A couple of years back, I remember getting into some very heated discussions with those who believed that ARM -based processors were poised to disrupt Intel's near monopoly in the sever processor market. While the jury's still out as to whether it will happen -- and I personally don't think so -- I remember one very interesting argument that was made.
The argument was essentially the following: ARM's open ecosystem and vast number of licensees would allow the ARM ecosystem to collectively beat Intel by offering specialized chips for different workloads. In other words, Intel's big weakness is that it can't build the perfect chip for everybody, and that multiple players all customizing ARM technology would win out.
However, Intel's recent moves have shown that it, too, can play the customization game.
Two kinds of customizationWhen people talk about "customization" with respect to ARM-based designs, there are actually a couple of different things that people are really referring to. First, a chip designer can take an off-the-shelf ARM-designed processor core, lay down as many of them as needed, and then customize the non-CPU parts of the system on chip (memory interfaces, on-chip workload accelerators, and so on).
The next kind of customization is actually at the processor core level. If the off-the-shelf ARM Cortex A53/A57 cores aren't quite what is needed, one can do what Cavium Networks and Applied Micro have done -- build custom ARM-compatible processor cores.
Bears have, in my view, argued against Intel's ability to customize on both levels. However, Intel's execution has shown that it is very capable of doing both.
Customization at the core levelIntel builds two main lines of processor cores. First is its "big core" line that it used for PCs, traditional server chips, and other higher-performance applications. Next is the small core, i.e. Atom, that's used for low-end PCs, tablets, phones, and even low-power servers. These two cores have allowed Intel to cover a very wide performance range with its server chips.
However, Intel has shown a willingness to build what essentially amounts to custom processor cores for particular workloads. For example, for Intel's next generation Xeon Phi processor aimed at high-performance computing applications, it features a heavily modified Silvermont (22-nanometer Atom) processor built on its 14-nanometer process.
The modifications for this core are quite extensive, according to heise.de, including four-way simultaneous multithreading, an out-of-order floating point unit -- Silvermont has a slower in-order floating point unit -- a 512-bit AVX3 unit, and larger caches. It's essentially a new processor core.
Intel has shown that it will gladly build the right processor cores for a given workload if what it already has isn't well-suited for the task at hand.
Customization at the system-on-chip levelAt the system-on-chip level, Intel has also shown that it's quickly ramping up its ability to customize parts. Take a look at this slide from the company's recent investor meeting:
Intel claims that it has integrated external customer IPs into four custom versions of the latest "Grantley" products, with two more in development. It also claims that it is engaged with more than 10 customers to build full custom CPUs based on customer requests.
Intel clearly has the ability to integrate whatever its customers need it to for the workloads that they'd like, whether that IP is developed by Intel itself, or by the customers themselves. So, when you see companies like Amazon or Google hiring CPU/SoC engineers, it's very likely that they are working on custom IP blocks for inclusion into a custom chip from Intel.
Why Intel's business model works so wellIntel's business model is essentially near perfect for this market. It develops very strong off-the-shelf chips for many of its customers; but for customers or market segments that need more, it has the resources to customize things down to the CPU level. Intel also has, arguably, the world's best manufacturing technology and enough volume to fill its factories, which helps product performance and price competitiveness.
As an investor, it's always important to keep an eye on the competitive landscape in any company. As an Intel shareholder, I'm pretty confident that Intel will very deftly defend its datacenter turf against the various ARM players.
The article In the Datacenter, Intel Corporations Business Model Is Near Perfect originally appeared on Fool.com.
Ashraf Eassa owns shares of ARM Holdings and Intel. The Motley Fool recommends Amazon.com, Google (A shares), Google (C shares), and Intel. The Motley Fool owns shares of Amazon.com, Google (A shares), Google (C shares), and Intel. Try any of our Foolish newsletter services free for 30 days. We Fools may not all hold the same opinions, but we all believe that considering a diverse range of insights makes us better investors. The Motley Fool has a disclosure policy.
Copyright 1995 - 2014 The Motley Fool, LLC. All rights reserved. The Motley Fool has a disclosure policy.