Medfield: the war begins x x86 ARM on smartphones

Introduction

With Intel Medfield starts another round in the fight between the x86 and ARM on smartphones and tablets. This is a fight that has been dragging on for some time, but so far no tangible progress for Intel. Apparently, the launch will mark the beginning of the fight for real.

In short, Intel has been a manufacturer of ARM processors, at the time of the Intel StrongARM and XScale, but she ended up selling the division for Marvel, considering the very low profit margins. By having a virtual monopoly in x86 chips, made more sense for Intel executives expand this market to compete with other manufacturers of ARM chips under equal conditions. Here we have an Intel XScale 80219, produced by Intel in 2006:

X86 superscalar processors like the Core 2 Duo that were successful in PCs and servers due to good performance were unsuitable for mobile devices due to the large energy consumption. To have any chance in the mobile market, Intel needed a simple processor with processing instructions in order. No time to create a project from scratch, they turned to the last processor in the processing order of business success: Pentium 1. Then came the Atom, which is nothing more than an upgraded Pentium 1, produced using current techniques of production.

Atom netbooks eventually produced which, although slow and ended up having limited success due to the relatively low prices. However, even the Atom was too wasteful for use in smartphones, and Intel had not yet mastered the manufacture of SoCs as well, offering only the processor and chipset, which need to be combined with several other chips (radio, etc..) Only to offer the ARM same features that manufacturers are able to offer in a single SoC.

Intel's first attempt to enter the smartphone business was the Moorestown . Technically, he was a two-chip solution, combining the Lincroft (which included the CPU, north bridge-and GPU) and Langwell (south-bridge chipset), however, he also had the Brierstown, a chip that ran the third management electrical and radio Wi-Fi/3G a PMIC, a total of five chips, not even rely on RAM, Flash and other components.

Clearly, the platform lacked the necessary levels of miniaturization, which explains the fact that he never reached the market. Intel came to present a prototype (quickly nicknamed Cyclops because of the odd shape), followed by prototypes from LG and Ahava, but none of them came to be used outside of demonstrations at trade shows:

The Medfield is Intel's attempt to throw a bucket of sand in the past and start anew. It offers a number of important improvements over Moorestown, which should give Intel at least a chance to show it belongs:

Instead of two chips, the Medfield is a SoC that integrates the CPU, GPU and chipset all components on a single chip, including any management circuits that were once part of Brierstown. He still does not include the radio nor the Wi-Fi/3G PMIC, but it already includes RAM integrated in the chip through a technique called the Intel package-on-package bonding, which is to "paste" a chip memory directly on the top of the processor die, which not only eliminates the use of trails, but also reduces access latency.

The initial model of Medfield, which will be marketed as the Atom Z2460 is based on a single-core Atom 1.6 GHz, with a SGX540 GPU, which is the same used in the Galaxy Nexus, but operating at 400 MHz in order to put the Medfield ahead in benchmarks. The combination of the GPU with a dedicated video decoder make the chip is able to decode 1080p video at 30 FPS, encode 720p at 30 FPS and also offer video output up to 1920x1080 via HDMI.

Besides the built-in memory, it supports LPDDR2 RAM-800 interface via a dual-channel 32-bit, which is similar to that currently used in high-performance ARM SoCs. As usual, "Medfield" is the name of the platform, which includes the SoC, radio and other components, the SoC itself is called "Penwell" and the processor inside a "Saltwell"

Intel released some numbers with the performance of Medfield compared with some recent devices, and some tests based on the reference platform had been posted by Anandtech, with similar results to those presented by Intel.

