Announcement

Collapse
No announcement yet.

Intel Linux NPU Driver v1.5 Released - Now Validated On Arrow Lake & Lunar Lake

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

  • Intel Linux NPU Driver v1.5 Released - Now Validated On Arrow Lake & Lunar Lake

    Phoronix: Intel Linux NPU Driver v1.5 Released - Now Validated On Arrow Lake & Lunar Lake

    Intel on Wednesday released version 1.5 of their Linux NPU driver, their user-space driver component for Linux systems in enabling the neural processing unit found with Core Ultra "Meteor Lake" processors and newer. This goes along with their upstream IVPU kernel accelerator driver for allowing a full open-source solution for AI workloads with the likes of OpenVINO...

    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
    “Meanwhile AMD still hasn't yet attempted to upstream their XDNA Linux kernel driver first posted earlier this year to GitHub nor commented too much on their mainline timing/plans for getting this support upstream and adopted by more software. Intel is winning the AI race when it comes to the timely open-source and upstream driver support for their NPUs and other hardware.”

    And people still give me shit when I tell them that AMD needs to stop ROCm because Intel’s oneAPI is the only viable competitor to CUDA. And that just about every initiative that AMD started since and including 3DNow! has failed to achieve market acceptance and has been abandoned. And with what Michael said above I couldn’t agree more. The only real benefit of AMD in the past and particularly post Lisa Su has been the manufacturing of cheaper and in most cares more performant hardware. But their software frameworks and toolkits suck and have no widespread market support. Lisa Su basically alluded to that fact earlier this year concerning drivers and ROCm.

    Comment


    • #3
      I'm fairly interested to see how Lunar Lake actually pans out in testing. NPU performance is part of that, but I'm more interested in the efficiency and iGPU performance. They were talking about ~50% raster performance gain, when Meteor Lake was already slightly faster than the 780M in many cases. If these actually end up being very efficient and there's a big iGPU perf gain, it could be an interesting choice for a handheld (yes drivers will still be an issue in some games). MSI had the Claw and a revision of that which I assume were / will be flops. I'm curious if they'll try again with Lunar Lake.

      Comment


      • #4
        Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post
        I'm fairly interested to see how Lunar Lake actually pans out in testing. NPU performance is part of that, but I'm more interested in the efficiency and iGPU performance. They were talking about ~50% raster performance gain, when Meteor Lake was already slightly faster than the 780M in many cases. If these actually end up being very efficient and there's a big iGPU perf gain, it could be an interesting choice for a handheld (yes drivers will still be an issue in some games). MSI had the Claw and a revision of that which I assume were / will be flops. I'm curious if they'll try again with Lunar Lake.

        The Meteor Lake GPU is a little faster than the GPU of AMD Hawk Point or Phoenix.

        This has been predicted before the product launch, because the Intel GPU has 4/3 times more ALUs and a clock frequency that is a little greater than 3/4 of the clock frequency of the AMD GPUs.

        The new AMD Strix Point GPU increases the number of ALUs by 4/3 times, matching the 1024 FP32 execution units of the Intel GPU, while keeping about the same clock frequency as Hawk Point and Phoenix, so an increase in performance of about 33% is expected. The leaked early reports of GPU benchmarks for Strix Point confirm the expected performance increase.

        Intel claims an even greater performance increase of 50% for the Lunar Lake Xe2 GPU, but it is not clear yet in what conditions can this be achieved.

        Lunar Lake will be manufactured using a TSMC CMOS process that is not only better than the process used for Meteor Lake, but it is also better than the process used for the new AMD Strix Point.

        This means that at equal power consumption, the GPU of Lunar Lake will have a higher clock frequency than that of the GPU of Strix Point (the two GPUs have about the same size).

        This higher clock frequency will give a performance advantage of Lunar Lake over Meteor Lake, but it is unlikely that this can be as high as 50%.

        The Xe2 GPU has additional execution units, e.g. for matrix operations, which must explain the rest of the 50%, but then such performance increases will exist only for certain programs, not for any program.

        This has to be clarified by more details about the Xe2 GPU and by benchmark results. For now, it is not clear how the comparison between the GPUs of Lunar Lake and Strix Point will look. It is likely that the GPU of Strix Point will be faster, but only at higher power consumptions. At reduced power consumptions, like in fanless devices or thin devices, it is likely that the Lunar Lake GPU will be faster.










        Comment

        Working...
        X