Announcement

Collapse
No announcement yet.

Concerns Over No PAE Kernel In Ubuntu 12.04 LTS

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

  • #46
    Originally posted by yogi_berra View Post
    I've never seen a dead VAX 8600, but I'm sure they exist. His point about cheap hardware crapping out sooner than it should is not countered by your experience with different manufacturers.
    But different manufacturers used Pentium Ms. One company was cited and countered with another. The point is that Banias cores are still in existence and in use. I personally was using until late last year a Thinkpad R50 with a Banias core for my job.

    Comment


    • #47
      Originally posted by sbergman27 View Post
      Originally posted by chithanh View Post
      Sorry to break the news to you, Ubuntu has no security support for most of its packages.
      You're trolling. Canonical provides full security support for 18 months on regular releases.
      [...]
      You can optionally install packages from Universe and Multiverse. And they may or may not get security updates.
      Last I checked, the universe repository was enabled by default in the sources.list. So IMO universe counts towards the Ubuntu packages. And it contains close to 3 times as many packages as main. Then there is restricted with its software manager which nags you to install proprietary software. Both universe and restricted come without any clear commitment to security updates.

      Originally posted by Shining Arcanine View Post
      This is not true. Gentoo still supports i486 processors and I imagine that doing a Gentoo installation on a i386 system is possible (although not supported).
      glibc stopped supporting i386 around 2.4 so it is not possible to install a "normal" Gentoo system on a 386 any more. uclibc will probably be possible, but has other drawbacks.

      Comment


      • #48
        Pentium M's are said to be old and slow in this topic but actually they can compete very well with today's lower end Atom models.

        They are absolutely not too slow for modern Ubuntu versions and many people are still using them. There are a lot of Pentium M laptops which came with 1GB of RAM.

        I think it's the best to use a bootloader that automatically detects the presence of PAE.

        Comment


        • #49
          Originally posted by AlbertP View Post
          Pentium M's are said to be old and slow in this topic but actually they can compete very well with today's lower end Atom models.

          They are absolutely not too slow for modern Ubuntu versions and many people are still using them. There are a lot of Pentium M laptops which came with 1GB of RAM.

          I think it's the best to use a bootloader that automatically detects the presence of PAE.
          I somewhat agree that they are still reasonably viable, but they are absolutely the bare minimum you can run an OS with, but the graphics are still not that great. AFAIK, they're actually still faster than Atom(except maybe the fastest dual-core), Atom is pretty slow. A $280 Brazos laptop absolutely slaughters a Pentium M.

          The Pentium M era was right at the peak of when Dell, HP, and most others were building their absolute worst quality PCs. I'd be willing to wager that at least 85% of Pentium Ms have been recycled by now, and the rest are hand-me-down PCs that are probably on their 5th owner.

          This is an excellent segway into the Bulldozer-bashing that's running rampant. Bulldozer is somewhere in the neighborhood of 30 times faster than Pentium M, unless you pull out the most wretchedly single-threaded benchmark you can find, then it's only perhaps 4x faster. I'm hearing from everybody now that Pentium M is good enough for people with basic needs. Does that make Bulldozer fast, or is it still slow?

          Comment


          • #50
            Exactly, Pentium M stuff was pretty craptacular, my old PM Banaias 1.5Ghz w/ GMA900 and 512Mb of ram is quite allot slower then my much older P4 Northwood 1.8Ghz w/ i845GL chipset that can only use the VESA driver, the only thing it has better is Gb of ram yet for day to day tasks that both are capable of the P4 seems much more responsive even without a working GPU driver.

            And yes, this is from the guy that runs Ubuntu on a 2002 800Mhz G4 w/ Radeon 7500 and 1Gb of PC133.

            Comment


            • #51
              Why are people only talking about Pentium M? Pentium II & Via C3 systems are often used as thin clients, and work well for that... (even better, I tried a Via C3 thin client as a regular desktop a year ago, and even that worked fine for light use like browsing the web).

              Comment


              • #52
                If the non-PAE Pentium M models work, you should also be able to get Pentium II working, unless Ubuntu will start requiring SSE as well, which is supported by Pentium III and M but not II. I don't know about VIA C3.

                I agree, for simple web browsing a Pentium II/III is enough, though it may not run modern desktops well. But there's LXDE to get around that.

                Comment


                • #53
                  Originally posted by JanC View Post
                  Why are people only talking about Pentium M? Pentium II & Via C3 systems are often used as thin clients, and work well for that... (even better, I tried a Via C3 thin client as a regular desktop a year ago, and even that worked fine for light use like browsing the web).
                  While thin client hardware that old may run the "bare ass basics" how much power are they using to do it and what things would you like to do that they are too slow to do? Are you going to recap the boards in the case of dried out capacitors? I'd rather get a bunch of 5.9w 1Ghz dual core 64 bit capable systems with a GPU capable of OpenGL4 and OpenCL like the embedded version of the AMD Z series. They're better in every aspect and will likely run for a solid 10 years since they can easily run fanless and be fully supported with the OSS drivers in that time.