For now they are just an indication that performance is competitive Medfield clock per clock compared to dual-core ARM SoC we have currently on the market, which explains the need for Intel to have stretched up to 1.6 GHz clock (instead of 1.0 or 1.2 GHz that would be expected in an ultra low power version of the Atom), so that the chip stay ahead of the competition in the direct tests. As usual, only the 1.6 GHz is the maximum frequency of the chip, since the Medfield system offers a very advanced power management, capable of dynamically adjusting the frequency in steps of 100 MHz to keep the processor operating at low power stages as long as possible. Even the nominal 1.6 GHz chip are kept only for short periods of time, exceed the TDP as most of the platforms to which it is dedicated. The maximum frequency that the chip is able to maintain for long periods is only 1.3 GHz

It is important to emphasize that despite being a single-core chip, the Medfield includes a processor with two processing units and support hyper-threading, which effectively allows it to process two threads simultaneously. Combined with the fact that he is a chip more complex than a Cortex A9, it is no wonder that the performance in many tasks is similar to or even superior to that of a dual-Cortex A9 the same clock.

Another point to bear in mind is that the Medfield is produced using a technique of 32 nm, while many of the ARM direct competitors are still produced using techniques of 45 nm 40 nm, giving Intel a temporary advantage, until Qualcomm, Samsung & cia. placing on the market of 28 nm chips, recover the advantage until the beginning of 2013, as Intel is expected to market 22-nm chips, and the cycle. It is also important to remember that for now is competing with the Medfield SoCs based on dual-core Cortex A9, outdated platforms front of quad-core SoCs like Tegra SoCs based on 3 and Cortex A15, which should arrive on the market shortly. Anyway, it'll be interesting to see direct comparisons between Medfield and ARM SoCs, since this is one of the first situations in which chips x86 and ARM chips will compete directly, and similar devices running the same operating system.

Preliminary figures for the reference platform, show a promising scenario for the Medfield in relation to electricity consumption, since the platform provides a total consumption in the home of a watt in basic tasks, similar to that offered by devices like the iPhone and Samsung Galaxy 4S S II, which are based on dual-core SoCs. Of course, part of it is because of the Medfield be a single-core processor, but if he is able to sustain a similar performance to competing dual-core, it will revert into an advantage.

Intel seems to be convinced that Medfield may be used in devices with the same size and weight of the ARM competitors, leaving the decision of the quarrel over details such as performance, price range and other features. The reference platform offers a sample of this: a smartphone "normal" that could even be launched on the market:

So far, Motorola and Lenovo have shown plans to produce Android based Medfield. Lenovo is announcing two models: K800, a high-end smartphone with a screen 4.5 "720p and running Android 4.0 and also the IdeaPad K2110, 10 a tablet," also based on Android 4.0. Motorola in turn announced a strategic partnership "for several years and multiple devices," and signaled that it intends to launch the first unit from the second half. These alliances are far from being something dramatic like in the case of Nokia + Microsoft, they are more a "let's see how it goes," indicating that the two companies are willing to complement their line of devices with some ARM-based products for Medfield test the market acceptance.

Interestingly, the lock-in in relation to the instructions of the processor, which Intel has always enjoyed in the x86 world is one of the problems she faces in competing with ARM as the architecture to be used more in the mobile world, is the ARM that enjoys greater familiarity on the part of developers and better compatibility with the operating systems and applications. Even in the case of Android, the build for ARM processors is the main, while the x86 is to build a smaller project, kept more strategic purposes, which receives far less attention and resources. Of course, with all the investment from Intel this situation will change, but the starting point is this.

Even in relation to applications there will be problems ahead, as though 90% of applications to run directly on the Android Dalvik (which is platform independent), the remaining 10% were written using the Android NDK (which lets you use low-level instructions ), not to mention a host of other applications that use a lot of ARM code, that also need to be modified and recompiled to run on x86 processors.

As an emergency solution, Intel is implementing a system of binary translation ARM> x86, which will run a large number of incompatible applications, in exchange for a reduction in performance. As we know, systems like this are a double-edged sword, because on the one hand increase the number of applications available at first, but then discourage developers to port their applications, making the performance degradation becomes a problem permanent.

In other words, is that Intel will have difficulty entering this new market, and to show real advantages over the ARM SoCs to achieve his place in the sun.

As much as the Medfield is successful, he can not expect anything more than being "another", competing with the dual-core ARM SoC and trying to win some battles. Intel knows it faces an uphill fight to win their share in smartphones and is working on a long-term strategy to go slowly gaining ground, with each new generation of architecture.