Intel’s third quarter earnings report comes out on Oct. 24, but some industry experts have set their sights on years down the line to when investments in artificial intelligence could begin to pay off.
“In three-plus years, the real competition will be in AI and machine learning-related systems,” said Jack Gold, an analyst at J. Gold Associates, in an email to FierceElectronics. “That means if you’re a semi company, it’s where you need to have viable products going forward.”
Some analysts have focused heavily on Intel’s competition with AMD and Nvidia for performance in the graphics space, often on a PC or workstation. AMD is growing much faster than Intel, but still has less than one-tenth the revenues.
Intel has already shown there is a relationship between graphics processing efficiency and AI. Intel has yet to fully reveal its roadmap on that front.
In May, Intel told investors that Intel will have a discrete GPU coming in 2020 based on its Intel Xe architecture. But the big news probably won’t come until 2021 when Intel introduces a 7nm general purpose GPU for data center AI and high-performance computing.
That general-purpose GPU “will embody a heterogeneous approach to product construction using advanced packaging technology,” Intel Chief Engineering Officer Murthy Renduchintala told the investors at the time.
The 7nm process technology is expected to deliver two times scaling and a 20% increase in performance per watt with a four times reduction in design rule complexity. Intel will use extreme ultraviolet (EUV) lithography to drive scaling for the first time commercially.
An Intel spokesman used a generally accepted definition for a discrete GPU: it usually refers to an add-in GPU card for desktops or an added GPU (separate from a CPU) in laptops. This would contrast with an integrated GPU like the UHD 630 IGPU that comes with the Core i7-9700K CPU.
The spokesman also said a general-purpose GPU like the one coming in 2021 refers to using a GPU for tasks other than rendering graphics, despite its name. That includes AI model training or other high-performance computing work.
As for the 7nm approach that Intel expects to use, Intel has taken it on the chin from various analysts for not moving fast enough on building a 7nm chip as both AMD and Qualcomm have shipped. However, Intel has argued that its 10 nm is comparable to the competitors’ 7nm nodes out today.
Analyst Gold buys Intel’s central point. “On the 7nm issue, while it’s true that Intel has lost a good deal of leadership in the process wars, the size of the chip alone is not a key determining factor in how well it works,” he said. “By simply saying you are at 7nm and must therefore be better is a false statement. So much of the chip design process goes into the circuit designs and related technologies like transistor designs, fast compilers, and power management that you can’t point to process alone anymore as leadership.”
Gold added: “That’s not say smaller nodes aren’t better. They just aren’t the only thing you need to do to make better chips.”
Aside from node size, Gold is a fan of Intel’s heavy investments in next-gen technology and specifically AI and ML with FPGAs (Altera), neural chips (Nirvana) and autonomous driving (Mobileye). “Those aren’t huge revenue makers yet, but the future is pretty clear that if you’re a chip company and don’t have this kind of tech, you’re in trouble.”
Intel’s second quarter results announced in July showed the company had invested $3.4 billion in R&D for the quarter, with total revenues of $16.5 billion, which were down 3% from the second quarter of 2018. Intel earned $8.8 billion from PC-centric sales (up 1%) and $5 billion from data-centric products (down 10%), with Internet of Things revenues at nearly $1 billion (up 12%) and Mobileye revenues at $201 million, (up 16%).
Intel, based in Santa Clara, California, has about 110,000 employees, an increase from 104,000 a year ago. It was founded in 1968.