ARM Tips Strategy to Crack Server Market
IP giant also talks about Intel, finFETs and the cloud
Needless to say, ARM Holdings is the dominate supplier of processor intellectual property (IP) in the booming cell-phone and tablet space. In a separate and bold effort, ARM is also putting the pieces in place to compete in the server world, thereby putting the IP supplier on a collision course with AMD and Intel in the x86 camp. To date, at least two major OEMs have announced ARM-based server programs. To get a handle on ARM’s server strategy—nd the various challenges ahead—Chip Design caught up with Ian Ferguson, director of server systems and ecosystem at ARM.
Chip Design: Which OEMs have announced ARM-based server programs?
Ferguson: In the latter end of Q2, Dell announced a platform, in which they call ‘Copper.’ This is a Marvell-based platform. And HP has been working with Calxeda. This is something called the ‘Redstone’ platform.
Chip Design: When will the ARM-based server market take off?
Ferguson: Boxes will start to ship this year. We’ve set the expectations at very modest volumes this year and next year. I am expecting we will see the analysts starting to measure our market share at a few percentage points in the 2014 timeframe. We look to grow from there. We think the addressable server market for ARM is about 20 percent of the overall server market. If I pick 2015 as a line in the sand, we can address about 20 percent of that market.
Chip Design: What server areas are you focusing on?
Ferguson: Our focus is on the cloud area, as opposed to your more traditional enterprise server. By cloud, I am talking about companies like Facebook, Google and Amazon. Traditional enterprise servers aren’t growing as fast, but it’s still the majority of the server market. The reason why I say that it’s really off-limits for us is that there is just a lot more legacy ties to x86 code. And that space is going to be a little slower to move.
Chip Design: What are the challenges in the overall IT arena?
ARM-based servers enter competitive market
Ferguson: The challenge in the IT infrastructure is security, whether it’s intentional or accidental. That results in systems going down and data being lost. You’ve also got the challenge of the cost of that infrastructure. I would argue that power is probably more of a concern. That’s forcing some slightly different decisions on how you fill that IT power budget with your equipment.
Chip Design: ARM brings a low-power solution to the party. But in the server world, ARM has the daunting task to compete against Intel. Intel is already making 22nm processors based on finFETs. ARM’s partners, and their foundries, are still doing 28nm planar chips. Isn’t ARM, and its partners, behind Intel?
Ferguson: Intel is making some claims about them having multiple years and multiple nodes in terms of leadership. I don’t think they are that far ahead. I think if you look at where our partners are with that, I don’t think it’s quite as big a gap as Intel is claiming. In addition, the foundries are making significant investments. There are some non-trivial investments from TSMC, Samsung and GlobalFoundries.
Chip Design: Can you elaborate?
Ferguson: ARM is working on finFETs. We have been looking at finFETs for quite awhile on some pretty aggressive nodes. You’ve seen quite a few announcements with finFETs with our foundry partners. One must also compare an SoC from Intel versus what the ARM community is doing. Doing a finFET on an SoC, we argue, is a different proposition than doing a finFET on a very-large, non-integrated SoC.
Chip Design: What do you see are the challenges in IC manufacturing?
Ferguson: There are incredible challenges for finFETs. You also have double-patterning and EUV. You have wavelengths of light that cause all sorts of challenges with layouts.
Chip Design: Getting back to ARM, how does ARM plan to crack the server world?
Ferguson: We see the emergence of a new type of server platform. We see people increasingly use higher level languages for programming their applications. People are using more Java. People are using more C++. That fundamentally starts to reduce their ties to legacy x86-based applications in servers. That allows people to consider new options such as ARM. That’s really our push around to what some people call ‘microservers.’ I particularly don’t like that term. I prefer the term that HP coined as the ‘extreme low energy server.’
Chip Design: What types of software is running on ARM-based servers?
Ferguson: One of those is a performance-optimized Java compiler from Oracle. They announced in August that the technology would be available with the same business terms as the x86 version. That puts the compiler and tools in the ARM community on the same level playing field as x86.
Chip Design: What’s the end game?
Ferguson: Software is king. There are now some ARM-based servers with that software being put up into the cloud. That allows customers to start testing their own applications running on an ARM-based platform. If ARM is to be successful, customers must see an advantage of that code running on an ARM platform, compared to an incumbent approach.
Chip Design: Where is cloud computing going?
Ferguson: The next challenge is how much of people’s infrastructure is going to stay inside the company versus outsourced to the cloud. People are talking about a great amount being outsourced into the cloud. Realistically, companies are going to keep some elements of their IT managed infrastructure in-house because of different reasons. It could be the security of the data, latency and reliability. I’m a believer of a hybrid cloud, as opposed to a complete outsourcing to the cloud.
Chip Design: What’s next for ARM and servers?
Ferguson: As we go into next year, you are going to see us doing more around 32-bit server platforms and 64-bit server platforms. 32-bit is certainly fine as a starting point. Marvell and Calxeda both have 32-bit chips today. For low-end server applications, 32-bit seems to be fine for Web hosting and storage servers. But clearly, 64-bit is needed for the larger portion of the server market. We have Applied Micro, which has announced a 64-bit ARM core for that space. Cavium has announced project ‘Thunder.’ We will be announcing our first 64-bit implementation cores. We will do that by the end of the year. We’ve talked about the names Atlas and Apollo, but we haven’t given any more details about those cores.
HP readies ARM-based server
Chip Design: Do the ARM-based chip makers need to be at the leading-edge of process technology to compete with Intel?
Ferguson: I think the majority of the ARM camp will do quite well a node or two back. Calxeda is not going to lose in the server business because they are not on 14nm finFETs. I think there is still a value proposition in the silicon. By integrating stuff on the silicon, I don’t think they have to be on the bleeding edge. I also see a lot of people at 40nm in my space. I think the 28nm node will probably stay for quite awhile.
Chip Design: What about multicore?
Ferguson: We are looking at how our technology scales up to multiple cores. If you look at the Cortex-A9, we could only have four processors on an SoC that were coherent. With Cortex-A15, we basically provided a path with an on-chip bus that allows you to scale up beyond four cores and keep them coherent. We are going to need to do that around 64-bit. To do that, we have an AMBA bus, which is the way people connect on-chip between the processor, the memory subsystem and peripherals. This can be a coherent connection.
Mark LaPedus has covered the semiconductor industry since 1986, including five years in Asia when he was based in Taiwan. He has held senior editorial positions at Electronic News, EBN and Silicon Strategies. In Asia, he was a contributing writer for Byte Magazine. Most recently, he worked as the semiconductor editor at EE Times.