Announcement

Collapse
No announcement yet.

Intel VPU To Be Found On All Meteor Lake SKUs, Intel Seeding Open-Source Projects

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel VPU To Be Found On All Meteor Lake SKUs, Intel Seeding Open-Source Projects

    Phoronix: Intel VPU To Be Found On All Meteor Lake SKUs, Intel Seeding Open-Source Projects

    Intel this week is using Computex 2023 to make some disclosures around next-generation Meteor Lake processors for laptops. The most exciting aspect relayed in advance during our press briefing last week was that all Intel Meteor Lake processor SKUs will feature their new Vision/Versatile Processing Unit...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    It's very interesting and refreshing to see Intel mentioning this new VPU block on "all SKUs" for Meteor Lake. It's great that they aren't artificially segmenting it to higher-end processors or particular SKUs.
    Since they have GPUs now, I think it's more likely that they only have so much fab time to make their CPUs so they aren't able to make a fragmented CPU launch like they used to do. From that perspective, it makes more sense to follow AMD's lead where your CPUs support as much as possible so the end-user is less confused since they only have to choose based on the amount of cores/performance level of the cores and if it is unlocked for overclocking or not. Since we're talking about an AIO SKU, I'm including GPU, CPU, AI, FPGA, etc when I say "cores/performance" generically like that.

    In that same regard, it's not the same market. It isn't 2004 or even 2016. Now they have to compete with not only AMD and x86, but multiple other companies using multiple architectures and processing technologies. They can't intentionally fragment the market to compete with themselves and AMD. Nobody can make multiple niche CPU lines these days. AMD all-in-oned their x86 line, Apple/ARM are doing it, and, by the looks of it, Intel saw the light and is doing the same. IMHO, that kind of fragmentation doesn't help one bit when both the lowest end and highest end have enough raw processing power to please a casual. All that does is piss off the casuals who bought feature limited hardware because they...pulling numbers...didn't realize there were 7 sub-versions of the 12th generation.

    Comment


    • #3
      AMD has any APU/VPU driver in the works? Is there any API like Vulkan that abstracts how programmers can interact with those APUs?

      Comment


      • #4
        Originally posted by skeevy420 View Post

        Since they have GPUs now, I think it's more likely that they only have so much fab time to make their CPUs...
        TSMC fabricates Intel's GPUs.

        Comment


        • #5
          Originally posted by tessiof View Post
          AMD has any APU/VPU driver in the works? Is there any API like Vulkan that abstracts how programmers can interact with those APUs?
          At the moment, they use Xilinx APIs. But AFAIK there is nothing unified like Metal.
          Last edited by brucethemoose; 30 May 2023, 10:42 AM.

          Comment


          • #6
            1. AMD needs to put XDNA in its upcoming Granite Ridge desktop CPUs, not just Phoenix and Strix Point APUs.
            2. I really want to use AI-powered The GIMP now.

            Comment


            • #7
              Originally posted by jaxa View Post
              I really want to use AI-powered The GIMP now.
              I bet he will be thrilled:

              Gimp2.jpg

              Comment


              • #8
                Originally posted by jaxa View Post
                1. AMD needs to put XDNA in its upcoming Granite Ridge desktop CPUs, not just Phoenix and Strix Point APUs.
                2. I really want to use AI-powered The GIMP now.
                I actually agree with this.


                ​​​​​​Two years ago this was a bad idea, but all of the sudden there are applications that need way more ram than desktop GPUs (llms, tts, diffusion), but need a bit more matrix compute than a CPU alone can provide.

                Zero copy support in whatever API they use is essential though. Copying weights from one memory pool to another in RAM will be counterproductive.
                ​​

                As for GIMP... what are you thinking of specifically? TBH other open source programs do inpainting/generation/upscaling about as well as Photoshop does now.

                Comment

                Working...
                X