Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Wednesday August 24 2016, @05:21AM   Printer-friendly
from the remember-when-'core'-referred-to-memory? dept.

Microsoft has talked about a "holographic processing unit" powering its HoloLens augmented reality device. Now it has released details about the device's processors at the Hot Chips 2016 conference:

Microsoft today revealed a first look at the inside of its Holographic Processing Unit (HPU) chip used in its virtual reality HoloLens specs.

The secretive HPU is a custom-designed TSMC-fabricated 28nm coprocessor that has 24 Tensilica DSP cores. It has about 65 million logic gates, 8MB of SRAM, and a layer of 1GB of low-power DDR3 RAM on top, all in a 12mm-by-12mm BGA package. We understand it can perform a trillion calculations a second. It handles all the environment sensing and other input and output necessary for the virtual-reality goggles. It aggregates data from sensors and processes the wearer's gesture movements, all in hardware so it's faster than the equivalent code running on a general purpose CPU. Each DSP core is given a particular task to focus on.

The unit sits alongside a 14nm Intel Atom x86 Cherry Trail system-on-chip, which has its own 1GB of RAM and runs Windows 10 and apps that take advantage of the immersive noggin-fitted display.

Also at PCWorld.


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by butthurt on Wednesday August 24 2016, @05:56AM

    by butthurt (6141) on Wednesday August 24 2016, @05:56AM (#392483) Journal

    Yes, except unfortunately Linux isn't a real-time operating system.

    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 0) by Anonymous Coward on Wednesday August 24 2016, @05:59AM

    by Anonymous Coward on Wednesday August 24 2016, @05:59AM (#392485)

    If Linux can't do it, it doesn't need to be done, because Linux is the greatest. If it's not Linux, it's crap!

  • (Score: 3, Informative) by q.kontinuum on Wednesday August 24 2016, @06:53AM

    by q.kontinuum (532) on Wednesday August 24 2016, @06:53AM (#392500) Journal

    Linux can be anything, given enough developers developing the right patch. Even Realtime [osadl.org]. Although with your nickname I'd have expected some other other statement, like "Linux can't, but hurd could.". GNU Hurd, as a microkernel system, should support real-time applications, or at least should be easier to add.

    --
    Registered IRC nick on chat.soylentnews.org: qkontinuum
    • (Score: 0) by Anonymous Coward on Wednesday August 24 2016, @07:03AM

      by Anonymous Coward on Wednesday August 24 2016, @07:03AM (#392502)

      Pretty sure there are at least 5 projects bringing realtime support to linux. Given that all 'realtime' really means is 'verifiable and ideally low latency', it isn't that hard to make a real time OS from a theoretical point of view. Actually ensuring it stays realtime and ensuring all hardware and drivers maintain response consistency is another matter.

    • (Score: 0) by Anonymous Coward on Wednesday August 24 2016, @06:05PM

      by Anonymous Coward on Wednesday August 24 2016, @06:05PM (#392697)

      This software for vehicle simulation and room-sized racecar simulators runs on Redhat Linux:
          http://www.vi-grade.com/index.php?pagid=vehicle_dynamics_carrealtime [vi-grade.com]
      Basic integration rate is 1000 steps/sec and obviously they are doing things to run realtime.

      Several years ago we wrote an add-on delivered as a shared object (*.so). Our stuff compiled OK, and ran with our own test code, but wouldn't run on their system. After pulling our hair out, they finally gave us a full virtual machine of their operating environment with all the right options and versions of everything (including some older versions). Then it worked.

      Must be a nightmare to maintain, but they probably ignore security updates, these machines are off-line (for the most part).

      Before you ask, I don't know if they are having any problems with systemd.

    • (Score: 2) by edIII on Wednesday August 24 2016, @06:23PM

      by edIII (791) on Wednesday August 24 2016, @06:23PM (#392707)

      I don't think that is the real issue at all, meaning Linux vs Microsoft. It's Freedom vs Proprietary instead.

      The Hololens sounds kick-ass, and I want. Yet.... I know the code is Microsoft, which means, NSA. While that may not be entirely fair, Microsoft's only response is completely transparency under the extreme toxicity of the current relationship between tech and national security. The Patriot Act literally means that no executive can ever be trusted again, and by extension, any corporation ever again. Moreover, I think the world has shifted towards completely and utterly UNTRUSTED status, and we must take it in our own hands to establish chains of trust similar to verifying SHA-256 hashes on downloaded packages.

      I would purchase a Hololens, assuming I'm not slave class in a dying America, but if and only if there are no blobs, binaries, or DRM, anywhere, in any circuit. Once vetted by the community as being reasonably clean with how-tos on how to perform verification yourself, I would have no problems purchasing hardware and software from Microsoft.

      Since we can very safely assume that Microsoft will NEVER release a clean Hololens, it in fact, doesn't exist. That's where I'm going. A future in which the hardware ecosystem I have access to may be 10 years behind state-of-the-art, but it will be free. It will be FREE.

      --
      Technically, lunchtime is at any moment. It's just a wave function.
  • (Score: 2) by FatPhil on Wednesday August 24 2016, @03:15PM

    by FatPhil (863) <pc-soylentNO@SPAMasdf.fi> on Wednesday August 24 2016, @03:15PM (#392599) Homepage
    Yeah, but I thought MS no longer have a RT Windows. The last one I know was definitely RT was nearly a decade back, and even the ones that hint in wiffle-waffley marketting speak to be "useful for real-time applications" or similar date back 3 or so years.

    And didn't Intel announce that they were killing Cherry Trail? Maybe MS just got a job-lot on the cheap?
    --
    Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
  • (Score: 2) by LoRdTAW on Wednesday August 24 2016, @06:10PM

    by LoRdTAW (3755) on Wednesday August 24 2016, @06:10PM (#392703) Journal

    Jesus christ. Does anyone bother to do a little research before posting a fucking foot in their mouth? Or do you prefer to look like a lazy sack of turds/idiot?
    https://rt.wiki.kernel.org/index.php/Main_Page [kernel.org]
    https://xenomai.org/ [xenomai.org]
    https://www.rtai.org/ [rtai.org]

  • (Score: 2) by RamiK on Wednesday August 24 2016, @06:24PM

    by RamiK (1813) on Wednesday August 24 2016, @06:24PM (#392708)

    That's ok. It's a modern x86 so it doesn't do real time anyhow.

    Real Time: The ability to write a piece of code, compile it to assembly, count the instructions, and tell with 100% certainty how long a computation will take.

    Intel Real Time: You write a piece of code. It compiles to x86 bytecode you think is machine code. It goes to a silicone interpreter that converts the instructions to machine instructions without any guarantees on completion time except for "It's over 3GHz so it's fast enough".

    After 20 years of this, people think Linux & Windows can do real time since you can tinker the plumbing enough to put your code on one core, the kernel on the other and have enough bus to put locks everywhere so nothing gets out of sync. And people wonder why avionics and automobile are so crappy nowadays...

    --
    compiling...
    • (Score: 0) by Anonymous Coward on Wednesday August 24 2016, @09:43PM

      by Anonymous Coward on Wednesday August 24 2016, @09:43PM (#392777)

      Real-time can also means a system with an guaranteed maximum on interupts service time and a deterministic tick based scheduler

      • (Score: 2) by RamiK on Thursday September 01 2016, @07:10PM

        by RamiK (1813) on Thursday September 01 2016, @07:10PM (#396318)

        Not when low latency or direct input processing (no buffers) is concerned.

        And without that, there's no point to real time.

        --
        compiling...