Announcement

Collapse
No announcement yet.

Ubuntu 15.04 Brings Some Graphics Performance Improvements For Intel Haswell

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

  • Ubuntu 15.04 Brings Some Graphics Performance Improvements For Intel Haswell

    Phoronix: Ubuntu 15.04 Brings Some Graphics Performance Improvements For Intel Haswell

    While we know Intel Broadwell performance is much faster on Ubuntu 15.04 than Ubuntu 14.10, with this week's release of Ubuntu Vivd Vervet, here's some fresh results looking to see how the Intel Haswell graphics performance has evolved over the past six months. For the many Intel Haswell owners out there, you'll be pleased that the performance has overall improved.

    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
    In the meanwhile, I'm still waiting for working Bay Trayl-T (Z3700 series) graphics. Intel wants to sell their Compute Sticks with Linux, at the very moment I don't see this happening.

    Comment


    • #3
      Originally posted by J?rnS View Post
      In the meanwhile, I'm still waiting for working Bay Trayl-T (Z3700 series) graphics. Intel wants to sell their Compute Sticks with Linux, at the very moment I don't see this happening.
      Support for Bay Trail has been there since forever. I have a Bay Trail-M netbook (Celeron N2806), works great. Then there's a thread at the Arch forums about the Asus X205TA which uses Bay Trail-T (Atom Z3735F), and the thread is about everything but graphics. So if you're having issues, it's specific to your situation, not a general lack of support for Bay Trail.

      Comment


      • #4
        Well not every device is affected, but quite a lot. For example my Dell Venue 8 Pro 3845. Screen is blank when KMS is used, without KMS no graphics acceleration.

        For reference: https://www.happyassassin.net/fedlet...trail-tablets/

        Comment


        • #5
          Originally posted by J?rnS View Post
          Well not every device is affected, but quite a lot. For example my Dell Venue 8 Pro 3845. Screen is blank when KMS is used, without KMS no graphics acceleration.
          If it's knows that several devices don't work, that of course is a problem. What is Intel saying about it? By that I mean can you provide any relevant links to bugzilla or mailing list discussions or similar?

          Comment


          • #6
            Here is a bug report for the kernel: https://bugs.freedesktop.org/show_bug.cgi?id=71977

            Haven't seen anything official from Intel up to now. Pretty sad, I would love to have Debian on that tablet. I could even live with an external Wifi-dongle, but accelerated graphics are a must-have. With nomodeset it "works", but that is not a real solution.

            Comment


            • #7
              GpuTest Triangle

              I think GpuTest Triangle improvement comes from switching X from DRI3 back to DRI2 (DRI3 was causing that test to be Vsynched).

              Comment


              • #8
                Uh, by my count there were more regressions than improvements, with only the synthetic gputest improving significantly. This is backwards, Intel.

                Comment


                • #9
                  Performance differences

                  I think article refers also to earlier results with CS:GO and TF2, which both improved, not just to results in this article.

                  GpuTest Pixmap Piano & Volplosion increases are due to improvements in Mesa shader compiler (those tests are fragment shader bound), Plot3D improvement is likely to be either DRI3->DRI2 change or kernel improvement.

                  As to the regressions in more complicated cases, I don't know. This is testing different Ubuntu versions, it's not a 3D stack version comparison.

                  Just by switching from Unity to gnome-shell I've seen >10 percent performance changes to both directions in some tests on Ubuntu 14.10 (not on HSW, but a machine with much weaker CPU), so e.g. changing from one Unity version to another may also affect performance, the changes aren't necessarily related to HW specific graphics stack changes. To do comparison for those, testing should change only those specific components, not whole distro.

                  Except for Tessarect change, the "regression" are small enough (few percent), that they could happen just by changes unrelated to graphics stack, e.g. scheduling, power management policy, memory allocation changes or configuration differences between distro versions.

                  Also, as test results don't include error bars, it's hard to know how accurate the results are.

                  Comment


                  • #10
                    Originally posted by J?rnS View Post
                    Here is a bug report for the kernel: https://bugs.freedesktop.org/show_bug.cgi?id=71977

                    Haven't seen anything official from Intel up to now.
                    There's some Intel names in this bug. But it's marked fixed, as is another bug linked from this one. However, there's this: https://bugs.freedesktop.org/show_bug.cgi?id=82880. Comment 7 describes a hacky workaround workaround, does it work?

                    Then there's this: https://bugs.freedesktop.org/show_bug.cgi?id=89573. So yeah, there's issues in Baytrail's DSI support (no wonder I didn't see it - DSI is a tablet thing, laptops use eDP).

                    Welcome to the wonderful world of modern embedded hardware. Linux is still a bit behind in this area, not just when it comes to video output. The fact that each device is wired slightly differently certainly doesn't help.

                    Comment

                    Working...
                    X