MP 7,1 GPU rendering on the new Mac Pro. News, updates, chat.

[deconstruct60]

I mean you're not wrong on a lot of things here....but one thing to point out is that the "regular" 6900XT you are comparing to has half the memory; double the memory in any graphics card is usually going to be nearly double the price. Add that to the fact that while the MSRP is $999 IF you can get one it's selling for double that, and the special form factor, and the built-in Thunderbolt AND the Infinity Fabric Link--all not available on a "regular" card and the price isn't really totally out of line.

Not just regular. AMD's relatively recent Pro W6800 card doesn't have Infinity Fabric link on it either. If Apple painted themselves into a semi-custom die then costs aren't going to match up at all. There is such a huge shortage of Navi 21 dies to dole out to the various flavors of 6800-6900 products that perhaps AMD is "holding back" their Pro W6900 with the IF link until the demand bubble is a bit more under control.

If Apple is getting something out of AMD that AMD isn't even shipping themselves then AMD is probably charging a substantive price premium if the dies are the same ( e.g, testing and binning AMD doesn't have to otherwise do for the rest of the add-in card builds ) . If it is a slightly different die then Apple blundered into a situation where assumed that AMD would have lots of extra wafer start fab capacity and it is exactly the opposite now ( and probably for next year or so).

The pricing is just a bit odd though. The 6800 duo (which has two of the same die , but lower active core counts ) is $1000 cheaper than just one die ( but binned to have far more active cores. ) . That somewhat points to Apple just not being able to get that many of the 6900's. ( And a bit of leveraging the inflation of the current widespread GPU scarcity. It is a markup on the current street prices not the suggested retail price. ). 2 to 3 years from now when widespread price bubble has likely largely diminished these prices will have even less justification. Apple may not care if trying to "kill" the market segment anyway long term.

In short, the Infinity Fabric is probably a huge cost multiplier. That amount of mac software that makes explicit calls in Metal to lean on that "value add" is pretty limited now. Somewhat similar to how Apple "bet the farm" on OpenCL coupled to the Mac Pro 2013 and then didn't delivery best in class OpenCL on the platform and number of software products that fully leveraged it was limited.

 

[rondocap]

Not just regular. AMD's relatively recent Pro W6800 card doesn't have Infinity Fabric link on it either. If Apple painted themselves into a semi-custom die then costs aren't going to match up at all. There is such a huge shortage of Navi 21 dies to dole out to the various flavors of 6800-6900 products that perhaps AMD is "holding back" their Pro W6900 with the IF link until the demand bubble is a bit more under control.

If Apple is getting something out of AMD that AMD isn't even shipping themselves then AMD is probably charging a substantive price premium if the dies are the same ( e.g, testing and binning AMD doesn't have to otherwise do for the rest of the add-in card builds ) . If it is a slightly different die then Apple blundered into a situation where assumed that AMD would have lots of extra wafer start fab capacity and it is exactly the opposite now ( and probably for next year or so).

The pricing is just a bit odd though. The 6800 duo (which has two of the same die , but lower active core counts ) is $1000 cheaper than just one die ( but binned to have far more active cores. ) . That somewhat points to Apple just not being able to get that many of the 6900's. ( And a bit of leveraging the inflation of the current widespread GPU scarcity. It is a markup on the current street prices not the suggested retail price. ). 2 to 3 years from now when widespread price bubble has likely largely diminished these prices will have even less justification. Apple may not care if trying to "kill" the market segment anyway long term.

In short, the Infinity Fabric is probably a huge cost multiplier. That amount of mac software that makes explicit calls in Metal to lean on that "value add" is pretty limited now. Somewhat similar to how Apple "bet the farm" on OpenCL coupled to the Mac Pro 2013 and then didn't delivery best in class OpenCL on the platform and number of software products that fully leveraged it was limited.

I agree - the biggest factor here for the W6900x pricing has to be the availability and the fact only Apple has this for now.

I have my W6800x Duo testing now, and it's consistently beating the 6900 XT in many cases - even beats dual 6900 XT in cases where the VRAM difference adds up.

The W6900x will of course outperform a single W6800x, but imo the W6800x Duo is generally much more powerful where the 2 gpus scale well.

