Announcement

Collapse
No announcement yet.

Intel Xe Graphics' Incredible Performance Uplift From OpenCL To oneAPI Level Zero To Vulkan

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
    smitty3268
    Senior Member

  • smitty3268
    replied
    Originally posted by mikkl View Post
    It is negative in various ways because it means Intel has a much stronger GPU, per clock it is much faster which is a good thing in the longer term. Intel has a much bigger headroom for future clock speed improvements than AMD.
    What longer term? Vega is dead, and Intel will be competing against a new architecture in a few months.

    In the here and now, all that matters is what they currently do. If Intel was able to run their gpu any faster while still hitting their various power, cost, etc. goals then they would. Slower isn't better.

    Leave a comment:

  • mikkl
    Junior Member

  • mikkl
    replied
    Originally posted by smitty3268 View Post
    You say that like it's a negative, but it's a massive positive that it's able to be clocked higher.

    It is negative in various ways because it means Intel has a much stronger GPU, per clock it is much faster which is a good thing in the longer term. Intel has a much bigger headroom for future clock speed improvements than AMD. For Intel it would be way worse if it was the other way around. Keep in mind AMD didn't clock their iGPUs that high until the Renoir generation. So with ADL-P (Xe LP 96 EUs) they can work on clock speed improvements without adding more units. Also keep in mind iGPUs are bandwidth limited, you cannot expect Intel to be 80-90% faster on DDR4/LPDDR4, this is unrealistic. Generally the faster iGPU has bigger bandwith constraints. Next year Tigerlake with LPDDR5 support is coming by the way.

    Leave a comment:

  • smitty3268
    Senior Member

  • smitty3268
    replied
    Interesting, and a useful contribution to this discussion. I think the tests I saw had the Renoir less power constrained which to be fair is probably overestimating it. However, note that the Xe is only about 10-30% faster in those tests. That's a decent lead, but it's not crushing Renoir like you might think if all you've seen is the marketing information coming out of Intel. It's basically the same type of lead that Renoir held over Ice Lake before this.

    Originally posted by mikkl View Post
    Renoir is manufactured on a better GPU process
    The only thing that matters is the final performance. If Intel wanted to use TSMC, they could - they deliberately chose not to, and their final product should be judged for what it is, not what it theoretically could be in some alternate world.

    Vega 8 needs a 50% higher clock speed
    You say that like it's a negative, but it's a massive positive that it's able to be clocked higher.
    smitty3268
    Senior Member
    Last edited by smitty3268; 26 October 2020, 12:04 AM.

    Leave a comment:

  • mikkl
    Junior Member

  • mikkl
    replied
    Originally posted by smitty3268 View Post

    Vega is also last-gen architecture, but the only thing that matters is actual performance.

    Renoir/Vega in a 4800U consistently outperforms Tigerlake/Xe when running at 15W. The amount ranges from a very small amount, to a pretty large one depending on the benchmark you're looking at.

    Bump the Tigerlake CPU up to 28W and it takes the lead, but it's also using more power.

    Not in this test: https://www.pcworld.com/article/3575...e-preview.html

    And also this is meaningless in regards to Xe LP because it's a package power and not GPU power and furthermore Renoir is manufactured on a better GPU process. Xe LP is a lot stronger than Vega 8, Vega 8 needs a 50% higher clock speed and still loses.

    Leave a comment:

  • smitty3268
    Senior Member

  • smitty3268
    replied
    Originally posted by mikkl View Post


    Xe is a lot better than Vega, even the 80 EUs version of Xe LP at 1300 Mhz easily beats Vega 8 running at 1750 Mhz. Vega in Renoir runs much higher clocked and loses. Another factor is that Vega has been optimized and tweaked for years and Xe is fresh new which means Xe has a disadvantage at the moment which will disappear more and more in the upcoming months.
    Vega is also last-gen architecture, but the only thing that matters is actual performance.

    Renoir/Vega in a 4800U consistently outperforms Tigerlake/Xe when running at 15W. The amount ranges from a very small amount, to a pretty large one depending on the benchmark you're looking at.

    Bump the Tigerlake CPU up to 28W and it takes the lead, but it's also using more power.
    smitty3268
    Senior Member
    Last edited by smitty3268; 25 October 2020, 05:18 AM.

    Leave a comment:

  • schmidtbag
    Senior Member

  • schmidtbag
    replied
    Originally posted by bug77 View Post
    Ahem, this is a Tiger Lake CPU you're looking at.
    It's 10nm with some patches for vulnerabilities. It's still otherwise a pretty old architecture.

    Leave a comment:

  • skeevy420
    Senior Member

  • skeevy420
    replied
    Originally posted by rmfx View Post
    Michael
    Phoronix
    Michael

    There is really a problem with the benchmarks on Phoronix.
    It's never really a pleasure the read them, too many pages, too raw presentation, no dynamic visuals...

    What about a graph with on top a series of comboboxes so we can filter multiple things, and removing the need to parse 15 pages ?

    On a positive note, thanks for these, it's interesting to see on the last page how they compare with other brands.
    Pay for Premium. You'll actually get some of those features.

    The less-that-pleasant presentation is a way to nag people to buy premium. The joke's on him. I'm dyslexic so I prefer smaller chunks of information separated by multiple pages and I don't like webpages where I sing to myself:

    Scrollin', scrollin', scrollin'
    Keep them bargraphs scrollin'
    Damn my finger's swollen
    Rawhide!

    Leave a comment:

  • dwagner
    Senior Member

  • dwagner
    replied
    I'm not surprised the Xe drivers worked well during this test. I mean, when I click through commits at
    https://git.kernel.org/pub/scm/linux.../i915?h=v5.9.1
    I can read plausible, concise comments on what was changed and why, and the code changes seem to reflect the stated intent, even on a superficial glance by the non-involved.
    Quite in contrast to this, when I click through commits at
    https://git.kernel.org/pub/scm/linux...m/amd?h=v5.9.1
    the comments there are mostly terse half-sentences devoid of essential information, and the code changes look incomprehensible and often like random attempts on doing something unexplained.

    Leave a comment:

  • mikkl
    Junior Member

  • mikkl
    replied
    Originally posted by smitty3268 View Post
    Xe is only slightly ahead of Renoir/Vega right now, and I expect they'll fall behind again once the next gen cpus launch next year.

    However, congrats to Intel on finally competing with their iGPUs. That's important because software devs can start relying on this as the new baseline GPU performance everyone will have in a few years.

    Xe is a lot better than Vega, even the 80 EUs version of Xe LP at 1300 Mhz easily beats Vega 8 running at 1750 Mhz. Vega in Renoir runs much higher clocked and loses. Another factor is that Vega has been optimized and tweaked for years and Xe is fresh new which means Xe has a disadvantage at the moment which will disappear more and more in the upcoming months.

    Leave a comment:

  • knweiss
    Junior Member

  • knweiss
    replied
    Michael
    Phoronix
    Michael The title "oneAPI Level Zero Tests - Test: Host-To-Device Bandwidth" appears twice.
    knweiss
    Junior Member
    Last edited by knweiss; 24 October 2020, 06:58 AM.

    Leave a comment:

Working...
X