Back in the “good ol’ days” of computers, processor speeds were measured in Megahertz. CPUs from different manufacturers could be compared with one another, and newer CPUs could be compared to older ones.
The MegaHertz Myth
Today Megahertz has given way to Gigahertz, and single cores are giving way to multiple cores. To complicate things, CPU’s do so much more than they did in the past, some even being called a “system on a chip” (or “SoC” for short, but more on that in a moment). These SoC’s might include CPU, memory controller, and various digital signal processors. There is a huge money savings by combining all these functions into one package, but doing so further complicates speed measurements.
Most SoCs are used in “embedded systems” and more recently in smartphones and tablets, and it seems like everyone has one: Samsung, Qualcomm, TI, Nvidia, and even Apple. Most are based on the ARM Cortex-A9 and -A15 (or similar technologies). ARM differs from traditional desktop CPUs in many ways, but primarily in the instruction set (the kind of code) they can process. ARM is a 32-bit RISC (reduced instruction set computer) which stands for Advanced RISC Machine (and before that the Acorn RISC Machine).
Intel, the long-standing king of desktop processors, has been trying to get into the mobile chip business for quite a while. Their attempts fall under the “Atom” moniker — but Atom chips are CPUs, Medfield is an SoC platform. The CPU that Medfield uses is a hyper-threaded processor called Saltwell, the SoC package goes by the name of Penwell, the SoC plus the radio, camera, SGX540 GPU, and other components is called Medfield. There are various different models of Medfield. The Z2460 version is quick, clocking in at 1.6 GHz.
And there we are back to GigaHertz for speed measurements.
History has clearly shown us that frequency isn’t very telling when it comes to speed. Intel even went away from naming their chips based on clock-speeds, to model numbers. So how does one tell the true speed of a chip these days? Benchmarks.
Unfortunately, benchmarks measure lots of different things and they have their own set of problems when it comes to speeds. Quadrant, for example, is one of our favorite benchmarking tools for Android, but many of our readers have been critical of the scores it returns… and like all benchmarks, Quadrant can be “cheated” using various tweaks and optimizations on the phone to get a better score, but not necessarily better performance.
The GPU in Medfield, unlike its CPU, is nothing particularly noteworthy. Medfield utilizes the PowerVR SGX540 GPU (which is the same as the GPU in the TI OMAP4460). In short, Medfield’s graphics handling is on par with the competition. That’s not to say it’s not impressive, it can decode and encode 1080p video at 30fps.
To show manufacturers that Medfield is a viable option, Intel has put together a reference platform. The device won’t be available to end-users, but looks stable and performs well — very well.
The reference platform includes a 4-inch screen at 600×1024 (but not the HD resolution that the Galaxy Nexus has). Interestingly, the reference platform is currently running Android 2.3.7 with a 2.x kernel (not the Android 4.0 OS with a 3.x kernel like you’ll find in the Galaxy Nexus). Even so, the reference platform scores 3,800 in Quadrant.
In theory, the 3.x kernel should add speed, as should being fully GPU accelerated (which isn’t in Gingerbread).
It seems ironic that Intel is putting so much effort into bringing x86 to mobile devices while Microsoft is busy making it’s next flagship desktop OS (Windows 8) run on ARM chips. The future seems to be evolving towards ARM-based architecture, or at the very least, it should see RISC win out over x86’s CISC-based computing.
That said, having competition is a very, very good thing. Luckily, with Android at least, the typical ARM vs. x86 compatibility problems don’t seem to apply. Once the OS can run on either foundation, Android apps shouldn’t have any problem running in either. Why? Android is somewhat unique, using a virtual machine paradigm (VM for short) in which apps are run — decoupling them from the actual architecture of the system.
One thing we haven’t discussed yet is power consumption, and its impact on battery life. Intel has made huge advancements in reducing their chips’ size. Currently built at 32nm, some sources are reporting that Medfield will be constructed at 28nm in the very near future. The smaller the construction, the less power is required to power the platform, heat is reduced as well. Both contribute to improved battery life. Unfortunately, it’s probably to early to comment on battery-life in new Medfield-powered smartphones and tablets, but we expect it will be on-par with today’s high-end devices.