Final HSA 1.0 specification now available

Since 2012, a group of industry heavyweights including AMD, ARM, Qualcomm, Samsung, and Imagination Technologies has been working on on a development platform that lets software more effectively use the diverse range of computational resources available in modern processors. Dubbed HSA, or Heterogeneous Systems Architecture, this open standard is designed to harness CPU, GPU, and even DSP resources. It also defines a virtual address space that allows CPUs and GPUs to access each others' memory.

Version 1.0 of the specification is now complete, and interested parties can grab it from the HSA Foundation's website. This initial release includes three documents: the system architecture specification describes how HSA-compliant hardware operates, the runtime spec details how applications interact with it, and the programming reference manual provides additional information for "software ecosystem, tool and compiler developers."

So, where's the compatible hardware? AMD revealed last month that its Carrizo APU is the "first processor in the world with HSA 1.0 support." We haven't heard about any other chips that are fully compliant with the final spec, but given HSA's backers, it seems likely that future mobile SoCs will get in on the action.

Comments closed
    • DarkMikaru
    • 5 years ago

    Great idea. To be able to better leverage GPU compute power for daily tasks on our PC’s. Imagine our browsers running directly from the GPU’s while the CPU’s do heavy lifting in the background. Very cool. The only thing that I don’t really understand is why specific hardware would be needed for such a thing.

    Isn’t HSA similar to OpenGL, DirectX and the like? Though I can understand the fundamental design of the CPU may need to be different for more efficient communication between processors. Ehhh…. I don’t quite get it. But happy to see the progress.

      • ronch
      • 5 years ago

      It’s not as simple as people think. You can look at the GPU as one massively parallel FPU a-la 3DNow! It’s hard enough to parallelize code for a few cores, let alone even just 320 compute units in a GPU. And remember, those are primarily number-crunching ALUs, not really designed to do heavy logic or move operations. That’s the CPU’s forte.

        • DarkMikaru
        • 5 years ago

        Thanks for that Ronch. Looks like I need to brush up on a few things.

    • UnfriendlyFire
    • 5 years ago

    By the time HSA really takes off, Intel and Nividia might have the resources to implement the hardware more efficiently than what AMD could do.

    Well, Nividia might take a little longer to support HSA.

    (And that assumes HSA doesn’t go the way of 3DNow!)

    • Tirk
    • 5 years ago

    Might also provide a compatible environment for Mediatek to better integrate the GCN GPU they are licencing from AMD.

    Its good an article like this mentions that its not just AMD working HSA and that HSA looks to take a much bigger role in the future as long as the other companies are serious about it as well. AMD stands to gain a lot if the ARM environment shifts towards HSA.

    • esterhasz
    • 5 years ago

    What I don’t get so far is the difference between compiling something written in C++ AMP to OpenCL or to HSAIL (the HSA intermediary language). Can it be that in the former case, the driver (edit: or rather the OpenCL compiler?) has to deal with HSA juggling (and may fall back on memory copy if HSA is not supported) and in the second shared memory is assumed? Or am I completely misunderstanding this?

    Since I use a lot of stuff like monte carlo, I really want to try getting into the whole heterogeneous programming thing (tried out the Bolt library a bit), but it’s ultimately not that easy…

    • MadManOriginal
    • 5 years ago

    Gee, that only took 9+ years.

      • ronch
      • 5 years ago

      +9 to you.

      • Jigar
      • 5 years ago

      2012 to 2015 is 9+ years ??? Gee, how are you holding up in this world with such maths skills ?

        • jaset
        • 5 years ago

        A wild guess is that @MadManOriginal is referring to CTM and that was originally announced/released in 2006. (CTM -> OpenCL -> HSA)

        Edit: grammar correction

        • Srsly_Bro
        • 5 years ago

        His name does say Mad Man in it…

        • MadManOriginal
        • 5 years ago

        I was referring to AMD’s acquisition of ATI. I know this is an industry group, but that’s when the whole heterogeneous compute idea became a thing that any had heard of.

      • anotherengineer
      • 5 years ago

      And the x64 extension has been out since 2003 and still waiting for 64-bit programs.

      And quad-core were out a decade ago and I still see single threaded programs.

      Tis the life in the PC software world.

    • ronch
    • 5 years ago

    I guess we can now all dip our toes in the HSA pond now that the FIRST HSA 1.0-compliant APU is coming out soon.

      • smilingcrow
      • 5 years ago

      Hopefully:

      “Kaveri APUs currently rely on the HSA 1.0 Provisional specification, and Carrizo (based on Excavator) will aim for 1.0 Final compliance if the tools are ready before Carrizo launch. Carrizo will not be held back in order to secure compliance before ramping up production, but the expectation is that it should pass and be used similar to Kaveri but with the minor adjustments required for 1.0 Final, such as GPU context switching.”

Pin It on Pinterest

Share This