                  I only keep my old Macs around as I'd been a Mac user since the IIe and have a ton of old Mac software and games I like to mess with from time to time. Commodity x86 on the other hand is dime a dozen and at least on open source OSs your old software will run with a minimum of modification.

                  Comment


                  • #54
                    No, even 10 years old thin client hw does not lose in power consumption much compared to stuff selling today. I have some fanless Via C3 hw that runs 7-22W (idle-full power).

                    That beats most Atoms, and is only slightly undercut by a modern AMD rig like the one you mention. Besides a 6W cpu, consider the chipset, mobo and psu.

                    Further, even a 486 can do thin client duty, these systems don't age that fast.

                    Comment


                    • #55
                      Originally posted by curaga View Post
                      No, even 10 years old thin client hw does not lose in power consumption much compared to stuff selling today. I have some fanless Via C3 hw that runs 7-22W (idle-full power).

                      That beats most Atoms, and is only slightly undercut by a modern AMD rig like the one you mention. Besides a 6W cpu, consider the chipset, mobo and psu.

                      Further, even a 486 can do thin client duty, these systems don't age that fast.
                      They are APUs, not CPUs, they have the memory controller and north bridge chips as well as the GPU all rolled into the GPU.

                      AMD rates their TDP as worst case scenario, so if it says 6w it'll be no more then that.

                      I know full well a 486 can do it, hell a 5Mhz 8086 can. But if you ever need more then the most basic of basic tasks you're SOL.

                      Comment


                      • #56
                        They're still not SoCs, though AMD hinted they will be in the future. Currently you still need extra chips for sata, usb, ethernet...

                        Comment


                        • #57
                          Happy to be on the Gentoo side here where I configure my kernels to my needs anyway. Didn't ever play with PAE since I never needed it. The only machine here with >4G RAM is an amd64 anyway.
                          I wonder if a PAE active kernel boots correctly on a non PAE machine. If not, well Geode LX, C3 and some Pentiums are still around and then the non-PAE-normal Kernel shouldn't be canceled.
                          Even on my old machines I use very recent kernels and I am happy that I can still use them.
                          Moreover I wonder if it is really a big burden for the Canonical devs to ship a kernel image without PAE. Maybe it's a small thing for them but a big relief for users out there.

                          Comment


                          • #58
                            Originally posted by Adarion View Post
                            I wonder if a PAE active kernel boots correctly on a non PAE machine.
                            As has already been pointed out numerous times, the answer is NO.

                            Comment


                            • #59
                              To get the best of both worlds you could program the CD's isolinux/syslinux bootloader to check for PAE, and then boot the proper kernel. Syslinux is capable of this (as has already been pointed out) so isolinux should be able to do it as well. Then the installer could check which kernel is running and install that kernel. In this situation NX bit will work if it's there, >4GB will work on PAE-capable machines, and Intel Pentium M and II, AMD Geode LX, and VIA C3 will still run Ubuntu.

                              Comment


                              • #60
                                Originally posted by Adarion View Post
                                Didn't ever play with PAE since I never needed it. The only machine here with >4G RAM is an amd64 anyway.
                                As I argued in a previous post in this thread, it's often better to use 64-bit kernel if you have 1GB or more memory. The reason is the 3G/1G memory split. By default the Linux kernel present to each process an address space with 3GB for the process and 1GB for the kernel. In the kernel space the physical memory is mapped. There are 128MB reserved for kernel use so only 896MB of memory can be mapped and used. There are 3 solutions to this problem.

                                1. You can try with 2G/2G or 1G/3G split. Then the address space for the kernel is increased at the expense of the process-usable memory
                                2. HIGMEM - there is a special memory zone called HIGHMEM where the memory over 896MB resides. If the kernel is to use it it has to map it into the kernel VM area first. This creates an overhead
                                3. Use 64-bit kernel

                                1 and 2 have nothing to do with PAE. PAE can be used on <4GB systems to get the execute disable feature. You only really need it if you have more than 4GB memory. Curiously, if you have 4GB RAM you have more than 4GB of physical memory area. Some of the physical memory area below 4GB is reserved for IO devices and most notably the video card. You can get down to 3GB or even less physical memory below the 4GB area if you have a mighty video card with lots of memory (although typically the values are between 3.25 - 3.5GB). Even with PAE you wont be able to address the memory above 4GB if the kernel is not compiled with the 64GB HIGHMEM option (I may be wrong about that but it makes sense to me). Usually those are the server kernels in Ubuntu.

                                There's a lot of confusion about these related but different issues. That's why I generally recommend using 64bit kernels on anything with 1GB or more. You get rid of the HIGMEM memory zone and the associated overhead. You can still run 32 bit userspace and every process can use the complete 4GB of virtual address space. There might be small overhead because 32-bit kernel calls have to be translated to the 64-bit ones. But the HIGHMEM zone incurs overhead too. And the 64-bit mode should offer better performance thanks to the extended register file.

                                Comment

                                Working...
                                X