Hey All! I'm trying to speed up my mod map development pipeline (thanks for the splitskybox fix Tony!) and have now hit a couple issues with vrad,
(I know this is the coders list, but since the issue is with Orangebox SDK and all the brains are here I believe I will get a sensible answer here!) Firstly I was looking for improved compile times with the assumed linear improvements from multithreading in vvis and vrad. The benches here: http://arstechnica.com/reviews/hardware/amd-phenom-x3- review.ars/5, here: http://www.anandtech.com/showdoc.aspx?i=3272&p=11 and here http://www.anandtech.com/printarticle.aspx?i=2868 had me convinced that upgrading from an e6600 to a q6600 would result in significant vvis and vrad execution time reductions. Sure enough, vvis times almost halved when going to quad-core. Awesome. But vrad times went up!! Here are some quick test results: Custom map compile, default settings, vvis only: 1 core: 5:09 2 cores: 2:35 3 cores: 1:45 4 cores: 1:22 Custom map compile, default settings, vrad only: 1 core: 4:13 2 cores: 3:22 3 cores: 3:18 4 cores: 4:08 Suspecting something was wrong with the orangebox vrad, I then set about compiling sdk_dm_lockdown with ep1 tools and compared that to sdk_dm_lockdown with orangebox tools. The results were surprising: Ignoring vvis (which improved linearly as expected for both ep1 and ob), the vrad times were: Ep1, 2 cores: 3:15 Ep1, 4 cores: 1:46 OB, 2 cores: 1:24 (zomg wtf!!!) OB, 4 cores: 1:44 Now whilst I am aware that the q6600 is 2 x c2d and not a true quad (and therefore can experience bottlenecks due to the shared bus) I am still suspicious that something is not right. Further to this, I sought more optimization by using vmpi. This works beautifully for both vvis and vrad in ep1 but only vvis works in orangebox. In OB vrad with mpi, the master crashes just after the worker requests '--plightdata--'. Is this a known issue? I would attempt to debug this myself, but unfortunately I cannot seem to compile the OB vrad due to a crapload of linker errors.... So I guess I am asking if: a) Is there something I am missing here? b) Has anyone else come across this? c) Are these known issues with OB vrad? d) Is it reasonable to expect a fix for at least the MPI crash on OB vrad? Thanks! CD _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlcoders
