Announcement

Collapse
No announcement yet.

Intel SVM Support Published For Linux - Another Step On The March To Xe GPUs

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

  • Intel SVM Support Published For Linux - Another Step On The March To Xe GPUs

    Phoronix: Intel SVM Support Published For Linux - Another Step On The March To Xe GPUs

    Ending out the week is an exciting development in the Intel open-source graphics driver space... Shared Virtual Memory (SVM) support! This is another step towards their upcoming discrete Xe GPUs and ultimately their exciting oneAPI conquest...

    http://www.phoronix.com/scan.php?pag...port-Published

  • #2
    Their software stack seems to come along nicely. And I am very interested what Intel brings to the table next year concerning their new Xe GPUs. The more competition the better.

    Comment


    • #3
      Originally posted by ms178 View Post
      Their software stack seems to come along nicely. And I am very interested what Intel brings to the table next year concerning their new Xe GPUs. The more competition the better.
      I'm glad I don't work on the Intel Xe team. If they don't turn out something price/performance competitive with AMD/Nvidia nearest equivalent in current generation they're going to get crucified. Again. That's a very high bar to meet. It won't matter if their software stack is in good shape if the hardware won't perform or they repeat the mistakes of their CPU counterparts.

      Comment


      • #4
        Originally posted by stormcrow View Post
        I'm glad I don't work on the Intel Xe team. If they don't turn out something price/performance competitive with AMD/Nvidia nearest equivalent in current generation they're going to get crucified. Again. That's a very high bar to meet. It won't matter if their software stack is in good shape if the hardware won't perform or they repeat the mistakes of their CPU counterparts.
        Intel seems to grasp the strategic importance of this effort. They created a whole new corporate division around it. Raja's talk is really worth reading through, to understand the scope and depth of the conceptual journey they're on.

        https://www.anandtech.com/show/15123...pm-mt-11pm-utc

        As such, I don't expect this to be a one-shot effort. Xeon Phi had 2 generations of commercially-released processors (and at least 2 unreleased predacessors + 2 cancelled successors waiting to go out the door) before that program got cancelled. Unless it's a worst-case, unmitigated disaster (which I really don't foresee), I fully expect Intel will have some strategic patience around this.

        That said, what they're trying seems a bit like a shotgun approach. They're attacking it from every angle: GPUs, FPGAs, deep learning ASICs, and possibly some kind of programmable matrix processors. Maybe they won't gain traction on one of those fronts, and we'll see a couple more product line cancellations. But, I really don't see this as flailing, shots in the dark, or a narrow initiative on a short leash - this is more like Intel re-inventing itself, as it fully comes to terms with the new realities of compute. It could even be another nail in the coffin of x86, if the result is Intel finally seeing its fortunes less intertwined with the fate of that scraggly old ISA.

        Comment


        • #5
          Originally posted by stormcrow View Post

          I'm glad I don't work on the Intel Xe team. If they don't turn out something price/performance competitive with AMD/Nvidia nearest equivalent in current generation they're going to get crucified. Again. That's a very high bar to meet. It won't matter if their software stack is in good shape if the hardware won't perform or they repeat the mistakes of their CPU counterparts.
          I agree, the hardware side is more important but nevertheless, you need both. It also seems that Intel needs a fair amount of pressure to get meaningful innovation out the door (think of an Athlon moment). My enthusiasm is certainly fueled by my expectations to get great compute/graphics capabilities on Windows and Linux at a fair price with a mature, well-tested and long-maintained software stack. But even though the pressure is high on them to deliver, some circumstances are in their favor now to succeed: They do have the talent and manpower, they now enjoy more attention of the management and I guess financial ressources than before and I hope they made the right technical decisions this time around. Xeon Phi always made the impression that they wanted to throw their x86-ISA at the problem, no matter how flawed that concept might have been from the start. I hope they learned that lesson and are more problem-focused this time around.

          Comment


          • #6
            Originally posted by coder View Post
            They're attacking it from every angle: GPUs, FPGAs, deep learning ASICs, and possibly some kind of programmable matrix processors. Maybe they won't gain traction on one of those fronts, and we'll see a couple more product line cancellations. But, I really don't see this as flailing, shots in the dark, or a narrow initiative on a short leash - this is more like Intel re-inventing itself, as it fully comes to terms with the new realities of compute. It could even be another nail in the coffin of x86, if the result is Intel finally seeing its fortunes less intertwined with the fate of that scraggly old ISA.
            It certainly is more risky than what they shipped to the market during the last decade. It also makes me wonder what they spend their R&D money on during that same time period. One might think they would have secretly developed something disruptive in the background, like a x86-ISA successor. The time certainly was there for them to come up with something like this! But appearently they underinvested into their core products and became complacent. On the bright side, competitiors took notice and see their chances to gain some traction ...
            Last edited by ms178; 11-23-2019, 05:28 PM.

            Comment


            • #7
              Originally posted by ms178 View Post
              It certainly is more risky than what they shipped to the market during the last decade. It also makes me wonder what they spend their R&D money on during that same time period.
              A failed push into mobile, aborted IoT effort, Xeon Phi (cancelled), OmniPath (mothballed)... and dividends to stockholders. They've also made a lot of big acquisitions, in the past 4 years.

              Originally posted by ms178 View Post
              One might think they would have secretly developed something disruptive in the background, like a x86-ISA successor.
              They do build research chips, but I think they can't undertake anything too massive without some level of reporting in their SEC disclosures.

              Originally posted by ms178 View Post
              But appearently they underinvested into their core products and became complacent.
              You mean their 14 nm and now 10 nm manufacturing delays? Yeah, they used to have an unassailable lead, and now they're lagging behind. Definite under-investment & complacency, there.

              Originally posted by ms178 View Post
              On the bright side, competitiors took notice and see their chances to gain some traction ...
              Competition was always going to be there - it's just that they're getting more oxygen as Intel has fallen behind.

              Comment

              Working...
              X