MediaTek has stepped up its bid for Qualcomm’s crown boldly, launching the mobile device world’s first 10-core applications processor, the Helio X20.
The product breaks all kinds of records, and highlights the intense innovation which is going on in the ARM-based market, as those firms blessed with an architectural licence come out with highly differentiated approaches.
In MediaTek’s case, this is not so much about the number of cores – which Qualcomm will no doubt dismiss as unnecessary at this stage of smartphone development, as it initially did with octacore and 64-bit. It is also about the ‘tri-cluster’ architecture of Helio X20, a different take on power management from Qualcomm’s, or from ARM’s Big:little.
With tri-cluster, MediaTek allows the chipset to scale its speed and power consumption up and down, depending what the device is doing at any one time. The CPU contains one cluster of two top-end ARM Cortex-A72 cores running at 2.5 GHz; plus two clusters of four A53 cores apiece. In one cluster the midrange A53s run at 2 GHz for medium processing power, while in the other, they run at 1.4 GHz to save power when the device is engaged in lightweight tasks.
Processing shifts dynamically between the clusters as the user moves from task to task. MediaTek gave the example of a Facebook user – scrolling through updates would use the medium cluster for the optimal balance of performance and power efficiency; watching a video post would require the high-end cluster; and when the customer paused and the screen went idle, the phone would switch to the low-end cluster.
All that is controlled by MediaTek’s CorePilot 3.0 heterogeneous computing software technology, the kind of system which most processor majors are seeking to develop (AMD, Freescale and others are prominent). The Taiwanese firm says CorePilot 3.0 results in 30% lower power consumption than a conventional dual-cluster approach (like Big:little). To improve battery life even further, the X20 also has an ultra-low power Cortex-M4 microcontroller integrated to support sensors and always-on applications like voice activation.
The SoC also contains a customized configuration of ARM’s latest Mali GPU cores, plus a dedicated hardware accelerator for 4K H.265 (HEVC and VP9) HD video decode and 4K/30Hz H.265 encode.
The product, which supports Category 6 LTE (up to 300Mbps theoretical peak download speeds), and two-carrier (2×20 MHz) aggregation, will go head-to-head with Qualcomm’s new flagship, the Snapdragon 820. The latter has a higher performance modem, moving to Category 9 and three-carrier aggregation.
That may reflect the US company’s particular strengths in modems, with would-be challengers more likely to find weak spots to attack in the area of processor design. It also reflects how the smartphone SoC providers may have dwindled in number significantly, but they still have to work harder than ever before to differentiate themselves and support new functionality for handset customers.
Inevitably, chipmakers will prioritize on different aspects of the increasingly complex smartphone chipset. Mohit Bhushan, general manager of US corporate marketing at MediaTek, acknowledged to FierceWireless that “in some areas the 820 will do better”, but insisted that Cat-9 LTE and 4K video were, at this stage in market development, unnecessary. They would add cost and design complexity but provide little benefit for the end user until more networks and applications support such features.
Qualcomm made very similar arguments when Nvidia beat it to quad-core and 64-bit, sticking to its habit of pushing the limits of modem technology while being more cautious around the processor (and look what happened in that battle …) MediaTek has considerable modem expertise too, however, and is a more formidable SoC challenger, though in 2014, its revenue share in smartphone apps processors was 14%, compared to Qualcomm’s 52% (according to Strategy Analytics). Apple, which designs its own mobile processor, had 18%.
However, MediaTek is making inroads on Qualcomm’s territory and achieved its first US carrier 4G certification last year, at T-Mobile, followed by the launch of its first US LTE device, the Alcatel OneTouch Pop Astro for TMo. Now the company wants to move into far higher value products than Alcatel entry level smartphones, which belong in the low end space MediaTek has traditionally inhabited. Adding Cat-6 and carrier aggregation was a must-have for Verizon and AT&T, Bhushan said, and the X20 will go to carrier labs, including US ones, for testing from September. The firm hopes to be in some commercial products in December, in time for Chinese New Year, though its powerhouse for the holiday quarter elsewhere will be its octacore family.
Some will question whether any smartphone really needs 10 cores, but the key to growth for MediaTek and Qualcomm will be to extend their technologies into other device categories with higher growth and margin potential than the hard-squeezed handset space. High end mobile SoCs are moving into products which would previously have relied on PC-class platforms and on separate apps and graphics processors – but which are now becoming highly mobile, and under pressure to reduce power consumption. Tablets, Chromebooks and other post-PC form factors are obvious examples, but also in-car systems and many embedded/IoT applications.
MediaTek’s challenge will be to prove that a 10-core implementation can be more power efficient than an octacore platform – if it does, that would not only wrongfoot Qualcomm and Samsung, but open up new potential applications, in areas such as mobile gaming and virtual reality, in which a few extra cores could drive new experiences, and for the first time, not carry an unacceptable trade-off in terms of battery life.