In case you've been living under a rock, artificial intelligence (AI) has become one of the biggest buzz phrases in technology. Unsurprisingly, then, many computer-chip companies have been rushing to build chips that can quickly and efficiently handle artificial-intelligence workloads.
One of those companies is chip giant Intel (NASDAQ:INTC).
At CES 2019, Intel talked up a new chip that it's building, called the Neural Network Processor for Inference, or NNP-I for short. According to the company, "[This] new class of chip is dedicated to accelerating inference for companies with high workload demands and is expected to go into production this year."
A little while back, Intel said that it expected the total market for AI-oriented chips to grow from around $2.5 billion to between $8 billion and $10 billion by 2022. If those growth projections are even close to correct, it's clear that there's a significant opportunity here for the chip giant.
Let's take a closer look at what Intel has disclosed about the NNP-I and one thing, in particular, that could help the product (and its successors) be commercially successful.
Intel's getting help from a potential customer
Generally speaking, it's a good idea for companies to build products that others want to buy. A good way to make that happen is to work closely with potential buyers of those products to build them. That's what Intel seems to be doing with the NNP-I.
According to Intel, one of its development partners for the NNP-I is social-media giant Facebook (NASDAQ:FB). Facebook is a major buyer of Intel's Xeon server chips today and, in general, makes massive investments in its infrastructure. Indeed, Facebook is set to plunk down between $14 billion and $14.5 billion on capital expenditure (capex) in 2018, with that figure set to grow to between $18 billion and $20 billion by 2019.
A big part of that capex is spent on data-center servers -- servers that tend to have a significant amount of Intel processor content inside.
While having Facebook as a development partner doesn't guarantee that the NNP-I will be a commercial success, I'd imagine that it does improve the odds that Intel will build a compelling product and have a pipeline of good stuff in the works, too.
What do we know about the NNP-I?
Intel hasn't disclosed that much about the NNP-I, but what's interesting to note is that, in a tweet post-CES, the head of Intel's artificial intelligence products group, Naveen Rao, disclosed some key technical details of this product. You can see the tweet below:
Ok, guys...some news for you. NNP-i is 10nm Intel process. It will also include IceLake cores to handle general operations as well as the NN acceleration. It'll be a great product :) You're welcome.— Naveen Rao (@NaveenGRao) January 8, 2019
The chip apparently will be manufactured by Intel using its in-house 10nm manufacturing technology. It'll also apparently embed the company's next-generation Sunny Cove CPU cores. (These new cores promise significant performance improvements over the current-generation stuff and also incorporate some AI-specific machinery.)
Given that CPUs tend to be pretty good for artificial intelligence inference workloads, it makes sense for Intel to leverage its CPU technology to build a more specialized accelerator product.
If Intel can build a compelling AI inference accelerator with the NNP-I and consistently deliver improved versions of the product at a rapid clip, it could help the chip giant better capitalize on the market for artificial intelligence chips, potentially boosting the revenue and profits of its data-center group (DCG) compared to what they might be without this product line.