Slashdot Mirror
Intel Lays Roadmap For 100-Fold AI Performance Boost With Nervana and Knights (hothardware.com)
MojoKid writes: Intel is laying out its roadmap to advance artificial intelligence performance across the board. Nervana Systems, a company that Intel acquired just a few months ago, will play a pivotal role in the company's efforts to make waves in an industry dominated by GPU-based solutions. Intel's Nervana chips incorporate technology (which involves a fully-optimized software and hardware stack) that is specially tasked with reducing the amount of time required to train deep-learning models. Nervana hardware will initially be available as an add-in card that plugs into a PCIe slot, which is the quickest way for Intel to get this technology to customers. The first Nervana silicon, codenamed Lake Crest, will make its way to select Intel customers in H1 2017. Intel is also talking about Knights Mill, which is the next generation of the Xeon Phi processor family. The company claims that Knights Mill will deliver a 4x increase in deep learning performance compared to existing Xeon Phi processors and the combined solution with Nervana will offer orders of magnitude gains in deep learning performance. "We expect the Intel Nervana platform to produce breakthrough performance and dramatic reductions in the time to train complex neural networks," said Diane Bryant, Executive VP of Intel's Data Center Group. "We expect Nervana's technologies to produce a breakthrough 100-fold increase in performance in the next three years to train complex neural networks, enabling data scientists to solve their biggest AI challenges faster," added Intel CEO Brian Krzanich.
BeauHD has reversed the blockquote on this. MojoKid wrote the summary and that is blockquoted. Then the actual quoted section from the linked article is not blockquoted. Should be the other way around if consistency is to be observed.
$5 / month hosted VPS on linux = awesome!
What a load of PHB bean-counter marketoid bullshit.
"Data scientists" still care jack shit about reality. 90% of today's "machine learning" [sic] work is about justifying the bias. The rest 10% is about tracking you and taking away your freedom, replacing your reality with an echo chamber, while making you pay for it.
We expect Nervana's technologies to produce a breakthrough 100-fold increase in performance in the next three years to train complex neural networks, enabling data scientists to solve their biggest AI challenges faster
100x faster generation of bullshit still achieves nothing but bullshit. It's not about always about how fast you do it.
Amazon (where Musk's OPen AI went) and Google already have a huge investment in focusing on video cards. Force them to use a new kind of logic and it will go really slow. Instead focus on the other technology driver. Get ten AAA games to advertise support for this card (with results visible in the game) and you'll kickstart development and sell fifty thousand units, while having the first new "must have" since Phys-X was folded in to graphics cards.
If video games influenced behavior the Pac Man generation would be eating pills and running away from their problems.
Is h1 2017? Half 1?
Neural Nets? What is this, the 1940s? All this AI "research" and neural nets are the best they can come up with? What a joke.
Please tell me some useful things that deep learning does?
Sounds like a rock group
Do the Knights include Sir Elton John, Sir Paul Macartney and Sir Rod Stewart?
"We'll kill the fathead, coughed the knight, so stick around"
Smells like teen spirit.
Show some love for Genetic Algorithms. Sure you may have to thoroughly verify that their 'better than a programmer' optimal solution is actually a correct solution but- they have existed for a long time even if as yet unspectacularly leveraged (in publicly visible ways that would have made them rank in your book)
The first half of 2017?
"Machine learning" is a computer algorithm and has little in common with the only real exemplar for AI .... us. It is being touted as progress in AI but it most definitely belongs with heuristic computing. This classification technique (one of a number all related in terms of distance measurement between samples) can benefit by using parallel techniques so that part of the story is OK. But please can we not fan the flames of AI again .. until we have understood that computers are only a tool in this activity and are not related to AI architecture. (IE we can use computers to make models for AI activities but the real system is tightly coupled to our environment via a suite of sensors and obeys a control system paradigm which handles events which are predominantly open loop).