Announcement

Collapse
No announcement yet.

A New & Exciting OpenGL 3 Benchmark To Run

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

  • #16
    There actually is a good demos pack but it's not very demanding and I can't figure out how to do a timedemo, there doesn't seem to be a way to do it without altering the code. There's no documented timedemo, only a GUI...
    If anybody is willing to help change that, it's probably best to start off by posting to http://www.ogre3d.org/forums

    I've checked out vegastrike and it looks ancient. I don't think anybody gives a damn how it performs nowadays. No timedemo support, of course.

    Comment


    • #17
      http://gamelab.mit.edu/games/a-slower-speed-of-light/ would make for a HELL of a benchmark, but again, no benchmark mode or timedemo support

      Comment


      • #18
        Originally posted by Shnatsel View Post
        There actually is a good demos pack but it's not very demanding and I can't figure out how to do a timedemo, there doesn't seem to be a way to do it without altering the code. There's no documented timedemo, only a GUI...
        If anybody is willing to help change that, it's probably best to start off by posting to http://www.ogre3d.org/forums

        I've checked out vegastrike and it looks ancient. I don't think anybody gives a damn how it performs nowadays. No timedemo support, of course.
        The other option is using APITrace - http://openbenchmarking.org/test/pts/apitrace - to reply a recorded GL dump demo, but that's not reflective of the entire game and at last try for some games there still was some performance issues in replaying the dumps, not to mention they end up being very huge.
        Michael Larabel
        http://www.michaellarabel.com/

        Comment


        • #19
          Okay, I have a question: why the hell are you still using MD5 for checking authencity of downloaded files which you then RUN?!
          MD5 was broken long ago and has already been used in hash collision attacks to push malformed Windows updates. Seeing you download ~1Gb Xonotic zipfile and check only the MD5 is daunting. At the very least you should check one of the SHA2 family functions, e.g. SHA512.
          Ideally you'd also add a detached digital signature distributed with the profile and checked after downloading the file.

          Which brings us to authenticating the test profiles and checking that they're not malformed to download and run an exploit instead of the test binaries. I hope they're at least downloaded over HTTPS?
          Last edited by Shnatsel; 07-10-2013, 05:24 AM.

          Comment


          • #20
            md5 is far from dangerous when combined with file size check and compression.

            Sure, you may be able to create a colliding file with a few weeks' computation. But is that colliding file going to be a valid zip/gzip/bzip2 file? And with the exact same size to the byte?

            Comment


            • #21
              Also, what is EnvironmentSize and in what units is it specified?

              Comment


              • #22
                Originally posted by curaga View Post
                md5 is far from dangerous when combined with file size check and compression.

                Sure, you may be able to create a colliding file with a few weeks' computation. But is that colliding file going to be a valid zip/gzip/bzip2 file? And with the exact same size to the byte?
                It's the ZIP that makes it trivial and negates the size checks. Ever heard of rarjpegs? That works for zip files too. You can place any kind of garbage in the beginning of the file and it will still be a valid zipfile. If your malicious payload is 1Mb size, you get the rest of the gigabyte for you to fill with hash-colliding garbage. That's 99% of the file. A hash collision attack is trivial in such conditions.

                Comment


                • #23
                  Sorry, messed up the numbers. You have 99,9% of the file to yourself to fill with hash-colliding garbage.

                  Comment


                  • #24
                    Excellent point, hadn't thought of that. While both gzip and bzip2 permit such data, they make it detectable; xz doesn't permit it, zip not only permits it but says nothing. So this is a valid hole for zips.

                    Comment


                    • #25
                      Originally posted by Shnatsel View Post
                      There actually is a good demos pack but it's not very demanding and I can't figure out how to do a timedemo, there doesn't seem to be a way to do it without altering the code. There's no documented timedemo, only a GUI...
                      If anybody is willing to help change that, it's probably best to start off by posting to http://www.ogre3d.org/forums

                      I've checked out vegastrike and it looks ancient. I don't think anybody gives a damn how it performs nowadays. No timedemo support, of course.
                      a) it is just under livid development.
                      b) if you checked it out - how much time did you spent on playing it?
                      c) timedemo is just a mission to script :P

                      but hey, a bunc h of 'benchmarks' all running basically the same eninge (*quake) are really saying sooo much.

                      Besides, why is ut2004 never included? Yes, it is old. But it doesn't look half as bad as most of the *quake descendants used. And it would give a much more rounded picture.
                      What about the Unigine benchmarks. Why aren't they run?

                      Comment


                      • #26
                        Unigine benches are not GL compliant and do not run on Mesa. But I bet you knew that.

                        Comment


                        • #27
                          inb4 someone complains you can set overrides. Pampering over non-compliant software is just wrong.

                          Comment

                          Working...
                          X