I am curious how two W6800x Duos will do, where they can be scaled - testing that next. I would imagine in Resolve if they are all used and connected with IF, 4 GPUs should easily beat even 2 W6900x.

I'm really curious how the 4 GPUs will behave in FCP. Usually it seems to only like 2 GPUs, where Resolve scales better with 4. If that's still the case, then you can make the argument that two W6900x are the best option for strictly FCP users that want the absolute fastest - but even then, the performance delta isn't that huge vs a single Duo, especially not for twice the price!

 

[vel0city]

Apple joins Blender Development Fund​

 

[Romanesco]

Would you mind helping me make up my mind?

I’m on a 16-core Mac Pro (2019) with 384 GB RAM and 1xVega II Duo. I am debating between adding a W6800X Duo or getting an RTX A6000 to use under Bootcamp/ Windows. I can’t get a hold of an RTX 3090 and have no intention of buying a tower PC.

I primarily care about the performance under Octane in C4D.

For my use case scenario, Octane X on Metal has been a gross failure. They’ve overpromised, underdelivered, and took way too long for a barely working release. Speedier or not, adding another MPX module won’t fix the constant crashes. On top of that, the developer is so slow on the progress that I can no longer rely on their word.

On the other hand, CUDA is more mature and would allow me to run deep learning algorithms outside my primary use case scenario.

In your opinion, should I go for:

• Option 1: Add the W6800X Duo to the Vega II Duo and continue struggling with the miserable Octane X experience.
• Option 2: Add the RTX A6000 and double boot to Windows for 3D rendering tasks/ CUDA.
• Option 3: Don’t add anything; wait for the next generation of NVIDIA cards/next-generation Mac Pro.
• Option 4: suggest your own.

Thanks!

 

[flowrider]

^^^^

Amazon.com: Rtx 3090

Amazon.com: rtx 3090

www.amazon.com

Lou

 

[vinegarshots]

Apple joins Blender Development Fund​

Well that took long enough. Not sure how helpful this will be on the Mac side of things, unless they're willing to fund a total rewrite of the viewport from OpenGL to Metal, and possibly a full rewrite of Cycles into Metal as well.

Would you mind helping me make up my mind?

I’m on a 16-core Mac Pro (2019) with 384 GB RAM and 1xVega II Duo. I am debating between adding a W6800X Duo or getting an RTX A6000 to use under Bootcamp/ Windows. I can’t get a hold of an RTX 3090 and have no intention of buying a tower PC.

I primarily care about the performance under Octane in C4D.

For my use case scenario, Octane X on Metal has been a gross failure. They’ve overpromised, underdelivered, and took way too long for a barely working release. Speedier or not, adding another MPX module won’t fix the constant crashes. On top of that, the developer is so slow on the progress that I can no longer rely on their word.

On the other hand, CUDA is more mature and would allow me to run deep learning algorithms outside my primary use case scenario.

In your opinion, should I go for:

• Option 1: Add the W6800X Duo to the Vega II Duo and continue struggling with the miserable Octane X experience.
• Option 2: Add the RTX A6000 and double boot to Windows for 3D rendering tasks/ CUDA.
• Option 3: Don’t add anything; wait for the next generation of NVIDIA cards/next-generation Mac Pro.
• Option 4: suggest your own.

Thanks!

If you need to work with CUDA/Windows, you really should just get a Windows machine. It's way less hassle. And at some point all Macs are going to be M1, and you're not going to be able to run Windows on a Mac at that point.

 

[Romanesco]

Well that took long enough. Not sure how helpful this will be on the Mac side of things, unless they're willing to fund a total rewrite of the viewport from OpenGL to Metal, and possibly a full rewrite of Cycles into Metal as well.




If you need to work with CUDA/Windows, you really should just get a Windows machine. It's way less hassle. And at some point all Macs are going to be M1, and you're not going to be able to run Windows on a Mac at that point.

The thing is, @vinegarshots, I don’t see a reason for spending extra for a tower when I already have one. Once I move to the Apple Silicon equivalent, I can seamlessly convert the Intel Mac Pro into a Windows machine.

Also, the thought of spending extra for another 384 GB of RAM and 8 TB of storage ain’t that pretty.

