Announcement

Collapse
No announcement yet.

Raptor Computing Is Working On More AMD Radeon Driver Improvements For POWER

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

  • Raptor Computing Is Working On More AMD Radeon Driver Improvements For POWER

    Phoronix: Raptor Computing Is Working On More AMD Radeon Driver Improvements For POWER

    Raptor Computing Systems, the libre hardware company behind the POWER9-based Talos II server board and Blackbird micro-ATX desktop, has been working to improve the open-source AMD Radeon graphics driver support for IBM POWER...

    http://www.phoronix.com/scan.php?pag...GPU-POWER-Bits

  • #2
    Ooh, nice! Definitely makes POWER systems more practical and useful.

    I wonder if the ARM64 improvements for the driver, are for Samsung phones.

    Comment


    • #3
      This is excellent. POWER-based systems are still the only non-backdoored, user-controllable option that is competitive in terms of performance.

      Comment


      • #4
        Can anyone speak to the experience of using POWER as a desktop? If I wanted to run Ubuntu as an example, I'd need to install the server version and then a graphical environment, which isn't that bad (I think Fedora even has a desktop ISO) but what is the experience like after that? Are there major pains or is it identical to an x86 system sans proprietary software?

        Comment


        • #5
          It's kind of a misnomer to say these are "liberated" systems. There was never anything here that needed "liberating" in the first place.

          cynical I've been looking at getting one myself for a while, working out the tradeoffs. Even within the realm of the open source, there's still quite a bit that doesn't get written or tested for POWER, and so doesn't work correctly. (this is an improving situation however)

          One that I ran into is the lwjgl library for the Java JVM is only available for x86 and arm. There isn't a POWER port for it yet, thought there has been some interest. Likewise, the open-source MultiMC launcher isn't normally run on anything other than x86, so there are probably issues there.

          The use case that made me start to consider it was a NAS and for that I'm told that ZFS works, and so should NFS. Beyond that, it requires research into what you want to do.

          Comment


          • #6
            Originally posted by cynical View Post
            Can anyone speak to the experience of using POWER as a desktop? If I wanted to run Ubuntu as an example, I'd need to install the server version and then a graphical environment, which isn't that bad (I think Fedora even has a desktop ISO) but what is the experience like after that? Are there major pains or is it identical to an x86 system sans proprietary software?
            While I am biased, as an individual user (I have a couple of POWER desktops, one personal and one larger one at work) I really don't see any day to day differences from x86, except for occasionally having to compile something from source because third party (non-distribution) repos are almost always x86-only. That being said, even on x86 I tended to compile such software from source anyway for security reasons, but I know I'm a bit unusual in that respect.

            Developer12 is correct in that Java has a few issues. Most of those issues are traceable in some way or other to Google software, as they don't tend to accept patches for architectures they don't care about (sadly, they only care about x86 and ARM, two quite closed architectures overall). That is more of a governance issue / issue with the community accepting "pushed source dumps" (for lack of a better phrase) vs. insisting on organic development to use Google tools and libraries -- it's not a technical problem as much of that software has gained unofficial POWER patches over the past couple of years (including Chromium itself, which I use pretty much daily on POWER).

            Having used x86, ARM, and POWER, I'll say that x86 is of course the easiest to use, but the nagging feeling of being monitored tends to change what you do / say / view on the computer (chilling effects). POWER is the next easiest; if you have a POWER version of the software it's just like x86 (speed, responsiveness, power, etc.). ARM tends to really feel like you're using a different architecture even aside from the custom kernel requirements for each board (yes this is changing with SBSA, but proprietary firmware size and scope is also growing very quickly on SBSA platforms, likely for silicon vendor profit reasons moreso than anything else). I'm not entirely sure why this feeling is so prominent, but I do know there tends to be some fairly nasty lag on many ARM devices, plus limited expandability to fix any performance issues encountered. Useable? Yes, within the limitations, but I wouldn't go so far as to say the existing ARM devices are going to be drop in replacements for x86 desktops / workstations the way that POWER has already proven to be.

