Announcement

Collapse
No announcement yet.

Intel Buys Out AI Startup Habana Labs

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

  • Intel Buys Out AI Startup Habana Labs

    Phoronix: Intel Buys Out AI Startup Habana Labs

    Well, here is some interesting M+A activity a week ahead of Christmas... Intel just announced they are acquiring AI chipmaker start-up Habana Labs...

    http://www.phoronix.com/scan.php?pag...ys-Habana-Labs

  • #2
    Do they admit once again that their former acquisitions in this area (Nervana) were not enough or the wrong technology to invest in? Or do they want to deny their competitors access to this tech?

    Comment


    • #3
      RIP Habana Labs.

      Comment


      • #4
        Originally posted by ms178 View Post
        Do they admit once again that their former acquisitions in this area (Nervana) were not enough or the wrong technology to invest in? Or do they want to deny their competitors access to this tech?
        Why not both?

        Comment


        • #5
          Originally posted by ms178 View Post
          Do they admit once again that their former acquisitions in this area (Nervana) were not enough or the wrong technology to invest in? Or do they want to deny their competitors access to this tech?
          I don't know the specifics here but AI is a very generic term, just like "soup" or "car".

          The underlying hardware can differ quite a bit depending on what kind of AI algorithms you are running. We are talking of more AI-specific hardware than a GPU is.

          Comment


          • #6
            Originally posted by starshipeleven View Post
            I don't know the specifics here but AI is a very generic term, just like "soup" or "car".

            The underlying hardware can differ quite a bit depending on what kind of AI algorithms you are running. We are talking of more AI-specific hardware than a GPU is.
            In the current market, it's mostly just a PR and marketing buzzword to sell clueless investors on the current big thing in the tech sector. We've been down this road before back in the late 70s and early 80s. Back then it was to get people to invest in LISP machines and other hyped hardware and software services, much like it is today.

            Less marketing fluff version: Intel bought out a patent portfolio to help develop and deploy certain potentially useful machine learning algorithms for specific domains. Specifically, Habana Labs created a processor platform designed to use the Glow compiler for image recognition, iirc. But just putting that in a press release is boring.

            It's also potentially a bad PR move considering how much skepticism image recognition in general is getting in the mainstream press.
            Last edited by stormcrow; 12-16-2019, 04:21 PM.

            Comment


            • #7
              Originally posted by stormcrow View Post
              In the current market, it's mostly just a PR and marketing buzzword to sell clueless investors on the current big thing in the tech sector.
              In this case it is actual AI, it's not just a buzzword attached on your run-of-the-mill webapp with a cloud server and some basic non-AI data analysis algorithms to give some nice graphs.

              Less marketing fluff version: Intel bought out a patent portfolio to help develop and deploy certain potentially useful machine learning algorithms for specific domains. Specifically, Habana Labs created a processor platform designed to use the Glow compiler for image recognition, iirc.
              That's what I said. This AI technology they just bought is specialized to do something, and it's probably not the same as the other AI technology they bought before.

              I don't see why you are so angry and salty.
              It's also potentially a bad PR move considering how much skepticism image recognition in general is getting in the mainstream press.
              If you think image recognition is at risk, I have bad news for you.

              Anyone trying to limit it will get lobbied so hard by more or less everyone in robotics, automotive, security, military and surveillance, plus all the usual suspects that live by data collection.

              You can at best expect some slaps on the wrist or "do good" laws that amount to nothing (similar to cookies laws in EU), like "if you don't want to be image recognized you can't use our services". Thank you EU, that's exactly what I needed, an additional popup bullshit about cookies and privacy bs that I can't turn off, god fucking damnit.

              Comment


              • #8
                If you can't beat them, drown the sellouts in $$$.
                Atleast someone is laughing all the way to the bank.

                Comment


                • #9
                  Originally posted by Redfoxmoon View Post
                  RIP Habana Labs.
                  Oh, be realistic. These kinds of startups are created to be acquisition targets. This is exactly what they wanted.

                  Comment


                  • #10
                    Originally posted by stormcrow View Post
                    Specifically, Habana Labs created a processor platform designed to use the Glow compiler for image recognition, iirc. But just putting that in a press release is boring.
                    Maybe it had something to do with this:
                    A standout feature of Gaudi is its interconnect for scaling out to hundreds of chips. It is the first AI chip with engines for Remote Direct Memory Access (RDMA) over Converged Ethernet with a bi-directional throughput of 2 Tbps (via up to 10 100 Gbps Ethernet ports). RDMA means that a chip can access any other chip’s memory directly (without involving the operating system). The advantage of Ethernet is that switches are readily available, and there is no need for PCIe switches or dedicated NICs.

                    In terms of performance, one 140W TDP chip has a throughput of 1,650 images per second in ResNet-50, 14% higher than Nvidia’s Tesla V100 at half the power consumption. This results in an efficiency of over twice the number of images per second per watt. The difference becomes even larger as the number of chips scales. Based on MLPerf v0.5, 640 chips achieve a throughput of 845,400,000 images per second, whereas the V100-based system with 80 DGX-1s reached 218.3 thousand images per second. A 512-chip Gaudi system has a relative efficiency of 80% compared to under 30% for a V100 system.
                    https://www.tomshardware.com/news/in...habana-labs-ai

                    Comment

                    Working...
                    X