 

[vinegarshots]

The thing is, @vinegarshots, I don’t see a reason for spending extra for a tower when I already have one. Once I move to the Apple Silicon equivalent, I can seamlessly convert the Intel Mac Pro into a Windows machine.

Also, the thought of spending extra for another 384 GB of RAM and 8 TB of storage ain’t that pretty.

But you're in a situation where you own a Mercedes sedan when you need a Ford F150 to get your work done. You can either keep jamming your tools into the passenger compartment of the Mercedes in an attempt to make your investment worthwhile, or you can just sell it and get the truck.

 

[Romanesco]

But you're in a situation where you own a Mercedes sedan when you need a Ford F150 to get your work done. You can either keep jamming your tools into the passenger compartment of the Mercedes in an attempt to make your investment worthwhile, or you can just sell it and get the truck.

I’m not seeing how selling at a loss then spending extra on inferior consumer components of a PC tower serves your case. I’m also not looking to switch from my Mac to any Windows-based machinery, but instead, use it for one particular task when 3D modeling.

What I would somewhat agree with you is that going PC for 3D rendering firsthand would’ve been worthwhile if it was the only thing I did, but given it’s the smallest % of all things I do, it just doesn’t add up.

 

[deconstruct60]

Well that took long enough. Not sure how helpful this will be on the Mac side of things, unless they're willing to fund a total rewrite of the viewport from OpenGL to Metal, and possibly a full rewrite of Cycles into Metal as well.

part of the announcement by Blender folks is that Apple is helping to covert Cycles.

Apple Pledges Support for 3D Graphics Tool Blender

Blender today announced that Apple has joined the Blender Development Fund to support continued development of the free open source 3D graphics tool....

www.macrumors.com

which has a screen grab of a twitter posting with a link .

Cycles Metal device

**Tasks** For tracking Metal device development, targeting 3.1 release: - [x] Kernel changes required to compile for MSL (Metal Shading Language): - [x] Explicit address space qualifiers - [x] Adapt cross-platform layer (device/gpu) for MSL - [x] MetalRT kernel changes - [x] Host...

developer.blender.org

Not done. but Cycles is covered by the Apache 2.0 license and Apple could fold things in over time.

The viewport is going to be trickier if it is covered by GPL v3... which Apple likely won't touch with a ten foot pole.
Money contributed so someone else writes code would help over time.


P.S. this hints that Apple isn't looking to start signing 3rd party GPU drivers for macOS 12 on M-series any time over interim horizon. Apple brings their 16-32 GPU core offering to the table and points at this announcement. Not just Metal, but Apple GPU tweakedd over time also. If Apple doesn't enable any 3rd party drivers over the long tern Metal will largely be the Apple GPU.

 

[vel0city]

Wonder where this leaves Cycles 4D?

 

[LymeChips]

I’m not seeing how selling at a loss then spending extra on inferior consumer components of a PC tower serves your case. I’m also not looking to switch from my Mac to any Windows-based machinery, but instead, use it for one particular task when 3D modeling.

What I would somewhat agree with you is that going PC for 3D rendering firsthand would’ve been worthwhile if it was the only thing I did, but given it’s the smallest % of all things I do, it just doesn’t add up.

Hi Romanesco, I’m really interested in your experience with Octane X. I’ve been eyeing a MP myself especially with the new 6800 duo cards and the possible intel bump.

What DCC are you using? Cinema? If so what version?


I too am frustrated to see they’re not updating as frequently as RS… BUT….

In a Slack group I’m in Jules said this yesterday “PR12 is next week, same features as PR11, but tons of bug fixes, mac os 12 support and keeps GPU support working all the way back to 2012, When we move to Octane 2022 we will lose these older devices, but will gain a ton of new features (obviously Octane X 2022 includes all features in 2021.1) and adds new kernels and CUDA interop, plus more things we can share shortly specifically on Apple Silicon”


If all else fails I think dual booting into Windows sounds like a good idea you could even use RNDR to render final frames while you switch back to macOS for post.

 

[Romanesco]

Hi Romanesco, I’m really interested in your experience with Octane X. I’ve been eyeing a MP myself especially with the new 6800 duo cards and the possible intel bump.

What DCC are you using? Cinema? If so what version?


I too am frustrated to see they’re not updating as frequently as RS… BUT….

