Has anyone with access to the new Mac Pro had a chance to try Houdini on it? We are considering adding one of the new Mac Pros for doing some video work to our set-up and I've been curious if it would also work well with Houdini? Specifically, I'm wondering about the AMD based graphics cards in it.

Thanks

-carlos

Only for Houdini, you can get Dual Xeon machines, which will serve you much better for Houdini, you will get almost double the speed compared to single CPU for the same price you would pay for a MAC PRO, dual Video cards will not give you any advantage inside houdini, AMD firepro drivers has something called Crossfire Pro , but I am not sure if Houdini can take advantage of it. Anyway, Houdini is the master of FX, thus better have two cpu then one, as far as I am concerned Apple build this machine for OpenCL video editing.

Thanks for the hijack, Tricecold, but I guess the question is “what is the new macpro like for houdini?” not “what would be a better machine than the macpro?”, to which we all know the answer, more or less…
I'm also curious about this one. Not that I would ever buy it though…

Cheers

I haven't tried it as of yet, but performance is probably similar to the performance of the Radeon 7950 you can get for the Mac. I'd expect roughly double the performance of the AMD 5770. You're still going to run into some of the inefficiencies of Apple's GL2.1 driver as we have not migrated to their core GL3 driver for H13.

I believe one GPU is reserved for graphics, and one for compute. As our CL support is disabled on the Mac due to driver issues, that means Houdini will only ever use the one GPU. But since you can't get a new Mac Pro with 1 GPU, that's a bit of a moot point.

twod
…You're still going to run into some of the inefficiencies of Apple's GL2.1 driver as we have not migrated to their core GL3 driver for H13.
I believe one GPU is reserved for graphics, and one for compute. As our CL support is disabled on the Mac due to driver issues, that means Houdini will only ever use the one GPU. But since you can't get a new Mac Pro with 1 GPU, that's a bit of a moot point.

Bummer…

I have a new mac pro, the 6-core model, with 32GB ram, 512gb Drive upgrade and the D700 video card upgrade… I also run an external Pegasus2 raid drive. I can't say much tho because as of yesterday it was my first time ever using houdini, I'm just playing around with some of the instructional videos because i've always really wanted to learn 3d / VFX type stuff, i am a sound designer / music producer… but if you give me maybe a file to load up i could test render it out and see how it does.

What i can say is this machine feels 1000x's nicer then my Mac Pro 2008 8-core i was running… its probably mostly the crazy hard drive, but everything is so responsive and nice, computer doesn't make any noise at all, although the RAID system spin drives do make a bit of noise. But i get basically 1000mb/sec transfers on my internal and external setup.. It really is a pretty amazing piece of machinery. Not saying its the most powerful thing for houdini specifically but its very nice. I am running 3 Dell u2711's with it as well.


Mark

I have a new mac pro, the 6-core model, with 32GB ram, 512gb Drive upgrade and the D700 video card upgrade… I also run an external Pegasus2 raid drive. I can't say much tho because as of yesterday it was my first time ever using houdini, I'm just playing around with some of the instructional videos because i've always really wanted to learn 3d / VFX type stuff, i am a sound designer / music producer… but if you give me maybe a file to load up i could test render it out and see how it does.


Mark

Great - thanks Mark - I'm interested in the graphics performance.

Can you test the ‘MeshedSea3Mil.hip’ in the link below, scroll down to get it, and see what the fps in the viewport reads when you tumble the view.

http://www.sidefx.com/index.php?option=com_forum&Itemid=172&page=viewtopic&t=29308&start=25 [sidefx.com]

There's a thorough review of the new Mac Pro by Dave Girard over at Arstechnica:

http://arstechnica.com/apple/2014/01/two-steps-forward-a-review-of-the-2013-mac-pro/ [arstechnica.com]

Couple of brief Houdini mentions, as well.

Thanks for the link to Girard's extensive review. Looks like Apple's inadequate OpenCL and OpenGL drivers are still holding back the capabilities of the new hardware:

The lackluster support of OpenCL is likely the biggest thing standing in the way of software transitioning from CUDA to OpenCL on OS X. Apple needs to stop just talking big about OpenCL and offer the aggressive support it needs to actually compete with CUDA. It's a sad irony that Apple invented OpenCL only to see it better supported on competing platforms and that dealing with Apple to resolve problems is “unpleasant” because the company is so opaque.

http://arstechnica.com/apple/2014/01/two-steps-forward-a-review-of-the-2013-mac-pro/5/ [arstechnica.com]

Not totally convinced about the drivers holding it back; my brief testing of a more optimised app, aka Mari on the older Mac Pro booted in Ubuntu and Mavericks showed they were about equal.

