Mac Pro, Metal & moving forward

   7402   89   4
User Avatar
Member
71 posts
Joined: Nov. 2015
Offline
goat
The Intel GPU is officially unsupported

I can’t be 100% certain if it really used the GPU.
But I think it would throw an error if Cl is on and it can’t use the GPU.
Not sure about that either though.
User Avatar
Member
362 posts
Joined: Dec. 2006
Offline
Christoph F
goat
Mac was 20-50% slower than Linux
Windows was 20% slower than Linux

what are those numbers based on?
meaning, is this sim time, disk load time, render time mantra, ..? or a combination of all?

Not to start the whole freakin OS discussion myself but Linux was always mentioned to be quickest at a lot of stuff and especially with disk operations. I wonder if this is still the same with ssds or even more so.

I think it is mostly viewport and OpenCL calculations. Houdini is entirely OpenGL app, I think even UI. And problem is that OpenGl and OpenCL and now CUDA are dead techs on OSX. It does not matter what hardware you use. Best GPU on world will not help you there. Especially now when Apple does not support them, neither is willing to do any bug fixes. You have worst experience in Houdini, Blender, MODO, MARI (compared to other OSs) … name it, every app that fully utilize OpenGL, OpenCL or CUDA (which at moment, basically all of them are). Blender for example can not be used for GPU rendering on Catalina, due issues with OpenCL and no CUDA support. Latest MARI for OSX is not updated for almost a year cause they had to work more on Qt and OpenGL which was easy thing to do on Win and Linux. See, all these things can affect you as artist. And this will be case until those apps are ported to Metal, if they ever (I doubt Blender ever will for example). It is really simple as that. We hope situation will change in future but imho it will never be as on Windows or Linux platform, at least you will be limited cause many developers will not find interest into porting to Metal. Those are really facts. Personal insults will not change that.
Edited by SreckoM - Dec. 9, 2019 16:14:36
User Avatar
Member
23 posts
Joined: Dec. 2019
Offline
SreckoM
Christoph F
goat
Mac was 20-50% slower than Linux
Windows was 20% slower than Linux

what are those numbers based on?
meaning, is this sim time, disk load time, render time mantra, ..? or a combination of all?

Not to start the whole freakin OS discussion myself but Linux was always mentioned to be quickest at a lot of stuff and especially with disk operations. I wonder if this is still the same with ssds or even more so.

I think it is mostly viewport. Houdini is entirely OpenGL app, I think even UI. And problem is that OpenGl and OpenCL and now CUDA are dead techs on OSX. It does not matter what hardware you use. Best GPU on world will not help you there. Especially now when Apple does not support them, neither is willing to do any bug fixes. You have worst experience in Houdini, Blender, MODO, MARI (compared to other OSs) … name it, every app that fully utilize OpenGL, OpenCL or CUDA (which at moment, basically all of them are). Blender for example can not be used for GPU rendering on Catalina, due issues with OpenCL and no CUDA support. Latest MARI for OSX is not updated for almost a year cause they had to work more on Qt and OpenGL which was easy thing to do on Win and Linux. See, all these things can affect you as artist. And this will be case until those apps are ported to Metal, if they ever (I doubt Blender ever will for example). It is really simple as that. We hope situation will change in future but imho it will never be as on Windows or Linux platform, at least you will be limited cause many developers will not find interest into porting to Metal. Those are really facts. Personal insults will not change that.

I actually 100% agree with you on these points. Apple banning these techs was absolutely ridiculous, even though I truly do believe in Metal 2 being the future. Because until that future is here, we as artists still need those techs to do our jobs as efficiently as possible. My hope is that these developers will fully embrace Metal 2 and take advantage of Optimizing these apps for Mac. As an example, due to Maxon acquiring Redshift, and C4D's biggest userbase by far being Mac, Redshift is going to take FULL advantage of the new Mac. I know personally that Apple is handling the majority of writing this for Metal 2 optimization, and as a result, we are going to see the full power of what is capable when these apps are optimized for Apple's new hardware and tool suite.

All of this said, of course it would've been much more pleasant and a much stronger transitional period for Mac users if Apple allowed the older techs to still be supported and fully developed in the Mac environment in the meantime.
User Avatar
Member
4130 posts
Joined: June 2012
Offline
Christoph F
goat
The Intel GPU is officially unsupported

I can’t be 100% certain if it really used the GPU.
But I think it would throw an error if Cl is on and it can’t use the GPU.
Not sure about that either though.

There are a few ways to check. In the Houdini terminal, located at:

/Applications/Houdini/Houdini18.0.310/Utilities

the command ‘hgpuinfo -c’ and it will show the OCL device selected. eg.

