Believe it or not, I actually really like ARM Holdings (NASDAQ:ARMH) (and its weaker sibling, Imagination Technology (OTCPK:IGNMF)). Yes, that's right. I think the company has a fantastic business model, and frankly, I think the management team is top-notch. Even after writing some decidedly negative pieces about the stock, the ARM guys have always been exceptionally good about answering any/all questions that I have about the company, the business model, and so on. The stock is surely expensive (for good reason, mind you), but I'd really want to wait on a pullback into the low $40s/high $30s before taking a long position myself.
Anyway, with that out of the way, I'd like to step back and talk about how the ARM hyperbole with respect to Intel's (NASDAQ:INTC) datacenter and mobile efforts has gotten way out of control. Mind you, I don't blame ARM for this - the company quietly plugs away developing slick, clean ISAs, and a very impressive array of CPU, GPU, and physical IP, but I do blame a misunderstanding of technology on the part of many investors and the mainstream press. With that in mind, I present the following thesis and then use the remainder of the article to "prove" it:
Intel's struggles in mobile would not be aided in the slightest if Intel today took out an ARM processor and/or architectural license.
Intel Can Do Low Power CPUs...
The big "issue" that everyone seems to think that Intel has is this fundamental "inability" to design processor cores that match/exceed what the good folks at ARM (and its various partners) develop. While there may have been some doubt before Intel launched its current generation "Silvermont" CPU core, there is absolutely no doubt that X86 isn't a significant barrier to the development of a high performance/low power CPU core. Further, there is no doubt now that Intel's low power CPU team has some serious chops. Don't believe me? Well, take a look at how the Intel "Bay Trail" platform performs in CPU-heavy tasks:
It's clear that when it comes to CPU performance, the "Silvermont" processor core is the fastest/most efficient CPU core in the industry. This makes sense given that CPUs are Intel's core competency, and given that the company can also co-optimize its CPU designs with its process (which is world-class), I have no doubt that this CPU leadership will extend far into the future.
So, what is Intel's problem?
It's More Than Just CPUs
While ARM provides the CPU instruction set or actual CPU design, depending on what kind of customer you are, there is far more to a system-on-chip than just the CPU core. It's all about the entire SYSTEM. What makes up such a system?
- Image Signal Processor
- Cellular baseband
- Connectivity (i.e. Wi-Fi. GPS, Bluetooth, NFC, etc.)
- Sensor hub
If you look at Qualcomm's (NASDAQ:QCOM) Snapdragon 800, you'll see all of these integrated in:
However, if you look at Intel's recently released "Bay Trail" system-on-chip, you'll see that a lot of those blocks are missing (although it's mostly there):
Take a close look at both of these, and you'll see that Intel has a pretty good SoC with *most* of what it needs, but certainly not everything. There's no integrated connectivity, there's no integrated sensor hub, and there's no cellular.
For a higher-end tablet chip, this kind of configuration is OK, but unworkable for phones/low-end tablets. I expect Intel's upcoming "Merrifield" platform sporting the "Tangier" SoC to have more integration (i.e. sensor hub), but it still won't be quite at the same level as what Qualcomm and Media Tek (OTC:MDTKF) are doing today.
In short, Intel's problem - ironically enough given the hype around ARM - is everything except for the CPU core.
The Server Question
Another thing I'd like to address is the whole "ARM server" business. In particular, I'd like to point out a comment that I received on one of my prior articles from a reader that I believe is worth addressing:
You are so right. The should build a moat around their server/datacenter business but the only way to do this, in my opinion would be to license ARM and offer it as an alternative solutions. The only problem with this is that they are too hard-headed make this move so other companies are going to do it to them...
I find this point of view exceptionally puzzling. First off, in this comment, does the reader want Intel to,
- Take an ARMv8 architectural license and build a custom CPU core around it
- Take an ARM processor license (i.e. Cortex A-series) and build an SoC around that
In either case, though, I fail to see what advantages Intel would gain from doing so. In the server space, X86 is the dominant instruction set architecture (so all of the software is already developed/validated for X86 and not ARMv8), and I've already shown that Intel's low power CPU cores are actually at least as good (if not significantly better) than the low power ARM cores that we've seen from the likes of Qualcomm and ARM itself.
Further, Intel is able to offer X86 products in either small core flavor (i.e. Atom-based) or large core flavor (i.e. Core-based), both of which are instruction set-compatible (barring the more advanced SIMD extensions). This means for a datacenter where the workload is mixed (i.e. you need some high performance CPUs for one workload, and smaller, lower power CPUs for another), it's actually more convenient to have one instruction set architecture everywhere rather than to mix and match.
In short, not only is there no need for Intel to do an ARM server chip, but it would very likely have an inferior value proposition. This is also one of the main reasons that ARM will have a very difficult time penetrating the higher end/compute-oriented portions of the datacenter.
In networking/comms, today IBM's POWER and Imagination's MIPS are the dominant architectures, but X86 is gaining significant momentum and ARM is expected to be a major player, too. However, X86 already has a much stronger foothold than ARM does, and is likely to grow its presence even further before the first ARM-based chips are available.
Intel's mobile problems aren't about ARM, nor are they even about the CPU core/ISA at this point. It's all about Intel's ability to build a world-class SoC (something that the company, admittedly, still hasn't quite mastered) and to iterate flavors/variants of these SoCs much more quickly. Licensing ARM would do precisely nothing to fix Intel's problems and would only serve to limit Intel's ISA flexibility (i.e. Intel can add in new instructions whenever it pleases with X86...not so much with ARM). The company also invests very heavily in C/C++ compilers and code optimization tools that really squeeze the best out of X86 (so all of that effort would go to waste if Intel transitions its product line to the ARM architecture).