No announcement yet.

How An Old Pentium 4 System Runs With Ubuntu 10.04, 10.10

  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Yeah - any of those three are probably quite usable.

    And configuring a 128MB ramzswap on 10.04 or 10.10's likely to help desktop performance w/512MB ram.

    The r9200's a questionable choice (could go with a 9600) and if there's enough HW around the SIS is quite dubious! There are a lot of boxes with i845/i865/i915 integrated graphics in use still, certainly way more than an sis+9200 combo, so it'd be good to see benchmarks with those.

    As for the CPU-related benchmarks... the P4 is a *very* odd CPU optimization wise and gcc dev is probably oriented towards the more general K8/K10 and Core 2/i7 cores.

    My own use? At SCALE I had a Compaq small-desktop i845 P4/2.66 doing a real time DV->MPEG(-1? I forget ATM) transcode for the main presentation theater (mostly to support the overflow room for keynotes) Ran 70-80% load for both days, no probs.


    • #17
      Originally posted by misiu_mp View Post
      Its really disappointing to see the new gallium drivers for old cards degrade in performance in comparison with the old mesa drivers. I have several of those old cards and was hoping to give them a new live sometime.
      Gallium basically requires a DX9 level card, so one'll need at least a i915 (someday), fx5200 or r9500. Noueavu devs were planning on getting it working on pre-DX9 nVidia chips but punted and made a regular DRI driver. r300g is faster than r300 in many things already.

      DRI2 cost 3D performance at first, and the catchup work probably hasn't been done on r100 or r200 yet. And yeah, it might even be running with vsync on...


      • #18
        Originally posted by ad_267 View Post
        Antiquated? I've got two computers with these sorts of specs in my house. And I agree with Zhick, this doesn't tell you anything about the desktop responsiveness.
        people nowadays think that anything older than 18 months is obsolete and antiquated. my pc is 4 years old (and a bit faster than pentium4 system, since it's amd64) and i still consider it an overkill for my needs.

        the main reason to get a new box nowadays would be to get a more energy efficient pc offering a similar level of performance as the old box.


        • #19
          Originally posted by bulletxt View Post
          mmm.......... quite interesting....

          Linux is really evolution. Evolution brings new "things", but loses others. Like humans loosing their tail
          GCC is also an evolution and who knows if it is not the main player here? It will be great to see how GCC makes a difference in those tests. Ext3 makes a big difference here, but maybe there's some payoff using it?


          • #20
            Originally posted by agd5f View Post
            The slower 3D performance is likely due to the tearing avoidance code in the dri2 support. A better comparison would be to look at newer distros with kms disabled.
            By, "tearing avoidance code", I assume you mean GLX Sync & Swap? That's only been introduced with the *.35 kernel (thus not effecting 10.04). I can't think of any other tearing avoidance functionality - especially because there doesn't appear to be any. KMS is a mongrel for tearing on my Radeon 9550.

            I'd suspect the performance drop would have more to do with redirected rendering.


            • #21
              i have fought with radeon cards and linux for some time now. so i am glad to see this article. i'm running a pentium 4 2Ghz, 1.5GB pc333 ram with a sapphire radeon 9600. dual sata 500GB drives handle disk i/o fairly well even at half their i/o at 1.5 GB/sec on an adaptec pci card. the radeon r250, r300 and r350 cards have been an up and down performers in linux for some time and i vacillated between debian and ubuntu, since they sometimes run different versions of xorg.

              ubuntu 9.04 performed very well with compiz and flash. 9.10 wasn't as stellar, but was acceptable. 10.04 compiz seems sluggish and flash is a jerky mess. 10.10 runs things fair, compiz is smooth and flash is playable, but not at full screen. i expected better support from an lts release. and i could run an older ubuntu, but why do that, it's all about the up to date software right. might as well run debian.

              this system has excellent specs, but the video performance has me making os decisions i am not fond of and i have several opinions on the matter, but i best leave them to myself. i'm saving my money for an intel i3! - thanks


              • #22
                What's it with Himeno? SMP optimizations? Looks scary


                • #23
                  Tests are good and shows some of the features of Pentium 4 architecture. However I think, you should do test on AMD64 platform with Athlon64 3000+ Venice and laters with this amd's architecture.