OpenCL Platform           Apple
Platform Vendor           Apple
Platform Version          OpenCL 1.2 (Aug 31 2019 04:58:52)
OpenCL Device             AMD Radeon RX 580 Compute Engine
OpenCL Type               GPU
Device Version            OpenCL 1.2 
Frequency                 300 MHz
Compute Units             36
Device Address Bits       32
Global Memory             8192 MB
Max Allocation            2048 MB
Global Cache              0 KB
Max Constant Args         8
Max Constant Size         64 KB
Local Mem Size            32 KB
2D Image Support          16384x16384
3D Image Support          2048x2048x2048


You can also see what is happening during a simulation with an environment variable. If you open the Houdini terminal and enter '
export HOUDINI_OCL_REPORT_MEMORY_USE=1
then
houdini
the terminal will echo:

===========OpenCL Memory Usage===========
Total Memory Allocated: 452 MB
In Memory Pool (cur / max): 0 / 512 MB
# of Buffers in Memory Pool: 0
In InUse List: 0 MB
Active Memory: 452 MB

When an error occurs, something like the following will show:


===========OpenCL Memory Usage===========
Total Memory Allocated: 452 MB
In Memory Pool (cur / max): 452 / 512 MB
# of Buffers in Memory Pool: 2
In InUse List: 0 MB
Active Memory: 0 MB
OpenCL Exception: Failed to create compute grid. (-4)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: Failed to create compute grid. (-4)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: Failed to create compute grid. (-4)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: clCreateBuffer (-61)
OpenCL Exception: Failed to create compute grid. (-4)
===========OpenCL Memory Usage===========
Total Memory Allocated: 452 MB
In Memory Pool (cur / max): 452 / 512 MB
# of Buffers in Memory Pool: 2
In InUse List: 0 MB
Active Memory: 0 MB
OpenCL Context error: [CL_INVALID_BUFFER_SIZE] : OpenCL Error : clCreateBuffer failed: No device in the context supports the requested allocation size (911141568 bytes)!
OpenCL Exception: clCreateBuffer (-61)

@callie_btw Thanks for the feedback. The invitation to show us the macOS specific accelerations in Houdini is always open.
User Avatar
Member
362 posts
Joined: Dec. 2006
Offline
callie_btw
I actually 100% agree with you on these points. Apple banning these techs was absolutely ridiculous, even though I truly do believe in Metal 2 being the future. Because until that future is here, we as artists still need those techs to do our jobs as efficiently as possible. My hope is that these developers will fully embrace Metal 2 and take advantage of Optimizing these apps for Mac. As an example, due to Maxon acquiring Redshift, and C4D's biggest userbase by far being Mac, Redshift is going to take FULL advantage of the new Mac. I know personally that Apple is handling the majority of writing this for Metal 2 optimization, and as a result, we are going to see the full power of what is capable when these apps are optimized for Apple's new hardware and tool suite.

All of this said, of course it would've been much more pleasant and a much stronger transitional period for Mac users if Apple allowed the older techs to still be supported and fully developed in the Mac environment in the meantime.

Yes, I always assumed that C4D will be first one to support Metal cause of size of their Mac userbase.
User Avatar
Member
7 posts
Joined: May 2015
Offline
So now that pricing is out on the mac pro, I wanted to see what a mid-range config would look like.

16 core xeon, 48Gb ram, 580x, 2Tb ssd - mac pro: £8,200
24 core amd tr, 64Gb ram, quadro rtx 4000, 2Tb ssd - scan custom prebuilt ugly black case: £3,800

For houdini, I really can't see anything justifying that price difference.
User Avatar
Member
23 posts
Joined: Dec. 2019
Offline
Matt Morris
So now that pricing is out on the mac pro, I wanted to see what a mid-range config would look like.

16 core xeon, 48Gb ram, 580x, 2Tb ssd - mac pro: £8,200
24 core amd tr, 64Gb ram, quadro rtx 4000, 2Tb ssd - scan custom prebuilt ugly black case: £3,800

For houdini, I really can't see anything justifying that price difference.

I feel like mid-range for a $60,000 machine is more around $20k - $40k

And in that case, we are looking at:

2.5GHz 28‑core Intel Xeon W processor, Turbo Boost up to 4.4GHz
192GB (6x32GB) of DDR4 ECC memory
Two Radeon Pro Vega II Duo with 2x32GB of HBM2 memory each
2TB SSD storage
Apple Afterburner card
Stainless steel frame with feet
Magic Mouse 2
Magic Keyboard with Numeric Keypad - US English

which comes out to $29,000

The comparison would have to be:

