Performance and Power
Since Intel’s previous platforms have been wholly unsuitable for smartphones, there was a great deal of skepticism surrounding Medfield. To dispel these notions (amongst other reasons), Intel has produced a reference phone platform (using an Infineon 3G modem) that is sleek, attractive and would accurately reflect real world products. The hardware is not available for independent third party testing yet, nor is the total bill of materials known. However, the reference Medfield platform at least allows Intel to make colorable performance and power comparisons to current smartphones. Once it is available for testing, the claims can be verified in greater detail.
Existing public mobile benchmarks are atrocious at best, and often misleading. Thus there is little motivation to provide exact numbers that might give a false sense of precision. Intel’s benchmark numbers for the Medfield reference design generally suggest a 10-30% CPU performance advantage and competitive graphics performance compared to the current crop of 40nm SoCs from Apple, Nvidia, Qualcomm and TI. Obviously, Intel’s claimed performance was higher; in some benchmarks the advantage was 50% or more, although those seem atypical. The energy efficiency advantage seems to be larger than the performance differences, as the dynamic power is lower due to Intel’s advantage in process technology.
Table 1. Medfield Power Consumption
In terms of power consumption for common tasks such as standby, talk, browsing, audio and video playback, Medfield has respectable numbers. Generally, the power seems to be lower than nearly all existing smartphones by around 5-15%. The one exception is 720p video decoding, where it is competitive and in the middle of the pack. That is likely because some of the latest phones are using newer and more optimized video codecs. Table 1 shows Intel’s claimed battery life for the Medfield reference platform, and a derived estimate for average power consumption. Again, the original numbers were provided by Intel, and can be measured once products are available.
The good news for Intel is that the Medfield reference design seems to be better than the current generation of smartphones. However, the first Medfield products are not slated until the second quarter of 2012. TSMC’s 28nm high-k/metal gate process is just beginning to ramp for complex designs like GPUs and SoCs, which means that in the latter part of 2012, the competition for high-end smartphones will be rather different. The current 40nm generation of SoCs from Apple, Qualcomm, TI, Samsung and Nvidia will be replaced by faster and lower power versions. Intel started sampling well ahead of their rivals (reportedly in February 2011), but the product windows will be fairly close.
This means that Medfield will face much tougher competition for most of its life. Realistically, Medfield will not have a decisive performance advantage over platforms like TI’s OMAP5 or the Snapdragon S4. At best, Medfield will be slightly ahead of the competition; but in many cases Intel’s performance may lag by 10-30%. The one area where Medfield is likely to retain an advantage is the ISP, which is highly programmable, unlike many competing designs. The main disadvantages for Intel will be microarchitectural, since competing SoCs will generally use the higher performance A15 or Krait CPU cores and updated graphics.
From a strategic perspective, this disadvantage is not a significant issue. Medfield is the first real smartphone SoC from Intel, and even if it had been released a year before, it would never achieve high volumes. To most of the handset vendors and carriers, Intel is not a well known partner. In contrast, TI and Qualcomm have been working on phones for a decade or longer. Few companies will risk a high-volume, flagship product on a relatively unknown partner. As Nvidia discovered with their entry to smartphones, the sole purpose of the first generation is to simply get a few design wins. These early collaborations are an opportunity to earn the trust of the vendors and carriers, which can be leveraged into commercial success in second and third generation products.
Medfield is a credible SoC for smartphones and is good enough to begin the process of building vendor and carrier relationships for Intel. This is particularly true, given Intel’s attractive roadmap. In 2013, Intel will ship a 22nm FinFET SoC with the new, power-optimized Silvermont CPU and the recently announced PowerVR Series 6 graphics. The rest of the world will ramp 20/22nm in 2014 at the earliest, a gap of 6-12 months. Judging by Intel’s plans for 14nm SoCs based on the Airmont CPU core in 2014, this process technology advantage is only likely to grow over time. Whether that advantage will yield a significant smart phone market share for Intel is uncertain, but Medfield clearly demonstrates that it is possible.
Discuss (68 comments)