In what looks like a noteworthy shift for the chip giant, Intel says it is inside the custom silicon business, showing off its current efforts big customers, and its future plans at a media event Monday. The event featured Intel’s webscale and cloud executives Jason Waxman (pictured) and Diane Bryant discussing how Intel is now offering custom chips for customers: something that would have been unfathomable even a decade ago at the storied company.
Waxman detailed how Intel made a custom chip for eBay and said it was doing the same with Facebook. Customized chips are gaining ground as data centers grow larger and as efficient computing becomes paramount for companies at which compute is often the primary cost associated with their services. Companies are building out custom servers for different workloads and eyeing the use of different chip architectures — from ARM to GPUs.
This represent a huge threat to Intel, which has struggled to release a credible massively parallel GPU-style chip (the Xeon Phi is what it has on offer today) and seemed to miss out on the benefits of using smaller, low-power cores for certain webscale workloads until it was pulled into the market by SeaMicro, a startup later bought by AMD. Intel’s biggest advantage in chips has historically come from its ability to churn out billions of them as cheaply as possible thanks to its massive economies of scale and huge investments in factories from the profits of the PC industry. Customization breaks that model.
Basically Intel is trying — with its Xeon, Atom and Phi families of different x86-based processors — to offer a modular set of capabilities that can be integrated onto a system on a chip (Intel touted its SoC capabilities today as well). But it’s unclear how customized Intel is willing to be for customers, especially when compared to the custom chips its rivals are making.
For example, AMD this year formalized a custom-silicon business that is making the chips inside the Playstation and the Xbox, combining different CPU cores with a graphics processor in a system on a chip. AMD said that the custom silicon business should make up 20 percent of its revenue this year.
It’s also not the same level of customization that GM and Corporate VP at AMD Andrew Feldman envisioned when he spoke of customers choosing to create custom chips using ARM cores. In that case one can build a unique ARM-based core for a customer.interview in May with The Register
“As far as the etching goes, we have done different things for different customers, and we have put different things into the silicon, such as adding instructions or pins or signals for logic for them. The difference is that it goes into all of the silicon for that product. And so the way that you do it is somebody gives you a feature, and they say, ‘Hey, can you get this into the product?’ You can’t do something that takes up a huge amount of die, but you can do an instruction, you can do a signal, you can do a number of things that are logic-related.”
But as far outside of Intel’s comfort zone this step might be, Feldman, who was the former CEO at SeaMicro, the startup that pulled into the microserver market, downplayed the impact of Intel’s change of heart. After I asked his thoughts on the topic he wrote in an email:
“These are minor tweaks of existing parts, predominantly, binning and soft strapping. What AMD did for the gaming business is create a fundamentally new chip—custom cores and different combinations of cores on a chip. What ARM enables is unique cores for a customer … The Intel model can’t do that. What they can do is try to make a superset chip, and enable and disable features for different customers … this is what they have always done.”
Even Patrick Moorhead, an industry analyst, questioned the depth of customization that Intel was offering, pointing out that it was very different from the customization that AMD is offering in its business.
When I asked for details from Intel, the company offered me an interview later this week with Waxman. So stay tuned.
While I wait, here’s why Intel talked about its custom chip business today. If Intel is really doing custom chips, it will do so because it has no other choice than to meet the needs of webscale customers, who are now in the position to demand changes in their data center infrastructure –something the server vendors have already discovered. If Intel is just spinning small variations made by turning off features on a chip using software, it’s because it’s scared the competition will convince data center operators that the competition has something Intel doesn’t.