System Configuration:
Chassis Model: Aventum 3 PRO
Exterior Finish: Black Metallic Matte Finish
Trim Accents: - Standard Factory Finish
Processor: Dual CPU 56-Core Total - 2x (Intel Xeon Platinum 8180 (28-Core) 2.50 GHz (Up to 3.80 GHz)
Motherboard: SuperMicro X11DPi-NT (Dual 10GbE LAN) (IPMI Access) (TPM Module) (Up to 3x PCI-E Devices)
System Memory: 192GB DDR4 2666MHz (LP ECC Registered) (4 Channel Memory)
Power Supply: 1600W Corsair AX1600i (Digitally Controlled Power)
Storage Set 1: 1x SSD M.2 (500GB Samsung 970 EVO PLUS) (NVM Express)
Storage Set 2: 1x Storage (2TB Seagate / Toshiba / Hitachi)
Graphics Card(s): 1x PNY Quadro P400 2GB (Outputs: 3 x Mini DisplayPort 1.4)

141816 – PROMOTIONAL LIMITED TIME BONUS DISCOUNT: $500 1 ($500.00) ($500.00)
Shipping: Calculator
Subtotal: $31,724.00

Unfortunately this is as close to 8PCI slots as they could configure…but to get that 8 PCI slots, with the 24-core, the 192 gigs of ram, 4 GPU's, and that dedicated coding card for 8k Prores Raw (later to include R3D raw, etc…), gonna have to go with the Mac Pro.

THAT SAID…

If you ONLY use Houdini, Octane, and other CUDA based softwares, there is ZERO reason for you to purchase a Mac Pro.

If however, you are a generalist and use Houdini, Octane, AND Cinema 4D, FCPX, the Adobe Suite, the Red Giant suite, Redshift, and regularly have to edit footage in the 6k - 8k range quickly, or run Logic Pro or Pro Tools and handle studio sessions and run 100's of virtual instruments next to live instruments and audio recordings and handle sound mixes, etc…in other words…if you are running a general post production setup, this may be something to look into.

I cannot understand anyone who ONLY runs Houdini and CUDA based softwares even discussing a Mac…at all, at any level. PC all the way, literally throw together that ugly black case custom system that Matt mentioned and you have an awesome rig for what you do.

For me though, this system is perfect and i'll be picking up the configuration I used as an example end of January.
User Avatar
Member
7 posts
Joined: May 2015
Offline
I can see the appeal for editing/colouring, particularly if you're already committed to the mac ecosystem, and FCPX. the other apps like octane & redshift I'm pretty sure will be faster on rtx/cuda. Looking forward to seeing benchmarks next year.

Not sure that I'd go with that config for the ‘mid-range’ pc though - something like: https://uk.pcpartpicker.com/list/493fyk [uk.pcpartpicker.com] would chew through pretty much any workload. 17k for 32 cores, 256Gb ram, 2 4tb drives, 2 quadro rtx 8000 (the expensive bit, memory pool of 96Gb though) Still has a couple of pcie4 slots free for extra cards.
User Avatar
Member
23 posts
Joined: Dec. 2019
Offline
Matt Morris
I can see the appeal for editing/colouring, particularly if you're already committed to the mac ecosystem, and FCPX. the other apps like octane & redshift I'm pretty sure will be faster on rtx/cuda. Looking forward to seeing benchmarks next year.

Not sure that I'd go with that config for the ‘mid-range’ pc though - something like: https://uk.pcpartpicker.com/list/493fyk [uk.pcpartpicker.com] would chew through pretty much any workload. 17k for 32 cores, 256Gb ram, 2 4tb drives, 2 quadro rtx 8000 (the expensive bit, memory pool of 96Gb though) Still has a couple of pcie4 slots free for extra cards.

I hear you. And yes, I'm equally excited to see the benchmarks, particularly after Redshift and Octane finish optimizing to take advantage of the new system.
User Avatar
Member
4130 posts
Joined: June 2012
Offline
callie_btw
The problem with that theory is assuming the fiber link doesn't get Houdini to register it as ONE device…as is the case with DaVinci Resolve, per their CEO, who was excited and shocked upon realizing this…maybe we should wait and here what SideFX has to say about it.

Found the reference for this claim; we can essentially rule out any special gains for Houdini on the Mac Pro duo, or infinity fabric linked, GPUs, as Blackmagic coded custom software themselves:

To take advantage of what Apple has created, Blackmagic had to implement ways to take advantage of the Mac Pro’s use of multiple graphics processing units on a single card.

“It was something we had to implement,” Petty said “It’s providing a huge improvement,” in part because the Afterburner system’s architecture avoids the relatively time-consuming process of sending information back and forth onto the main processor.

https://www.forbes.com/sites/dbloom/2019/06/06/apples-pricey-new-mac-pro-will-be-a-hollywood-hit-says-blackmagicdesign-ceo/amp/ [www.forbes.com]
  • Quick Links