Maybe someone has done more extensive testing though.

AMD and Nvidia provide the low-level driver, but it's Apple that writes the front-end implementation of OpenGL and OpenCL that acts as the middle-man between the application and the driver. Because it's a different GL/CL implementation, it has different performance characteristics, features and bugs than the proprietary AMD and Nvidia implementations. One of the glaring omissions in their OpenGL implemenation is GL_KHR_debug (or GL_AMB_debug_output) which would be very useful for debugging when GL drops to software. The Q&A with Neat Video on page 5 of that review sounds very familiar, just replace “CL” with “GL” So I believe that's what he's referring to, not so much the raw performance itself, but the ease of maximizing performance.

I'd imagine every engineer/software developer working at Apple would agree with this from page 5:

We would ask them to let AMD directly support OpenCL on AMD hardware on Mac, so that AMD could develop a full AMD APP SDK for Mac OSX, make the performance of OpenCL on Mac match that of the same GPUs in Windows/Linux, and implement all the missing features that are currently available in other OSes but not on Mac.

..but they lack the authority to do it

I had hoped that in the post-Jobs era Apple that the GL/CL implementations might have opened up a bit to allow third party ones, but I suppose they have their reasons.

Update to the Ars review - the nMP openGL was being held back by a ‘bad’ .env setting in Maya.

2013 Mac Pro FirePro D700 OpenGL is better than we thought it was

http://arstechnica.com/apple/2014/02/2013-mac-pro-firepro-d700-opengl-is-better-than-we-thought-it-was/ [arstechnica.com]

Have been doing some render tests with the new MacPro, an iMac, and my older linux box. (Have not tested the graphics performance of the MacPro yet.)

MacPro: 12-Core Xeon E5 2.7 Ghz, 64 GB, latest Mavericks
iMac: Quad-Core i7 3.1 Ghz, 16 GB, latest Mavericks
Homebuilt: Quad-Core i7 2600k 3.4 Ghz, 16 GB, Debian Squeeze

The linux box is the license server and the ifd's and image files are being written to the linux box. I've tried heavy-geometry scenes and low-geometry scenes and the relative render times seem consistent, (network overhead doesn't seem to be much of a factor).

Times for a test render:
MacPro: Render Time: 1:17.09.45u 26.27s 3:16.87r
iMac: Render Time: 57:47.42u 5.26s 7:17.35r
Homebuilt: Render Time: 46:28.23u 4:34s 5:52.02r

I know there are other factors besides Ghz, but these numbers don't add up for me. I'm used to Ghz-relative speed-ups in render times from previous upgrades to newer linux systems.

I was expecting the Mac Pro to render somewhere around 2.35 times faster than my linux machine. The render times for the MacPro and iMac are pretty close, relative to their processor speeds.

Is OSX just generally slower running mantra when compared to linux?
Are there efficiency issues with high core-count processors?
Are there some special settings for rendering on OSX?

Cheers

fgillis
Is OSX just generally slower running mantra when compared to linux?
Are there efficiency issues with high core-count processors?
Are there some special settings for rendering on OSX?

Very interesting. Thanks for testing. The very best test would be to install Linux on the nMP and see if it's the processor or OS.

Have finally been able to test the graphics performance of the new macpro side-by-side with my 2 year-old linux box.

MacPro: 12-Core E5 2.7 GHz, 64 GB, dual AMD FirePro D700 6GB each
Linux: 4-Core i7 2600K 3.4 Ghz, 16 GB, GeForce GTX580 3GB

Tested with houdini 13.0.343.

Loaded up meshedsea3mil_111.hip, (link from an earlier post), and tumbled in the viewport using GL 2.1 on the MacPro and GL 2.1 on Linux, (GL 3.3 looked terrible with High Quality lighting on my linux box).

Headlight Only: D700: 86fps, GTX580: >120fps
High Quality Lighting: D700: 37fps, GTX580: 45fps
High Quality with Shadows: D700: 36fps, GTX580: 41fps (but no shadows)

GTX580 also had noticeably faster playback on a couple of Houdini animation projects I've loaded up for testing.

Let's hope that the OsX GL3+ drivers are faster than the GL2 ones, for the next Houdini

fgillis
Is OSX just generally slower running mantra when compared to linux?

On a like-for-like test - same machine and scene - Mantra appears to ~8% slower on OsX 10.9.2 vs Mint 16