Announcement

Collapse
No announcement yet.

51 GPUs Tested, From The Radeon HD 2900XT To RX 580 & R9 Fury: Testing The 2017 Linux Driver Stack

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

  • MacgyverF
    replied
    You always make great reviews, I just miss more AAA's that challenge the mid-high GPU, like GTX1060, 70, 80, RX580, RX Vega. When launching RX Vega, I would like a review including Hitman, Deus Ex Mankind Divided, Shadow of Mordor, Mad Max, Borderlands: The Pre-Sequel, BioShock Infinite, Alien: Isolation, The Talos Principle; Using high or ultra preset, 1080p / 1440p / 2160p. Of course it does not need as many boards, it could just be 1050ti / RX470 to up.

    Leave a comment:


  • coder
    replied
    Originally posted by existensil View Post
    It's kind of surprising to see that in a way I was wrong... in certain workloads 10-year-old CPUs are indeed ~10x slower.
    Certainly not in single-thread performance. His benchmarks don't show that.

    Leave a comment:


  • leipero
    replied
    Tho test were quite limited, it still shows that ATI/AMD GPU's age better. The difference since nvidia 600 series did/will drop, but it all depends on their drivers. Interesting comparison.

    Leave a comment:


  • eydee
    replied
    A year or two ago these graphs usually looked like this: All nvidia cards on top, even the slowest one beating the fastest AMD card. It is nice to see that linux performance is now much closer to what the actual hardware can do.

    Leave a comment:


  • valici
    replied
    Thanks Michael. Another benchmark i wanna see is how NVidia and AMD card scale with resolution. Like overlaid bars for 4k, 1440 and 1080 is modern games.

    Leave a comment:


  • Michael
    replied
    Originally posted by schmidtbag View Post
    Wouldn't it be easier if you plugged in maybe 3 or 4 GPUs into a single motherboard at the same time, and test them individually? Most GPUs can't even saturate PCIe 2.0 at 16x, so it's not like any slower slots are going to skew results that much.
    No, really wouldn't be any easier, aside from automating PTS to then run 4 times before needing to intervene and swap, but still really wouldn't be much of a time saver and could open the situation up to driver problems or other hypotheticals I am sure some people would then complain about.

    Leave a comment:


  • schmidtbag
    replied
    Originally posted by Michael View Post
    One at a time... A number of days.
    Wouldn't it be easier if you plugged in maybe 3 or 4 GPUs into a single motherboard at the same time, and test them individually? Most GPUs can't even saturate PCIe 2.0 at 16x, so it's not like any slower slots are going to skew results that much.

    Leave a comment:


  • Azrael5
    replied
    you have to consider these tests only for gaming purposes, while you have to consider OPENGL specifications for the main desktop tasks. Nowadays a VGA able to support up to 3.3 opengl specification should be enough.
    Last edited by Azrael5; 02 June 2017, 02:36 PM.

    Leave a comment:


  • dungeon
    replied
    Originally posted by Holograph View Post

    "once legendary HD 2900XT"

    I assume you mean legendary for how much the HD2000 series sucked.
    Hey, respect HD mark there... most people were running CRTs or 4:3 or 5:4 LCDs when that appearead
    Last edited by dungeon; 02 June 2017, 01:31 PM.

    Leave a comment:


  • Holograph
    replied
    Correction:

    "Dota with its OpenGL renderer would run on the card back to the HD 3850, albeit very slowly... The RX 580 is 22x faster than the Radeon HD 3850"

    It's 20.2 times faster, or 21.2 times as fast. If you want to round a .2 up, fine, I guess, but it would be 21x faster, not 22x. But I'd recommend saying 20x. It's the most correct statement possible while rounding to an integer.

    If something is "faster" than something else by a specific amount, you add that to the original amount. I mean, otherwise, with your logic, something that is "50% bigger" would generate math of "y = x * 0.5" which would actually be 50% lower in value because you didn't add it to the original value. (Note: this is language-independent.)




    Sidenote:

    "once legendary HD 2900XT"

    I assume you mean legendary for how much the HD2000 series sucked.
    Last edited by Holograph; 02 June 2017, 01:26 PM.

    Leave a comment:

Working...
X