Announcement

Collapse
No announcement yet.

GCC vs. LLVM Clang Compiler Performance On The AMD Threadripper 2990WX Benchmarks

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

  • hajj_3
    replied
    It would be nice if you could re-do your test Michael with Clang 7.0 that was released a couple of weeks after your benchmarks and try the latest nightly builds of gcc and clang too.

    Leave a comment:


  • andrei_me
    replied
    chithanh Yeah, adding to your argument, if you take the best perf of 2990WX on Clear from Phoronix's graph, AMD would be the winner at ~100k points instead of Intel at ~93k

    Leave a comment:


  • chithanh
    replied
    Originally posted by edwaleni View Post
    he was deferring to those he felt represent Linux better in the benchmark space.
    I think it was rather looking for figures which support his argument. What I find particularly notable is that he used this graph from Phoronix:

    And this one from techspot:

    to conclude that the 7980XE is ahead of the 2990WX in Linux. However, Phoronix tested the 7980XE too and found it to score slightly lower, around 93k points compared to the 2990WX's 94k points on Ubuntu.


    Leave a comment:


  • edwaleni
    replied
    Originally posted by chithanh View Post
    I posted this in the original 2990WX benchmark thread earlier: This is not an excuse. I mean if you can't make informed statements about or perform tests on Linux fine, don't do that then. If you want to make an informed statement, do it like Hardware Unboxed who actually installed Linux for the first time and tried their best.
    All I got out of it was that he made a premise that there is an inconsistency in WX test results and used Michael's graph as merely an example. To me rather that install Linux himself, he was deferring to those he felt represent Linux better in the benchmark space. In this case he referred to Phoronix.

    If anything, it showed to me he reads Phoronix and tipped his hat to Michael's work.



    Leave a comment:


  • chithanh
    replied
    Originally posted by edwaleni View Post
    When Gordon worked for Future, (MaximumPC/PC Gamer) he probably wouldn't have let this drop. He was good at getting details on the issues, even when it included Linux. But he had a lot more help and editorial freedom there too.
    I posted this in the original 2990WX benchmark thread earlier: This is not an excuse. I mean if you can't make informed statements about or perform tests on Linux fine, don't do that then. If you want to make an informed statement, do it like Hardware Unboxed who actually installed Linux for the first time and tried their best.

    But do not post asinine stuff like the comment about the 7-zip scores. As the Techgage reviewer noted, 7-zip results can be very unstable on Windows (varying between 55K and 85K MIPS between runs), so any single measurement is of questionable validity anyway.

    Leave a comment:


  • edwaleni
    replied
    Originally posted by DavidKL View Post
    In case you haven't seen the mess of an article that just hit from PCWorld...

    • Uses a Microsoft product (Visual Studio) to check whether or not another Microsoft product (Windows) is hampering the 2990
    • Clickbait headline doesn't include mention of potential Windows performance hit problems, or the possible issue with 7-zip code changes
    • Mentions that your 7-zip code is older but claims testing twice with only the latest version proves that 7-zip code changes can't be the culprit
    • Doesn't bother to do any Linux testing
    • Doesn't bother to try a different compiler

    On a good note, at least he mentions your testing and posts one graphic from it, before seeming to try to dismiss it.
    In defense of Gordon Mah Ung at PC World:

    The guy does just about everything there, podcasts, webcasts, editing web and print. He relies heavily on Brad Chacos on looking under the covers.

    PCWorld is a computing generalist publication and I wouldn't expect it to start pulling code out to prove something. But in this case Gordon did a basic level of research.

    There is no doubt Gordon knows who butters his bread. To call out Windows to have an explicit issue, when AMD, Intel and Microsoft are all heavy advertisers in PCWorld and other IDG publications, he clearly had to dance a line and still come out and say there is a problem.

    He used Michael's data (and duplicated one of his tests) to prove that not only TR has a bandwidth issue, but Windows has an issue.

    When Gordon worked for Future, (MaximumPC/PC Gamer) he probably wouldn't have let this drop. He was good at getting details on the issues, even when it included Linux. But he had a lot more help and editorial freedom there too.

    After reading all of these reviews across the ether, and especially seeing Phoronix results, I did post at PCWorld that Windows clearly has an issue with TR and someone needed to look at it. I knew they were Windows focused. They have a large readership. If anyone should do it, they should.

    Fortunately Michael is not beholden to the corporate world, only his premium subs. Microsoft has made it clear in the commercial world that if you make hay out of benchmarks at our expense, we can make it ugly for you. The SQL Server benchmark wars in the 90's. The COM library slow downs in Win98/NT. Microsoft started putting a crimp on commercial exposes on Windows and performance. The only exception was security.

    Unfortunately, the computer "press" has shrunk to its bare threads. In another era, PC Magazine would be all over this. Today, its up to Phoronix and other online media to call it out. I am glad they have, I don't care if its Linux or Windows. The numbers don't lie and can't be ignored.

    Leave a comment:


  • orome
    replied
    Originally posted by austin754 View Post

    Note: The Phoronix Test Suite includes two solutions of the n-queens problem. One of them is named "m-queens" and has the following description:

    Run Identifier: pts/m-queens-1.0.1
    Description: A solver for the N-queens problem with multi-threading support via the OpenMP library.
    ah, a clever wordplay ...
    thanks. I stand corrected.

    Leave a comment:


  • DavidKL
    replied
    In case you haven't seen the mess of an article that just hit from PCWorld...

    • Uses a Microsoft product (Visual Studio) to check whether or not another Microsoft product (Windows) is hampering the 2990
    • Clickbait headline doesn't include mention of potential Windows performance hit problems, or the possible issue with 7-zip code changes
    • Mentions that your 7-zip code is older but claims testing twice with only the latest version proves that 7-zip code changes can't be the culprit
    • Doesn't bother to do any Linux testing
    • Doesn't bother to try a different compiler

    On a good note, at least he mentions your testing and posts one graphic from it, before seeming to try to dismiss it.

    Leave a comment:


  • celrod
    replied
    Originally posted by Anty View Post
    One thing to report. When I was doing some tests with GCC/G++ ver 8.0.1 (and sometimes 7.3.0) using -march=znver1 in some cases would generate worse code than just adding -mavx2 -mfma. I would need to test ver 9.x.x if it is still the case.
    I found the same. For numerical code, -march=haswell (or adding those flags) tends to do a lot better than -march=zenv1.
    Part of the problem is zenv1 prefers 128 bit vectors over 256 bit, but -march=haswell is still often faster than -march=zenv1 -mprefer-vector-width=256.

    Leave a comment:


  • johanb
    replied
    Originally posted by anarki2 View Post



    iPhone SE, Safari, iOS 12 beta 12 or something, but it's been like this since forever. Does the same in Firefox too.

    No, you can't scroll horizontally or "zoom out" either, the whole site just scales down:
    I'm on firefox on android, same layout but scrolling horizontally works fine.
    Sounds like a buggy browser.

    Why are you even testing both Safari and Firefox on iOS anyway?
    Apple has forced that all browsers in AppStore has to use Webkit, so firefox on iOS is just Safari with the ability to sync your bookmarks with a firefox account and a different UI.

    Leave a comment:

Working...
X