                  • #24
                    Originally posted by pmorph View Post
                    What's it with Himeno? SMP optimizations? Looks scary
                    On a single core CPU?


                    • #25
                      interesting tests. until very recently, i was running lucid on a an athlon xp 2500+ (@1.8GHz)-based setup with an nvidia 6800. The only thing it couldn't do well was HD video. otherwise, it worked fine, even games -- at least, i don't see a significant improvement in games and i'm running 2 9800 GTX+ in SLI now...

                      wish i thought to try to benchmarking before the upgrade -- it would have been interesting to see my own results.


                      • #26
                        I don't really understand the usefulness of this test. A P4 2.8ghz with 512mb of RAM isn't an "antiquated" machine. While clearly not the newest thing out there, I certainly wouldn't think it relevant for testing an OS's performance on older hardware.

                        If we really want to get some interesting results, do the tests on a P3 800mhz machine with 256mb of RAM and onboard video with 8mb of video RAM. There are still many, many people running such machines with Windows 98 or even XP.


                        • #27
                          Originally posted by kraftman View Post
                          On a single core CPU?
                          Could SMP optimized kernel bring that big penalty when running single core? No idea really, it just amazes me it got so much slower with a pure CPU/memory test.


                          • #28
                            Well I recently installed Lucid on an 800Mhz PIII 390ish Mb ram .... it runs like crap but it does run.

                            The X11 proformance is very bad as in most of these machines you end up having a cheap mach64 based card (8 mb vram if you have 4mb don't even thing about it X.x)

                            by the way the mach64 driver and a few others are *greatly* in need of maintainace they aren't bad cards and work fine for 2D and very basic 3D acelleration but the developers of these cards seem to be short on time to work on keeping them working.


                            • #29
                              GCC has changed between 4.2.4 and 4.4.3, and I suspect the march and mtune flags Ubuntu uses have changed too (or gcc's defaults). That would affect cpu-bound performance a lot, if it used to optimize for netburst and now optimizes for atom or old athlon.

                              I have a Pentium III 733 with that video card and 1GB ram. It runs Ubuntu 10.04 but the performance is worse than it was a few years ago (subjectively). I'd start with compiling everything for the right processor, using ramzswap as already recommended and maybe disabling some fancy stuff that became fashionable with faster computers (background indexing for example). A comparison of gcc 4.2.4 vs. 4.4.4 vs. 4.5.1 (4.5 branch from svn) with -march=pentium4 on that machine for cpu-bound code would be interesting. Also, try disabling HyperThreading and making sure kernel and glibc are not SMP or try playing with the HT-aware scheduler in the kernel. Benchmarks of those could be interesting.

                              The chipset is suspect. Maybe there was a driver regression and nobody noticed? Try another motherboard.


                              • #30
                                Stickies would be nice for these types of forums.

                                Five major points
                                1. Compile the kernel for Pentium 4
                                2. Set the scheduler for deadline
                                3. Drop the kernel option for ram down to less than 1 Gigabyte
                                4. Recompile GLibc to i686 and march=pentium4
                                5. Choose the Ext3 filesystem

                                You will notice a difference even if only you just set the kernel processor type to Pentium 4 and set 1 Gigabyte or lower ram.

                                I'm not up here to argue the Force hokey stuff or any of that galactic imperialism that those jerks over in the nothing wrong here camp does. We have to take a lesson from the GenT0o.

                                I have two of the P4's, a celeron and a full fledged. Speaking on behalf of my experiences with Nvidia and the 10.04 version of Ubuntu I can say at the moment I've been unable to enable compiz.

                                Lot's of people are still running these on XP. That is P4's.

                                I noticed not long ago that Ubuntu complains about the intel_gart=blacklist parameters which are the only ways I get one of my P4's to boot other than pulling the pci card out to let the intel one nurse the installations.

                                I think Canonical should just release a statement that older hardware may need to stay with an older branch of Ubuntu....