ARM’s diverse group of partners will beat the monolith of Intel, says ARM R&D chief Krisztian Flautner
Although Intel is still the undisputed king of silicon, it’s a surprising fact that the ARM processor design is in more 32-bit chips than Intel’s x86 design, thanks to ARM’s dominance of mobile devices and tablets.
The British company has no silicon fabrication plants, but licences the design for the ultra-low power ARM core to other companies, including Samsung, Qualcomm and Apple, to make processors for their devices.
The competition with Intel is heating up right now, for two reasons. Smartphones and tablets are increasingly dominant devices for people to actually use, and at the same time, the low-power demands which make ARM such a good fit for smartphones and tablets now look a good fit for data centres.
In the giant data centres run by the likes of Google and Amazon, electrical power is the main consideration, and a lot of the work can be done best on masses of cheap, low power devices. So ARM-based processors are being proposed for data centre servers, by the likes of Calxeda and Dell, and through a deal with the number two server processor maker, AMD.
With all this going on, can ARM really keep ahead of Intel? We figured the best person to ask might be Krisztian Flautner, head of ARM’s research and development – but his answer was that it’s not the technology that is hobbling Intel. It’s the way it does business.
“It is really about the business model, and less about the product,” Flautner told us. “We are enabling a broad set of people to compete with Intel, and they are not really used to that.” So while AMD has struggled on its own to keep enough research and development going to compete with a company the size of Intel, the collective efforts of the ARM ecosystem can stay ahead in their chosen markets.
ARM’s deal with AMD is just one of many, and that sort of competition is not what Intel is used to, said Flautner. “They emphasise the core technology and try to push the lock-in business model. We are the David of the situation, with Intel as Goliath. That is how disruption happens.” The ARM ecosystem is one where everyone’s specialisations help the others, he explained: “It is much better to do this work in collaboration, rather than in a monolith.”
Are there ever strains in the ecosystem? For instance, what if Apple wanted special smartphone features for its own ARM-cored chips? “Our interest is in licensing the product broadly,” said Flautner. “Our model is about breadth, we want to proliferate. Managing expectations between partners? That is what we are good at! We have to be sensitive to their concerns and we are very good at extracting the common benefit.”
In the end the ARM process is a partnership of people with different ideas, and comes out with a consensus that works – “at least partially” – for everybody, he said.
So is it possible that ARM could win? “I am hoping that the ARM partners are going to be the winners in this conflict,” said Flautner. “I don’t see fundamental reasons why they shouldn’t be. ARM is really quite good compared to what Intel has been doing. They are a good tech company and will be improving their product and getting some wins.”
But if the business model is the problem, does he think Intel is doomed to fail against ARM unless it starts doing business differently? “Their business model will make it difficult for them unless they come up with a different strategic plan.”
What about the technology then? How will ARM chips enter the server arena and how will this change data centres? “It will be a mix,” he said. “It is quite heterogeneous, and there are lots of different server requirements and use cases. For example, we have looked at a number of workloads for the memcached memory caching system used by people like Facebook.
“There are large efficiency gains available if you build around ARM cores. What will end up happening is those niches will have ARM products, and it will become a broader play. If you look at how ARM has proliferated, it has gone into one segment after another and we will see the same behaviour, as an SoC [system on a chip] optimised for each segment is developed. But it is early days right now.”
At first, the use of ARM cores won’t change the design of data centres: “You won’t have to change – but there may be an opportunity to build things better in future,” he said. “That’s usually the way these things go.”
AMD is a big new partner, with something of an iconic name as Intel’s longest term competitor, but it won’t get any special treatment in research and development said Flatner: “I don’t see AMD differently from our other silicon partners. We try and engage them in conversation and areas of mutual interest. I don’t see that much of a difference from any of our other partners.”
AMD is very welcome, he hastens to add – as its strength in the server and high-availability markets is not something ARM has historically had.
How is ARM’s research and development organised, with this diverse set of partners? “The way R&D is organised in ARM is we do a little bit of everything We work on silicon technology, we build test chips, and engage foundries early to drive it forward. There is a lot going forward in the next generation of micro-architecture, and we are trying to engage on all levels for a clear understanding of the problem.”
ARM’s research includes work on predictive process models, which have the ability to predict accurately what might happen at a silicon node. “This determines how to build the core, and what material you should use, what will be the bottleneck and what characteristics will the processor have,” he said.
ARM also works on silicon modelling having put a lot into the gem5 open source simulation effort. “We have made it R&D ‘s objective to have high-quality simulations of ARM cores available for design studies,” he said. “Engaging people in conversation at this technical level, avoids problems down the line in future – and guides our own research objectives.”
Flautner is very excited about the Raspberry Pi. He agrees it’s a niche product – but it is encouraging the kind of people that he wants in the ARM community. “We were very impressed by what they built and the attention they get. It engages an important sector of people: people who are tinkerers, and the younger generation. We are very happy that they pulled off their project – and a lot of ARM people gave their time to help out.”
Where will all this go in future? What are his research and development goals?
“We want to know what will drive computing workloads down the line, we need to try to get a handle on what will be the requirements five years from now,” he said. “With augmented reality and similar things, do we have to do things differently?”
He wants to follow questions of improving efficiency, and using GPU computing in a mobile device. Also, while there’s a lot of focus on servers which are big (at least by ARM’s standards), he is keen to look at the other end of the spectrum, and push ARM down into smaller and cheaper forms, where it can be part of the “Internet of Things”, in micro-controllers and monitors: “We are not just innovating at the high end.
“If you only have a few tens of thousands of transistors, what does a micro-controller need to enable the next generation of the Internet of Things?” he asks. It’s the usual story of making something that is cheaper – or more functional for a cost of around ten cents – and as small as it can be, but someone needs to think about what extra functions will be needed in sensors.
One thing Flautner believes is that sensors for the Internet of Things may need to have security built in, no matter how small the transistor budget. “We are trying to understand what extra functionality you might need, and we think there is a need for securing some of these sensors, there is always an energy efficiency angle, and an issue of how you integrate with wireless comms and sensors.”
So when Flautner talks about diversity he is not kidding. With everything from sensors to supercomputers, ARM wants to be everywhere.
Euro Story: each week, we publish a selected story from across NetMediaEurope’s network of European sites. This week’s story came from TechWeekEurope UK.
How well do you know ARM? Try our quiz!