art3mis Posted December 24, 2016 Share Posted December 24, 2016 (edited) I ran two different simulations for exactly 15 minutes on the exact same hardware and simply tracked the number of frames reached, repeating several times i7 6900 k 64 gb RAM Titan XP GPU Indie v 15.5.673 Linux - 15.5.632 Windows Windows Simulation 1 OpenCL on -140 frames OpenCL off -138 frames Simulation 2 OpenCL on -53 frames OpenCL off -49 frames Linux(Ubuntu) Simulation 1 OpenCL on -169-170 frames OpenCL off -160-165 frames Simulation 2 OpenCL on -61 frames OpenCL off -51 frames The hip for Simulation 2 attached. Its from the H15 workflow enhancements webinar. I used default cache settings (attached) Also I think that the Open CL comparisons might not be 100% accurate since AFAIK there is no global on-off switch for OpenCL and I only toggled the single Use OpenCL switch on the FlipSolver node. CompressedFlip_benchmark.hiplc Edited December 27, 2016 by art3mis Quote Link to comment Share on other sites More sharing options...
shawn_kearney Posted December 24, 2016 Share Posted December 24, 2016 Mind uploading your hip? Quote Link to comment Share on other sites More sharing options...
art3mis Posted December 24, 2016 Author Share Posted December 24, 2016 (edited) Shawn, please see my edited post. The hip for Simulation 2 is from the H15 workflow enhancements webinar. The only edit I made was to remove the whitewater nodes. Edited December 24, 2016 by art3mis 1 Quote Link to comment Share on other sites More sharing options...
shawn_kearney Posted December 24, 2016 Share Posted December 24, 2016 (edited) Cool, I'll check it out. Like you I've seen improvements on Linux I had not seen previously using Windows, but in my experience it was pretty substantial. While your core speed is faster, I do have more threads (I think FLIP is multi-threaded?) and your GPU should be substantially faster than my M4000, so I'm a little surprised you're not seeing the openCL performance that I am. It will be interesting to see how Quadro compares against high-end GTX for FLIP. Does anyone know if FLIP is calculated at single or double-point precision? Edited December 24, 2016 by shawn_kearney Quote Link to comment Share on other sites More sharing options...
LukeLetellier Posted December 24, 2016 Share Posted December 24, 2016 I want to go Linux if only for the memory improvements; it's miserable on windows. At the same time, I always get rather frustrated with Linux. I just want to double click something to install a program, not enter lines of code. I have anough technical things to learn in Houdini & other 3D programs that are directly related to earning a living; I don't want to have to learn a whole new OS Quote Link to comment Share on other sites More sharing options...
shawn_kearney Posted December 24, 2016 Share Posted December 24, 2016 As a student I kind of feel liek I have to learn linux, at least as an end-user. So I went in head first and deleted my windows partion completely so I wouldn't fall back on it. I think that's the only way to really learn it, otherwise I'd just boot back into Windows when it get tough. But if I weren't learning, I doubt I'd switch at all. And #$%&ing pulseaudio! I really wish that they'd just use jack instead. It's SO much more robust and reliable and doesn't get in the way of everything. Quote Link to comment Share on other sites More sharing options...
edward Posted December 26, 2016 Share Posted December 26, 2016 Did I read that right, that the tests are being run with caching to disk? If that's the case, the performance of Windows NTFS might be a bottleneck and then you're not really benchmarking compute work at all. Quote Link to comment Share on other sites More sharing options...
Mandrake0 Posted December 26, 2016 Share Posted December 26, 2016 i think you can test all day long and the end result for usage would be always: LINUX -> Production in a 24x7 Environment (Example: CG Movies) WINDOWS -> Mix Production Setup (Example: Games, Small Studio) Mac OSX -> Small Production / Learning Quote Link to comment Share on other sites More sharing options...
art3mis Posted December 26, 2016 Author Share Posted December 26, 2016 (edited) edward, yes perhaps. Please feel free to do your own testing with the attached .hip and share your results. Regarding OpenCL, perhaps someone more knowledgeable than I can explain why there doesn't seem to be any sort of global override 'Use Open CL' switch? If you allow editing of contents on the FlipSolver node there are easily a couple dozen nested nodes, each with their own 'Use Open CL' toggle. Puzzling. Edited December 27, 2016 by art3mis Quote Link to comment Share on other sites More sharing options...
Guest tar Posted December 26, 2016 Share Posted December 26, 2016 On 25/12/2016 at 6:58 AM, shawn_kearney said: As a student I kind of feel liek I have to learn linux, at least as an end-user. So I went in head first and deleted my windows partion completely so I wouldn't fall back on it. I think that's the only way to really learn it, otherwise I'd just boot back into Windows when it get tough. But if I weren't learning, I doubt I'd switch at all. Another option is to use virtual machine to learn/keep your Linux skills up to date. Linux is all fun until it breaks, i.e. you update and upon the restart you are left in Grub! That gets boring real quick when all you really want to do is vfx... With a VM you can also try all flavours of Linux, and, even BSD. Re: OsX I don't recommend it but it can and does work too in a pinch, but, booting the same machine into Linux is like getting a car tuned. Quote Link to comment Share on other sites More sharing options...
Guest tar Posted December 26, 2016 Share Posted December 26, 2016 3 hours ago, art3mis said: Regarding OpenCL, perhaps someone more knowledgeable than I can explain why there doesn't seem to be any sort of global override 'Use Open CL' switch? If you allow editing of contents on the FlipSolver node there are easily a couple dozen nested nodes, each with their own 'Use Open CL' toggle. Puzzling. OpenCL isn't a panacea for speed. It requires memory to be laid out differently. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.