Hi, yesterday I did a lot of tests to compare the overall render and OpenCL performance of latest production build of Houdini on latest Mac OS X 10.11.1 and CentOS 7. And what I've found shocked me. Render time on CentOS was 15-27% faster then on Mac OS X and OpenCL was 44% faster! I love Mac OS X but this get me to switch permanently I used same scenes, same settings and on the same hardware (the same machine). For example one render on OSX took 9:57 min but on CentOS 8:00, other 3:01 vs 2:27, OpenCL on Mac 1:52 min vs 1:18 Linux etc…
I was using MACs for abut 20 years, but I think this is for me a game changer - to get up to 44% boost in performance for free only by switching to another operating system. I think it could help people decide too.
Do you have similar experience? How could be so big differences technically?

Yoop , I did some tests before too and I take same result like you.

I tested a same project on Linux (Mint) , Windows 7 and Mac.
Linux is faster than others.

Also you can check reading and writing speed on the normal and SSD drives too.
writing and reading time on the Linux is faster than others ,(Specially windows)
this can be useful when we have large cache files for Pyro or Flip simulations 8)

My MAC is on SSD while my Linux is on classic HDD and still beats the OS X. I am sad about that because I love Macs.

Perhaps send in some tests to SESI to try to help the developers optimise OsX Mantra rendering.

Bullet simulation could be 50% slower in H14 but in H15 it now matches Windows and quite possible Linux.

Francoise
My MAC is on SSD while my Linux is on classic HDD and still beats the OS X. I am sad about that because I love Macs.

Also i did another test for memory usage between Windows 7 and linux.
Linux used 25% fewer memory than Win7.
Of course I heard Microsoft use different method for memory management in windows 10 , But I didn't test it !

Memory Management in Linux is perfect for Dynamic Simulations 8)

welcome to the real world :-)

now you know why apple can show speed up's of 20-50% and more. they let the api rotten for some years and then they do a update and praise in there show we are the best we made speed ups of xx %….

i don't think that sesi can fix the apple faults of updating or optimizing there frameworks….

Hi Mandrake, so you as a developer can definatelly say, that this is Apples fault? I mean - why is that exactly? What exactly is causing to be Mac OS X so slower then Linux? I just want to be clear. I am trying now to make from CentOS my friend, but after a lot of years with Macs easy of everything it is quite hard for me

I'm curious what machine you're using that boots Linux and OS X. It's not supported by Boot Camp

You can install Linux on any machine. It has nothing to do with bootcamp. Bootcamp is for win.

But my question is still unanswered - What exactly is causing to be Mac OS X so slower then Linux? When I look into “about Houdini” the output is quite the same, only difference is Mac OS X 10.11 is using maximum OpenGL 4.1 but CentOS ver. 4.5 - but it has nothing to do with CPU compute time or OpenCL which specs are the same on both.

The CPU executes code that is written by the system and application programmers -therefore there are two possibilities, Apple's OsX code is not efficient or the SideFx code is not as efficient on OsX.

You can also ask how Windows is not as fast as Linux too.

Yes, sure there are just these two possibilities, so I just wanna know, which one is correct If it is Apples code or Sidefx code, what is causing OS X to be slower. If it is sidefx fault, then I believe they can or do their best. But if it is Apples fault, then the correction never comes I think

I can say for sure that OpenGL is both an older implementation, as you wrote, but also Apple does not supply good debug tools, so you waste a lot of time guessing at what is going wrong, instead of optimising. I hear that OpenCL is the same too.

For C++ coding, we a real coder to answer that

On CentOS I am running my Quadro on OpenGL 4.5 version, the same card on OS X is running just 4.1. I am discussing here right because I don't want to waste my time. If there will be any chance the render and OpenCL speed on OS X in near future will be the same as on CentOS, I will stay on OS X, because the system is much more friendly for everything. Otherwise I will move to Linux. Thats all.

Did you post that Mantra test file into to SideFx? That's the most sure way to get the devs to look at the file to see if there is a simple fix. Also do the same for OpenCL.

Overall there is nothing from Apple or SideFx that indicates an imminent speed up on the OsX platform as H15 was released 7 weeks ago, and major changes optimisations usually happen on full releases. Apple's OsX has improved in OpenCL previously, i.e 10.9.2/3 to 10.9.4/5, so it may be possible but nothing has been announced afaik.

I would stick with Linux to be sure for the greatest speed.

No I didn't post any test file, beacuse the speedup is markant on EVERY scene I have tested, so anybody can do their own test with ANY scene. SumaSumarum Aproximately 20% faster render time and 40% faster OpenCL on CentOS 7 even with slower disk vs OS X 10.11 on SSD. Linux is great performer, but has terrible controls. So better render times I got earned I spent right away with messing with UI, shortcuts, Terminal, backup etc.