            FWIW here's the Debian archive coverage per architecture:
            https://buildd.debian.org/stats/graph-week-big.png

            POWER's within a couple percent of x86_64 in terms of coverage, which is pretty good considering the Debian maintainers don't usually make any effort to bring in existing POWER enablement patches for a number of the packages in the archive -- they generally wait for upstream to merge them, then for a new upstream release to happen, which means that if a Debian package isn't available for POWER there's a good chance you can pull the upstream source and build it on POWER anyway.

            While this may not exactly be a fair comparison, I do feel the need to compare the two main open ISA systems in terms of software availability. POWER's at ~97% coverage in the main repositories (comparable to x86_64 which is at ~99% coverage), while RISC-V is sitting at around 82% coverage and is relegated to the ports repositories. Note ppc64 is not ppc64le; almost all large commercial development of software for POWER is centered around ppc64le right now.
            https://buildd.debian.org/stats/grap...s-week-big.png

            I hope this helps!
            Last edited by madscientist159; 12-08-2019, 07:41 PM. Reason: Clarify ARM / SBSA

            Comment


            • #7
              Originally posted by jacob View Post
              This is excellent. POWER-based systems are still the only non-backdoored, user-controllable option that is competitive in terms of performance.
              they are not competitive in terms of performance compared to members of top500. if you've meant "performance per dollar" instead of "performance', then they aren't competitive either

              Comment


              • #8
                Originally posted by cynical View Post
                Can anyone speak to the experience of using POWER as a desktop?
                linus can. he used power for several years to make sure it gets attention from kernel, but ditched it in the end

                Comment


                • #9
                  Originally posted by pal666 View Post
                  linus can. he used power for several years to make sure it gets attention from kernel, but ditched it in the end
                  Sorry, I know for a fact that's not true.

                  While I get you're probably rooting for Team Red (how's that mandatory PSP on your Zen 2 chip treating you anyway?), try not to spread baseless FUD. It doesn't help anyone.

                  Comment


                  • #10
                    Developer12, madscientist159 Ty both! That was very helpful. It sounds like POWER is in a good place then, actually. I don't mind recompiling if necessary (I used to run Gentoo) I just prefer the convenience of binaries for most things these days.

                    Originally posted by madscientist159
                    is correct in that Java has a few issues. Most of those issues are traceable in some way or other to Google software, as they don't tend to accept patches for architectures they don't care about (sadly, they only care about x86 and ARM, two quite closed architectures overall). That is more of a governance issue / issue with the community accepting "pushed source dumps" (for lack of a better phrase) vs. insisting on organic development to use Google tools and libraries -- it's not a technical problem as much of that software has gained unofficial POWER patches over the past couple of years (including Chromium itself, which I use pretty much daily on POWER).
                    Strange given that they actually use POWER servers. I guess they aren't that enthusiastic about open platforms.

                    Originally posted by madscientist159
                    Having used x86, ARM, and POWER, I'll say that x86 is of course the easiest to use, but the nagging feeling of being monitored tends to change what you do / say / view on the computer (chilling effects).
                    Yes but not only that, the idea that someone might be able to take control of my hardware remotely due to some backdoor is alarming. (even if it was an unintentional thing, like an exploit for their remote management code) I have some interest in cryptocurrency, but how can you store value on hardware that you do not trust? I'm a developer mostly interested in the web and Java, which are nice open platforms for the most part. I stopped gaming and don't run any proprietary software any longer so I just need a reliable, secure, and fast system. Most of the chat stuff you might use to work with colleagues has web or open source clients anyway, so I think I am covered.

                    Originally posted by madscientist159
                    I'm not entirely sure why this feeling is so prominent, but I do know there tends to be some fairly nasty lag on many ARM devices, plus limited expandability to fix any performance issues encountered.
                    I might consider ARM for a laptop, but never a workstation. I really dislike their lack of a standardized boot process.

                    That graph is especially helpful. I think 97% of Debian should satisfy me lol.

                    Originally posted by pal666
                    they are not competitive in terms of performance compared to members of top500. if you've meant "performance per dollar" instead of "performance', then they aren't competitive either
                    You trolling? The top two spots in the top500 are POWER systems, and they have three of the top10. Having 30% representation in the ten fastest systems in the world is pretty good.

                    Comment

                    Working...
                    X