AMD, the chip company that has spent its history in Intel’s shadow as the only other holder of an x86 license, said on Monday that it has taken an ARM architecture license. That license will allow AMD to design its own ARM-based chips as opposed to just using the available ARM cores. Qualcomm, Apple, Samsung, Applied Micro and a dozen others hold such licenses. The first of AMD’s chips built using that architecture license would be offered in 2016.
AMD also showed off its first chip using a standard ARM core (the A57 core) designed for servers and said it would be out later this year. Significantly, although not surprising for those following the ARM server progress, AMD showed off the server running all kinds of traditional workloads including running WordPress, playing videos and more. This is a big deal because for decades, Intel helped keep its x86 dominance because server jobs weren’t built to run on the ARM instruction set.
A marriage of two instruction sets
Finally, AMD also announced what it calls Project Skybridge, an effort to make an x86 and ARM-based chip that is pin compatible. This means you could swap out an ARM-based CPU with an x86 CPU on the same board. At first the Skybridge framework will be aimed at the embedded and client market including tablets and computers, but it could also bleed into servers over time.
Patrick Moorhead, founder, president, principal analyst at Moor Insights & Strategy looks at Skybridge as a competitive differentiator for the existing hardware players offering laptops or tablets that currently must choose to design for either the x86 or the ARM platform.
AMD’s architecture license for ARM plus single-socket SkyBridge capability makes them unique in their approach. This can save both OEMs and ODMs a lot of money by only having to design, build one platform. Customers will still have to do software validation, but it’s a huge hardware savings.
AMD is wrapping all of these efforts under what it calls its ambidextrous computing roadmap. It’s an admission of how much the chip world is changing as a result of the rise of mobile, the cloud and of our ever-increasing need for computing. The question now is how can AMD best bring its experience in x86, ARM and even graphics processors to the market? Unless Intel takes an ARM architecture license or starts handing out x86 licenses, AMD is the only company that can design on both instruction sets.
A new future for AMD’s webscale efforts
On the server-side the architecture license and the possibility of AMD bringing a custom core to market becomes more interesting given the attention that the webscale world has paid to the ARM architecture, with Google, Amazon and others hiring chip designers familiar with ARM architecture in a purported bid to build custom server chips.
While, it’s not clear if Google and Amazon have taken an ARM license that they would need to build a custom core, they have hired people or are looking for people with the ability to design a core. What if those engineers end up working with AMD? AMD has a custom-chip business that makes the processor inside the Sony PlayStation 4 and AMD estimated that the semi-custom chip business would generate a fifth of its 2013 sales.
Bringing the ARM architecture license, plus the experience AMD hopes to get building the Skybridge framework offers clients a significant amount of design expertise and flexibility. Combine that with AMD’s experience building chips aimed at the server market and it’s possible that Google and other webscale companies could see a reason to turn to AMD instead of building chips all on their own.
As we have said many times prior, building a custom server chip isn’t simple. There’s the core design, the design of the entire system, the software running on top of it and an understanding of how those chips might interact in a data center environment where there are millions of possible nodes.
In writing last week about Amazon’s plans, Moorhead suggested that Amazon might be thinking not just about the chips, but about a fabric to span its data center. AMD has that fabric experience thanks to its purchase of SeaMicro, which made a very dense server using low-power Intel Atom chips. The magic however, was in the fabric that allowed those small chips to communicate together to perform jobs.
The elephant in the room
Other sources in the chip world have suggested that this focus on fabrics is why Amazon and others are after as they hire chip designers and folks with fabric expertise. It’s not crazy to see that AMD is setting itself up as the design shop prepared to offer all of the pieces for custom-silicon to go into the data centers running the web’s highly distributed applications.
But will these webscale clients, who are among the top clients for Intel today, commit to AMD? Even with that uncertainty, AMD’s announcements put that ball back in Intel’s court. I’m eager to ask Diane Bryant, the head of Intel’s server division about its plans at our Structure 2014 event in June. While it may feel like Intel is under siege, the direct competition is still a ways off.
AMD isn’t going to launch the Skybridge compatible framework until 2015 and is looking at a 2016 time frame for a custom ARM-based core. Moves like AMD’s and IBM’s efforts with the Power architecture, provide a looming thereat, but Intel still has time to counter with a more modular or custom strategy of its own. I can’t wait to see what happens next.
Updated on May 6 with a headline that offers more clarity on what AMD is doing with the two instruction sets.