Continuing our coverage of Intel’s 14nm Technology, another series of press events held by Intel filled out some of the missing details behind the strategy of their Core M platform. Core M is the moniker for what will be the Broadwell-Y series of processors, following on from Haswell-Y, and it will be the first release of Intel’s 14nm technology. The drive to smaller, low powered fanless devices that still deliver a full x86 platform as well as the performance beyond that of a smartphone or tablet is starting to become a reality. Even reducing the size of the CPU package in all dimensions to allow for smaller devices, including reducing the z-height from 1.5mm to 1.05 mm is part of Intel’s solution, giving a total die area 37% smaller than Haswell-Y.

The first wave of three Core M parts will all be dual core flavors, with HD 5300 graphics and all within a 4.5W TDP. For Core M Intel is no longer quoting the SDP terminology due to the new design.

Intel Core M Specifications
  Core M-5Y70 Core M-5Y10a Core M-5Y10
Cores / Threads 2 / 4 2 / 4 2 / 4
Base Frequency / MHz 1100 800 800
Turbo Frequency / MHz 2600 2000 2000
Processor Graphics HD 5300 HD 5300 HD 5300
IGP Base Frequency / MHz 100 100 100
IGP Turbo Frequency / MHz 850 800 800
L3 Cache 4 MB 4 MB 4 MB
TDP 4.5 W 4.5 W 4.5 W
LPDDR3/DDR3L Support 1600 MHz 1600 MHz 1600 MHz
Intel vPro Yes No No
Intel TXT Yes No No
Intel VT-d/VT-x Yes Yes Yes
Intel AES-NI Yes Yes Yes

The top of the line processor will be called the Core M-5Y70, which is a bit of a mouthful but the name breaks down similarly to Intel’s main Core series. ‘5’ is similar to i5, giving us a dual-core processor with Hyper-Threading; ‘Y’ is for Broadwell-Y; and ‘70’ gives its position in the hardware stack.

The CPU will leverage both processor graphics and CPU Turbo Boost, allowing each of them to turbo at different times and different rates depending on the workload and overall power usage. Of particulary interest is that the 5Y70 features a base clock of 1.1 GHz, with turbo for both single-core and dual-core use listed as up to 2.6 GHz. The new HD 5300 GPU similarly has a 100 MHz base frequency with an 850 MHz turbo. The 5Y70 is different from the other two models in both clock speeds and features, as it will be part of Intel’s vPro program and also supports Intel TXT.

The other two processors in the stack are the 5Y10a and 5Y10, with dual-core + HT configurations and 800 MHz base frequency with turbo up to 2.0 GHz. There doesn't appear to be any major difference between the two parts, though Intel's presentation notes that the 5Y10 supports "4W Config Down TDP" (cTDP Down). The graphics is clocked slightly lower on the turbo, giving 800 MHz.

It's interesting to note that Intel informed us that the 1k unit pricing will be the same for all three processors: $281. Obviously these chips are going to end up in hybrids, tablets, and laptops that come pre-built, so the actual pricing will vary by OEM and whatever deals they have with Intel. But in general, Intel seems to be saying that OEMs can choose any of the three chips based on their power/thermal targets.

The HD Graphics 5300 is the new processor graphics and as part of the brief behind Core-M, a die shot was supplied with the important areas marked:

In the processor graphics section in the shot above, there clearly looks like 12 repeated units, with each representing two EUs (Execution Units). In our dive into the architecture in early August, it was stated that the minimum configuration here would be as a result of Broadwell taking 8 EUs per sub-slice, with the minimum configuration being three sub-slices, making 24 in total. This comes in combination with an increase in the L1 cache and samplers relative to the number of EUs, allowing for 25% more sampling throughput per EU.

Intel's Tick-Tock Cadence
Microarchitecture Process Node Tick or Tock Release Year
Conroe/Merom 65nm Tock 2006
Penryn 45nm Tick 2007
Nehalem 45nm Tock 2008
Westmere 32nm Tick 2010
Sandy Bridge 32nm Tock 2011
Ivy Bridge 22nm Tick 2012
Haswell 22nm Tock 2013
Broadwell 14nm Tick 2014
Skylake 14nm Tock 2015

The fundamental architecture of the GPU does not change from Haswell, albeit on a smaller process node. The GPU is confirmed as supporting DirectX 11.2, OpenGL 4.2 and OpenCL 2.0, with UltraHD (3840x2160) supported at 24 Hz through HDMI. This opens up possibilities of fanless tablets with UHD panels.

One of the main graphs Intel was pushing in their briefing was this one, indicating what power is required for a fanless tablet:

For a chassis of 7, 8 or 10mm, to have a maximum skin temperature of 41C at load, the above TDPs are required depending on the chassis size in order to go fanless. The first batch of 4.5W Core M processors aim at either the 11.6-inch, 8mm thick fanless tablet design as indicated in the graph above, or similarly a 10.1-inch 10mm thick tablet will also be suitable. Intel wants Core M to have a range of possible TDPs based on increasing or decreasing the frequency as required for a thin fanless tablet.

Intel is going to support extended docking functionality, especially with its business partners to allow features such as WiGig and additional I/O. Intel is also bringing a new 802.11ac design in the form of AC 7265, a lower powered version of the 2T2R 7260 for tablets. This will also support WiDi 5.0, and overall the platform aims to offer 1.7 hours longer battery life. Intel got to this ‘+1.7’ hour number with a reference design compared to a clocked-down Haswell-Y. I would like to point out that despite these numbers, a clocked-down part usually represents moving outside the optimal efficiency window, especially when dealing with low powered tablets.

