MediaTek Helio P60 SoC offers AI acceleration for all

In the mobile SoC world you mostly hear about Samsung's Exynos, Qualcomm's Snapdragon, and Apple's A-series SoCs. Those companies aren't the only ones making miniature microprocessors for mobile systems, though. At MWC, MediaTek announced its Helio P60 SoC. The new chip seems aimed to compete with the heavywights discussed above, as it comes with an AI accelerator not unlike those found in recent ARM-based offerings.

First off, the base specs. The Helio P60 is an eight-core SoC with four ARM Cortex-A73s and four Cortex-A53s. As usual, they're set up in ARM's big.LITTLE configuration, meaning that the Cortex-A73s spin up to handle computationally-demanding tasks and let the ultra-low-power Cortex-A53s run the rest of the show. All eight cores can be configured to run at up to 2.0 GHz. The graphics hardware in the Helio P60 is likewise based on an ARM design, in this case the Mali-G72 MP3.

More specifically, the company claims the Helio P60 has 70% faster CPU and GPU performance than its predecessor P23. The new Helio P60 should also be around 12% more power-efficient as its older brother and is 25% more efficient while gaming. MediaTek says the Helio P60 is fabricated on TSMC's 12-nm process, and that the new chip is its most efficient Helio P-series offering yet.

Of course, the implementation details will be up to device vendors, but those companies will be able to hook up the Helio P60 to up to 8 GB of LPDDR4X memory running at up to 3600 MT/s. The SoC also supports a single camera with up to a 32-MP sensor or dual cameras with sensors as big as 20 MP and 16 MP. Like most of its contemporaries, the Helio P60 has myriad specialized application processors built in, and its image signal processor supports HDR recording as well as AI-powered image enhancements.

Those image enhancements come courtesy of the newest feature to grace the Helio line, MediaTek's NeuroPilot AI accelerator. Like similar functional units in other companies' SoCs, NeuroPilot is a DSP intended to accelerate AI inferencing. MediaTek suggests that besides improving pictures, NeuroPilot could also be used for real-time overlays and AR acceleration. NeuroPilot supports Android's NNAPI as well as the TensorFlow, Caffe, and Caffe2 AI frameworks.

MediaTek aims the Helio P60 directly at the smartphone market, and to that end it includes a 4G LTE modem with dual-VoLTE support. You won't have to wait long to see the Helio P60 in phones, either—MediaTek says you should start seeing devices with it pop up as early as next month.

Comments closed
    • albundy
    • 2 years ago

    sounds like this will be used on the China brand phones. i’ll definitely be moving to one soon, as the koreans have decided to copy apple in every single way and remove as many features as possible, thus making them literally the most boring phones ever made, while still holding astronomical pricing. I’m thinking the Asus Zenfone 4 Pro or go with a Huawei P20 on my next upgrade.

    • ronch
    • 2 years ago

    I honestly wonder why we need all this newfangled AI thingies in our phones. I just need to check my text messages, make phone calls, surf the web, watch videos. Do I need AI or does someone need it to keep an eye on me?

    • Tom Yum
    • 2 years ago

    Given that A75/A55 were released in May 17, little surprised that MediaTek are just releasing a A73/A53 SOC now. The Kyro 385 used in the 845 is based on A75 according to Anandtech, so not sure how this SOC is meant to compete with the current ‘heavyweights’ , maybe the previous gen ones.

      • Shobai
      • 2 years ago

      Someone call chuckula, he’ll tell you all about it!

      They may just be behind the times, but another consideration is that the A73 core isn’t affected by Meltdown while the A75 will be (if / when any devices with A75 based cores release), apparently .

        • NTMBK
        • 2 years ago

        Surely most Chinese phone companies would see the ability to snoop on all your activity as a feature, not a bug.

          • Shobai
          • 2 years ago

          Heh, unlike anyone in the US, I guess?

        • tipoo
        • 2 years ago

        It doesn’t seem like
        1) Meltdown would be enough to bridge that gap in 85% of workloads and 99% of phone workloads
        2) They could have reacted that fast in building an SoC, they learned of it before the rest of us but SoCs take a few years to design and validate.

        Just running late seems most likely, or else cost/licencing fee deals. Low margin players = low working R&D is my guess.

          • Shobai
          • 2 years ago

          A renewed focus on their core business, as per their announcement last year, is the likely cause.

      • watzupken
      • 2 years ago

      MediaTek is known for producing cheap SOCs, and not into fighting with the top dogs. In fact, I believe they recently mentioned that they will not compete at the high end for now, and will focus more on their mid end SOCs. So that probably answers your question.

        • Shobai
        • 2 years ago

        [url=https://gearburn.com/2017/11/mediatek-break-flagship-chips/<]That they did[/url<]

    • tsk
    • 2 years ago

    AI will be the most misused word of 2018.

      • soccergenius
      • 2 years ago

      [url=https://www.theverge.com/2018/2/27/17057582/asus-zenfone-5-price-release-date-lite-mwc-2018<]The Verge: The Asus Zenfone 5 is a shameless iPhone X clone with loud speakers and overreaching AI claims[/url<] [quote<]When I queried Asus on what’s “AI” about those functions, which already exist in other phones, I was told that the company is “adopting a broad definition of AI.”[/quote<]

      • Redocbew
      • 2 years ago

      Everytime I hear about an AI processor I think “so what does it do, exactly?”. A thing does not just “do” AI, and most things that have some learned component or otherwise spun off from what might be considered research in AI end up having names of their own like facial recognition, or voice automation.

      I guess I’m weird.

        • fyo
        • 2 years ago

        “What does it do” is very, very simple in the context of SoCs aimed at Android: Accelerate NNAPI.

        I’m really looking forward to some benchmarks comparing the various solutions out there.

      • fellix
      • 2 years ago

      Yet another popular buzzword to be slapped on any remotely related product or service. Just like “3D” was beaten to death for years and “digital” before that.
      It’s just an ordinary DSP accelerator with fancy instruction set.

      • strangerguy
      • 2 years ago

      But somebody needs to authoritatively empowering “outside the box” sustainable cloud-centric with AI-accelerated synergistic machine learning IoT platform.

      • watzupken
      • 2 years ago

      It is surely one of the most misused words. The term is loosely used and slapped on marketing materials to make it look good. May be we should have a benchmark to measure how smart is the AI now.

      • Stochastic
      • 2 years ago

      I think a more accurate term would be pattern recognizer. Isn’t that essentially what these deep learning “AI” systems are?

      • dpaus
      • 2 years ago

      Well, you’re the first offender: ‘AI’ is not a word 😛

        • chuckula
        • 2 years ago

        I beg to differ!

        — Weird Al

      • psuedonymous
      • 2 years ago

      Easy mode: every time you see “AI”, read “Neural Network”, and every time you read “Neural Network” think “like a Floating Point Unit but designed to do matrix math in parallel”.

      And if you’re thinking “that sounds useful and will probably end up ubiquitous, but it;s not really ‘exiting’ or a fundamental change in computation” then congratulations! You now have a good handle on the current state of ‘deep learning’.

      • wingless
      • 2 years ago

      Go watch a Computer Chronicles from the late 1980s. The TERM ‘Artificial Intelligence’ cannot be more misused than it was back then! LOL

      • Chrispy_
      • 2 years ago

      RGBLEDAI or GTFO.

Pin It on Pinterest

Share This