"Nervana’s Product Overview Deck on their AI & Deep Learning Capabilities" in Humanizing Tech
Created in April before their acquisition by Intel in August 2016
We’ve done a lot of research into the field of deep learning and artificial intelligence, mainly because at its core it’s really just math. It’s a trendline for some data that helps you predict something. Deep learning is very good at very specific tasks, like reading handwriting, or keeping a car in between two brightly colored highway lane markings, but it’s not so good at more general things.
Most of the media doesn’t really understand that we’re dealing with a scatter plot chart in Excel, where you can right click and select “add trend line”, and that is the basis of what’s happening with artificial intelligence.
Deep learning will hit a limit, but in the meantime, it’s the hottest thing on the street. Heck, even I’ve profited from investing in NVIDIA early before their recent deep learning chips and software. I’ll write more on that in the coming weeks, with another announcement following shortly after that.
II. Intel Catches Up
Yesterday news broke that Intel made their move to catch up to NVIDIA by acquiring a young startup in the deep learning space called Nervana. The founder came from Qualcomm and had been doing some interesting things in the biological neural networking arena. The goal was low power usage without sacrificing massive computing capabilities, but Nervana’s focus was a platform directed towards regular enterprise business.
It proved a smart move in the short term. Nervana went on to raise $24.4M before the acquisition for a reported $350M. That’s a pretty good hourly rate and 10x return for a few years work.
The problem still remains that it’s expensive and time-consuming to train deep learning models on millions of data points. That’s with the innovations by Nervana. Just to get started with them it’s going to cost you $10,000 per month.
That’s why I believe eventually these large cloud services houses will start charging for power or data throughput instead of just “compute” time which is a bit of a black box. More on that in the Read More section below.
III. Nervana’s Product Overview Deck
We managed to get our hands on the deck they used for talking with prospective customers, which will give you some sense as to how they’re marketing themselves. Much of this is likely to get subsumed by the Intel mothership, but I imagine that the platform will remain independent.
Meanwhile I fully expect Intel to proceed quickly with integrating Nervana’s chip innovations into their own superstar GPUs.
Nervana’s Product Overview Deck on their AI & Deep Learning Capabilities was originally published in Humanizing Tech on Medium, where people are continuing the conversation by highlighting and responding to this story.
from Stories by Sean Everett on Medium http://ift.tt/2aLNrF9