Intel used the above slide in their presentations and drew particular attention to the power consumption of the audio during HD video playback (the orange bar on the top comparison). As part of Core M, Intel is reducing power consumption of the audio segment of the system from 100s of milliwatts down to single-digit milliwatts by integrating an audio digital signaling processor (DSP) onto the die. This is what Intel refers to as its Smart Sound Technology, and is designed to shift the majority of the audio processing onto a configured part of the die which can process at lower power.

If you think you’ve heard of something like this before, you have: AMD’s TrueAudio sounds remarkably similar in its implementation and its promotion. We asked Intel if this new DSP for Broadwell would have a configurable API similar to TrueAudio, however we are still waiting on an official response to this.

The platform controller hub layout was also provided, showing USB 3.0 support along with SATA 6 Gbps and four lanes of PCIe:

The PCH is also designed to be dynamic with power, meaning that disabling features on a design could yield a better-than-expected increase in battery life. The design will support NFC, and it is worth noting that the two USB 3.0 ports are in a mux configuration which may limit bandwidth. With a number of PCIe lanes in tow however, there are a number of controllers that could be used to expand functionality in a design.

Intel will be showing off the Core M at IFA in Berlin this week, with over 20 designs based on Core M from OEMs in the known pipeline – including designs like ASUS’ Transformer Book T300 Chi announced back at Computex. The T300 Chi was specified as a 12.5-inch fanless tablet in a 7.3mm thickness design, with LTE support and a 2560x1600 display. With the 12.5-inch size and 7.3mm width, it sounds like the T300 Chi will be modifying the Core M CPU to be around 4W in order to keep the 41C skin temperature as a maximum. Intel also listed the following Core M devices at IFA:

The CPUs will be in volume production before the end of the year (we seem to have differing reports whether volume production has started already or is just about to), with systems from ~5 OEMs available in Q4, starting in late October. Intel lists both consumer and business designs for this timeframe, however volume production is expected in Q1 2015.

POST A COMMENT

115 Comments

View All Comments

  • Morawka - Friday, September 5, 2014 - link

    can you guys not afford a dslr for the front page photo? This looks like it's been taken with a flip phone. Reply
  • fokka - Friday, September 5, 2014 - link

    haha, what i was thinking. like it has been run through a bad hdr filter. Reply
  • haukionkannel - Friday, September 5, 2014 - link

    It would be interesting to see these tested against Intel Atoms. How much more there is speed in these and how much more these eats electricity compared to atoms... These would be very interesting in for example Surface pro mini tablets! Reply
  • azazel1024 - Friday, September 5, 2014 - link

    I'd guess the power delta is pretty minor, at least 22nm Bay Trail versus 14nm Broadwell M. Now, against the upcoming 14nm Airmont/Braswell, it would probably be a pretty significantly delta.

    IIRC Bay Trail-T is a 4.5w TDP as well (or roughly that), though I suspect in normal use is likely to bat a fair amount lower than Broadwell M.

    Performance wise, see my comment above. The proof is in the puddling, but some basic extrapolation I'd guess that both are going to edge closer to Atom than they are to increase the current performance gulf. However, Airmont especially in Cherry Trail configurations, is likely to reduce the performance delta rather significantly, especially in GPU ability, but probably in CPU as well (its a 200Mhz increase, and probably/possibly IPC improvements...and they might be more than very minor IPC improvements, though I am betting on minor). Where as Broadwell M is a step back in CPU frequency at least (compared to ULT Haswell, which is the only baseline I have for Atom versus Core performance). Though compared to Haswell Y, it is probably a significant increase in performance.
    Reply
  • mkozakewich - Saturday, September 6, 2014 - link

    Considering the CPU in the charts above was running at 1 W and was only a fifth of the total platform power, things are definitely getting to a point where we have to shift our focus to the high-power aspects. Like, at full power, how much performance per Watt do we get with Broadwell? How about with Atom. Yeah, I'm definitely waiting for that. Reply
  • Krysto - Saturday, September 6, 2014 - link

    Who cares about "full power", whatever that's supposed to mean. What matters is how it does in its TDP range. I don't care if Broadwell can be 10x faster at 200W, if 99 percent of the Broadwell chips they want to sell are limited to 4.5W TDP. THAT'S where you wanto compare them. The rest is irrelevant. Reply
  • Krysto - Saturday, September 6, 2014 - link

    Exactly. That's what I want to see too. Let's compare this $280 chip to an Atom and also the $25 Nvidia Denver CPU and Apple A8. Then we'll see how truly pathetic these chips are.

    I bet they aren't more than 50 percent faster.
    Reply
  • Klimax - Saturday, September 6, 2014 - link

    Another jump to conclusions. My prediction is: Even massively constrained Broadwell will make short work of any of those chips. Reply
  • darkich - Saturday, September 6, 2014 - link

    Lol, no.
    Just no. Mark my words it won't be even 50% faster.
    And just wait the hilarity of the 8 core A57+A53 Mali T760Mp 12 20nm/16nm chips 6 months from now, almost matching the Core M..in a phone.
    Reply
  • Speedfriend - Monday, September 8, 2014 - link

    An A57+A53 8 core will be up to 1.5x faster than an octa A15 chip, which was probably 10-20% slower than a Baytrail. So you are assuming that Core-M will only be 30% odd faster than a Baytrail? And lets not forge tthat the graphics are likely to be significantly better in Core M. That seems ridiculous, but I suppose we will have to wait and see. The only performance guidelines for core M that I have seen is that it will be 2x a fast as a 2010 Nehalem at 1/4 the TDP. Reply

Log in

Don't have an account? Sign up now