Announcement

Collapse
No announcement yet.

Ubuntu 32-bit, 32-bit PAE, 64-bit Benchmarks

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

  • #16
    Originally posted by Drago View Post
    I also think that this huge difference is due to i486, rather than i686 optimization. Does anyone knows how 64-bit programs impact CPU on-die cache memory. Doesn't addresses in it, eat twice as much, than 32-bit code, effectively half usable cache size. How this affects smaller CPUs, like Athlon II for example?
    not everything doubles. Only integers and pointers. Everything else has the same size. So amd64 code is a bit bigger but not very much.

    Comment


    • #17
      Originally posted by clavko View Post
      It would be really, really great if someone could test Gentoo x86
      vs Gentoo x64 and show whether there are any gains to achieve.
      Both with SSE2/SSE3, gcc optimizations and so on. Phoronix is
      just full of 'Ubuntu this, Ubuntu that' - well Ubuntu is just one
      member of a big GNU/Linux family, give some love to other distros
      as well.

      "So pretty please... with sugar on top. Clean the f-car."
      Winston Wolf

      do it yourself :P

      Comment


      • #18
        Originally posted by Drago View Post
        I also think that this huge difference is due to i486, rather than i686 optimization. Does anyone knows how 64-bit programs impact CPU on-die cache memory. Doesn't addresses in it, eat twice as much, than 32-bit code, effectively half usable cache size. How this affects smaller CPUs, like Athlon II for example?
        There's certainly an increased code size and memory usage, but keep in mind that the default operand size is still 32bit. Addresses are 64bit, so pointers and function calls takes twice the memory, but overall the memory footprint isn't doubled.
        I haven't seen a real analysis on this, so I can't give you numbers...

        Comment


        • #19
          AFAIR when I switched on disk usage increased by ~25%. RAM usage increased by roughly the same amount.

          But that was many many years ago

          Comment


          • #20
            We're demanding Phoronix to provide a i686 benchmark too! )
            Fedora 12 is i686!

            Comment


            • #21
              Originally posted by energyman View Post
              AFAIR when I switched on disk usage increased by ~25%. RAM usage increased by roughly the same amount.

              But that was many many years ago
              What do you mean "increased by same amount" ... 100% increase?
              However, I am not concerned about system RAM usage. I am concerned about L1/L2 cache usage on my future Athlon II.

              Comment


              • #22
                no, in memery increased by 25% too, if I remember correctly.

                it will be roughly the same. Since data and commands are seperated in L1 at least data will not suffer.

                Also, while you don't have L3 with Athlon II you will get a bigger L2 (except if you are coming from an old X2 6000). 1mb/core is not too bad. More than most old X2s. And the Athlon II Regor core is faster on the same clock than the old Kuma or Windsor/Brisbane. So.. don't worry too much.

                Comment


                • #23
                  Originally posted by energyman View Post
                  1mb/core is not too bad. More than most old X2s. And the Athlon II Regor core is faster on the same clock than the old Kuma or Windsor/Brisbane. So.. don't worry too much.
                  These new Athlon IIs, have only 512KB/core, so I am worrying about it.
                  Maybe, I just should install arch i686 and x86_64 versions, and test it myself. So now I have to figure out how to run this PTS on my box
                  Do you know any other test suits, with..smaller disk footprint?

                  Comment


                  • #24
                    Using a car analogy (since we all love them)

                    We're going to compare 3 petrol types

                    Unleaded
                    Super Unleaded
                    Chemical X

                    but then not mentioning that Chemical X uses different engine components

                    Which isn't like for like

                    I guess it's difficult to compare these type of benchmarks when the target hardware is so different

                    The best way to compare would be to compile using -march=native on a 32bit and 64bit environment keeping as much the same as possible

                    Then testing pure 32bit, pure 64bit and 32bit with 64bit kernel

                    I currently have two gentoo systems set up like this would anyone be in
                    me making some benchmarks for a core2 + GM45 system and a athlon X2 + Radeon 4650?

                    Comment


                    • #25
                      Originally posted by Drago View Post
                      What do you mean "increased by same amount" ... 100% increase?
                      However, I am not concerned about system RAM usage. I am concerned about L1/L2 cache usage on my future Athlon II.
                      It is higher but you should take into account that the fastest "memory" available are the cpu registers, and you have twice as many when running 64bit code.

                      Also you need less instructions to do the same tasks, and function parameters are passed via registers rather than via stack.

                      So it should be a net gain for most applications.

                      Comment


                      • #26
                        Originally posted by Drago View Post
                        These new Athlon IIs, have only 512KB/core, so I am worrying about it.
                        Maybe, I just should install arch i686 and x86_64 versions, and test it myself. So now I have to figure out how to run this PTS on my box
                        Do you know any other test suits, with..smaller disk footprint?

                        the dual core have 1mb/core. The quad have 512kb/core L2. But as several sites have found out, no L3 does not hit the Athlon II too hard. The quad Athlon II are the fastest procs in their prize range.

                        You loose a lot of performance if you go 32bit (to conserve cache) by loosing half of your registers. And registers are the stuff you really want to have.

                        Comment


                        • #27
                          x86-64 CPUs don't actually store everything in 64 bits, that would be stupid and wasteful.
                          Code:
                           ~ $ cat /proc/cpuinfo | grep sizes
                          address sizes	: 48 bits physical, 48 bits virtual

                          Comment


                          • #28
                            Originally posted by energyman View Post
                            the dual core have 1mb/core. The quad have 512kb/core L2. But as several sites have found out, no L3 does not hit the Athlon II too hard. The quad Athlon II are the fastest procs in their prize range.
                            I am up to Athlon II 435, tripple core with 1.5MB cache total. But may change my mind, if you make me. I am a developer and compilation will be common task. Also how eclipse, and java respect >2 cores?
                            Last edited by Drago; 12-30-2009, 12:25 PM.

                            Comment


                            • #29
                              Originally posted by Ant P. View Post
                              x86-64 CPUs don't actually store everything in 64 bits, that would be stupid and wasteful.
                              Code:
                               ~ $ cat /proc/cpuinfo | grep sizes
                              address sizes	: 48 bits physical, 48 bits virtual
                              That is what is present in hardware, the ISA does use 64bit pointers and therefore all pointers are 64bit.

                              And no that is not "stupid and wasteful", also early x86_64 cpus could only address up to 40bit of memory (1TB)

                              Comment


                              • #30
                                I'd be interested in seeing a java app in your future benchrmarks (one running in a 32bit JVM vs 64bit JVM).

                                Comment

                                Working...
                                X