair of ECC-LRDIMM (Load Reduced DIMM) as packing in 512GB DDR4-2133 would normally slow systems this doesn't penalise you either in RAM speed/amount of CPU frequency. Taking care of storage comes courtesy of four Samsung enterprise-class PM863 1.92TB (MZ7LM1T9HCJM specs) SSDs, peak system power consumption is 3.2KW with each P100 consuming 300W.


Signed from Jen-Hsun Huang: "To Elon & the OpenAI team! To the future of computing and humanity, I present you the world's first DGX-1"


“Artificial intelligence has the potential to be the most positive technology that humans ever create,” said OpenAI Chief Technology Officer Greg Brockman. “It has the potential to unlock the solutions to problems that have really plagued us for a very long time.”

Although the alternative could have been accomplished via distributed computing offering mixed results, OpenAI Research Scientist Ilya Sutskever said “Our advances depend on GPUs being fast. Speed of our computers is, in some sense, the lifeblood of deep learning."

Tackling AI is one area that would dominated by the few leading to a stranglehold of patents that could stagnate progression for potential fear that governments may have.



Share this post

Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in

Sign In Now