Today there are two primary players making the brains inside servers: Intel and AMD. And really, many argue, AMD is so beholden to Intel that AMD can’t effectively steer the market or compete even when it does innovate with something cool. But that’s about to change if Thursday’s announcement from ARM has the impact the chip-IP-licensing firm is hoping for.
While only Intel, AMD and Via Technologies currently can make x86-based chips, Mike Muller, CTO at ARM, says the ability to license an architecture freely could expand the number of companies building server chips to around 10. Plus, unlike with Intel’s x86 architecture, any company getting a certain type of license to the ARM architecture can tweak it to its heart’s content, which could mean far more innovation.
“ARM partners can build very different-looking solutions that take that core and integrate in different parts of the system … and that can significantly change things,” Muller said. “Another thing that could be interesting is they can offer a $100 chip and still get good margins, because they may come from a world of making $10 chips.”
As Muller alluded to, there could also be cost innovation in addition to technological innovation. For example, AppliedMicro will manufacture chips using ARM’s next-generation architecture that will deliver up to 3 gigahertz of performance for far less power than a comparable Intel chip. And the AppliedMicro chips are projected to cost less than what Intel sells its comparable chips at today. It is one of those ARM licensees that is used to making products for the embedded world where chips cost much less than those found inside PCs and servers.
However, when everything will be connected and require its own compute, both the power efficiency and the economics will need to change. “What’s interesting is it’s as disruptive from a business perspective as it is from the technology side,” Muller aid. “We have a different business model and [chips using our IP] will be sold in a different way that enables innovation as well.”
One big question will be if the webscale market will welcome such a wide variety of hardware platforms for its applications. Vendors have come and gone in the specialty computing space for the last few decades, and even the most successful couldn’t make much of a dent in the x86 market.
At scale, where computing seems to be trending, it’s hard to justify a lot of specialty hardware, so even variations on the same ARM architecture might be suspect. But they may not, and the resulting years should prove anything but boring in the server space.