No announcement yet.

Zstd 1.5 Released With Big Performance Improvements

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Phoronix forums: where (some) great minds fight like in kindergarten. Michael, when will you change this forum software to be more reddit like? (works great for arstechnica)


    • #32
      Originally posted by birdie

      I don't remember ever opining about G-Sync. Isn't it amazing that you're putting words into my mouth? Go on with your vendetta against me.

      Yet, not a single AMD fan here has anything to say about Mantle or about the fact that AMD doesn't properly support its ... CPUs under Linux. Yeah, your idol open source company.
      What's there to really say about Mantle? I had one game that used it, Battlefield 4, where on my crappy ass R7 260x it performed better than DX11. That was my only Mantle experience. It was a good one. I guess your experience wasn't a good one?

      However, very soon after than it was forked into the open standard we know as Vulkan and depreciated outside of the games that used it. Mantle was around long enough for people to go "Well that kicks ass. That needs to be done in a way that benefits everyone." And now we have Vulkan that everyone, including AMD, contributes to. Vulkan 1.0 is basically Mantle 2.0 in spirit. Even MS based DX12 around it and Vulkan. While on one hand it could have been seen as lackluster at the time depending on what GPU you had, on the other hand it became Vulkan that is, IMHO, one of the biggest game changers we've ever had on Linux.

      I do agree that their hardware support under Linux could be a bit better. I'm trying not to be too critical since they've either hired or are trying to hire Linux developers. If you have the skills and abilities they'll pay you to fix all the complaints you have. I don't have those abilities nor do I think they take apprenticeships from 36yo dudes who have been contractors for the past 16 years.

      Point is: Now that they have money coming in they're spending some of it on Linux. I'm going to try to stay positive for the next year or two because of that.

      Many 10th generation Intel CPUs are better in terms of value per dollar than AMD offerings - not to mention they come with a stock cooler (which only the 5600X has) and have an iGPU while AMD requires a discrete GPU.
      That's disingenuous unless you're only talking about the only 5000 CPU line there? I don't think the 5000 line has APUs based on them released yet. AMD tends to tick-tock their newest CPUs and APUs. The CPU is the tick and a bit later they'll tock with the same thing with an APU added on...sometimes with less cache like my 4650G APU. I assume they take CPUs that didn't cut it as a higher end model and slap an iGPU on. I reckon that you'll start seeing Zen 3 APUs in the coming months...probably named 6000 series because AMD does that for some confusing reason. For example, my 4650 is a 3600 with an iGPU and less cache. IMHO, it should be named R5 3650G. That is what it is.

      And on that price/performance chart --- all the newer CPUs from both brands have crappier price/performance ratios as more cores are added. Even my grey market APU was priced like that. The 4 and 6 core APUs were reasonably priced but the 8 core one had an extra premium price extra $70 more than it should have been. They know people are willing to spend money on the high end parts.


      • #33
        Originally posted by birdie
        1) I have never advocated for G-Sync
        2) I've never said it's a good tech
        3) I indeed called it a vendor lock-in but not entirely since current G-Sync monitors are VSync compatible for fuck's sake. The first GSync monitors only worked with NVIDIA cards. I've never denied that.
        Points 1) and 2) have absolutely nothing to do with my posts. I have never claimed that. As for 3), if you want to use G-sync feature (and not alternative solutions), you are locked to NVIDIA GPUs. You were pathetically trying to defend NVIDIA on this point by claiming that this is not a vendor lock-in and you have failed.

        Either way, I have no intentions in arguing with you any further, so I've deleted my posts to de-escalate this of topic. Let's not make a shit show in this thread, like you always try to do in this forum.

        BTW, have some self respect and actually ignore me since I am in your ignore list. My original post was a reply to another user.
        Last edited by drakonas777; 15 May 2021, 09:26 AM.


        • #34
          birdie, you are so terrible. I can't believe it.

          If you were to stop going off-topic, insulting us, f**king us off and advertising as your last resort defense tools to keep your beloved evil safe on EVERY thread (even if it has nothing to do about it, like this very one) it would be great.
          I have shut the door for you. Have a nice day.
          Last edited by tildearrow; 15 May 2021, 01:13 PM.


          • #35
            Originally posted by perpetually high View Post
            Is it time for me to switch everything in my kernel config from LZ4 to zstd? Any reason not to?
            Well, XEN does not support zstd and I have to handbuild all kernels because every single distro like imediatelly switched over.


            • #36
              I don’t get how it’s better or more interesting than lz4. It compress barely better but it’s still way slower.


              • #37
                Originally posted by cl333r View Post
                I just don't see why it's called Zstd,
                because it's a shortening of Zstandard.
                Because Yann Collet wanted to make a new standard for compression (zipping).

                Originally posted by cl333r View Post
                where's the Facebook in it?
                It's not called Facebook-{wahtever}, because back when it started Yann Collet wasn't even working at Facebook.
                He has pretty clearly documented the developemnt on his blog.
                The idea was to combine the extremely fast dictionary search (including the exhaustive, tough still fast-ish one variant) as he previously designed LZ4, together with more modern and faster implementations of entropy coding (including the FSE he designed out of Jarek Duda's ANS papers).
                Basically something more modern but roughly in the same ballpark as GZip(*).

                Facebook hired him around the time cyann released the first beta of ZStd.


                (*) as opposed to where most of the research in compression was heading back in those days, which is using a ton of expensive machine learning/modelling to better predict entropy: PAQ can use some neural-nets, PPMd owes its name to partial-prediction-match and in LZMA, MA stands for MArkov chains.
                These can compress much better than the dictionary+entropy combos such as Gzip's deflate or as Zstd, but at large computationnal cost.
                Meanwhile Zstd is "just" GZip done better / more modern.


                • #38
                  Originally posted by birdie View Post
                  A corporation developed a really nice fast and efficient free software compressor for Linux and this is how you're thanking them?
                  No one should be 'thanking' face book for anything. They're as bad as Google, Microsoft, and Apple, perhaps more so given their proclivity to censor and ban people and ideas at will. FckFB.