For Super-Dense, Extremely Rugged, High-Performance Embedded Computers, Intel beats ARM Hands Down
We weren’t doing that because we wanted to be good citizens, although it was nice not to be beholden to any one company. No, we were doing that because it’s what the customers demanded. But over the last decade or so, the landscape has changed and these options have been diminished or removed. When it comes to the high-end open platforms that run myriad software and have to maintain an evolution path—forwards and backwards—there’s no disputing that Intel® is the king.
On performance alone, ARM is getting closer. There are some 64-bit multicore versions of ARM processors that are pretty powerful, but they’re not yet at the levels that our customers demand. In all the benchmarking we’ve been able to get access to for our kind of server-based data processing, Intel’s Xeon® is still the clear winner in terms of performance, although it does burn more power. That’s a tradeoff our customers are willing to live with.
In full disclosure, GMS does develop products with ARM-based processors for some limited embedded applications, such as subassemblies of our smart graphics RuggedView LCD panel computers, but there’s generally nothing on top of that—it operates semi-independently without impacting the overall system design. Again, it’s a fixed-function application that’s “closed” with no requirement for upgradeability or backwards compatibility.
A key reason that our customers are not asking for ARM is because it limits their choices of hardware, peripherals, and drivers. To implement something like the latest generation of PCI Express with ARM, or other high-end interfaces, you have to do a fair amount of design work.
One of the secrets to our success—with a thanks to Intel® —is how seamlessly we can evolve, for example, from fourth-generation Core™ i7 to fifth, sixth and future seventh generations, and to the new Intel® Xeon® D (Broadwell-DE). We have products based on all of these architectures in some form, either in production, sampling, or in design.
Note that Intel® gives GMS these processor evolution-based price and performance points that we extend to our customers. ARM’s architecture could never do that, mostly because the SoCs developed around ARM cores change from vendor to vendor, as do their pinouts and functions. Intel’s openness allows common software, drivers, as well as this critical evolution path. In the Xeon® world (as opposed to the ARM core world), it’s the same thing. We now have Ivy Bridge Xeons and Broadwell-PE Xeons coexisting. If Intel® does a Skylake-based Xeon®, we can do that, too, probably within a few months.
That’s not to say that the ARM architecture doesn’t have a place. But high-end open architectures will stay with Intel® for the foreseeable future. As will General Micro Systems.