In a Slack group I’m in Jules said this yesterday “PR12 is next week, same features as PR11, but tons of bug fixes, mac os 12 support and keeps GPU support working all the way back to 2012, When we move to Octane 2022 we will lose these older devices, but will gain a ton of new features (obviously Octane X 2022 includes all features in 2021.1) and adds new kernels and CUDA interop, plus more things we can share shortly specifically on Apple Silicon”


If all else fails I think dual booting into Windows sounds like a good idea you could even use RNDR to render final frames while you switch back to macOS for post.

Hi @LymeChips,
I’m on C4D S24, and the amount of crashes experienced with PR11 is beyond acceptable for professional work. And that’s just another mishap in a long series of disappointments by both Apple and OTOY.

Octane X arrived considerably later than promised, with an underwhelming built that still feels more proof-of-concept than a usable release. And the iterative progress is underwhelming, to say the least.

I know which Slack you’re talking about, and in my experience, Jules tends to overpromise and underdeliver.

One of the reasons I went with the Mac Pro (2019) was the superior user experience of macOS, complemented by the seamless interoperability with the rest of the Apple products. Plus, I’m running one 6K XDR + 2x5K monitors. Has it delivered? Overall, yes, except when it comes to 3D rendering.

What I’d suggest is to wait till Monday and see what gets updated/ introduced. If it’s nothing groundbreaking, then it would be safe to assume no revamp of the 2019 model is expected until WWDC22.

If you have the budget, the latest (parts) maxed out Mac Pro (2019) will serve you well. But if your primary use case scenario for this machine is 3D modeling/ rendering, you should strongly consider a workstation-caliber PC tower. To be clear, not consumer, but workstation parts (as in not RTX 3090 but rather A6000).

I hope that helps.

 

[vinegarshots]

The 3090 performs at a similar level to the A6000 in benchmarks. If you don’t need the extra 24GB of video memory, a good 3090 can function just fine as a workstation card TBH.

 

[awkward_eagle]

Hi @LymeChips,
I’m on C4D S24, and the amount of crashes experienced with PR11 is beyond acceptable for professional work. And that’s just another mishap in a long series of disappointments by both Apple and OTOY.

Octane X arrived considerably later than promised, with an underwhelming built that still feels more proof-of-concept than a usable release. And the iterative progress is underwhelming, to say the least.

I know which Slack you’re talking about, and in my experience, Jules tends to overpromise and underdeliver.

One of the reasons I went with the Mac Pro (2019) was the superior user experience of macOS, complemented by the seamless interoperability with the rest of the Apple products. Plus, I’m running one 6K XDR + 2x5K monitors. Has it delivered? Overall, yes, except when it comes to 3D rendering.

What I’d suggest is to wait till Monday and see what gets updated/ introduced. If it’s nothing groundbreaking, then it would be safe to assume no revamp of the 2019 model is expected until WWDC22.

If you have the budget, the latest (parts) maxed out Mac Pro (2019) will serve you well. But if your primary use case scenario for this machine is 3D modeling/ rendering, you should strongly consider a workstation-caliber PC tower. To be clear, not consumer, but workstation parts (as in not RTX 3090 but rather A6000).

I hope that helps.

For what it's worth, I've been using 2x W6800x Duos in C4D r25 with Redshift and it's been a great experience so far. I typically use Arnold on both Mac and Windows and it's IPR is generally reliable but can crash on heavy scenes if its updating too fast.

Redshift has been solid so far and has feature parity with the Windows version which is a big plus. Was experimenting with OctaneX but stopped as soon as I saw there was no ACES support on the Mac yet. I liked the simplicity and "Arnoldness" of Octane, but you might want to give Redshift a try if you want to stay on the Mac until Octane sorts itself out.

 

[Romanesco]

For what it's worth, I've been using 2x W6800x Duos in C4D r25 with Redshift and it's been a great experience so far. I typically use Arnold on both Mac and Windows and it's IPR is generally reliable but can crash on heavy scenes if its updating too fast.

Redshift has been solid so far and has feature parity with the Windows version which is a big plus. Was experimenting with OctaneX but stopped as soon as I saw there was no ACES support on the Mac yet. I liked the simplicity and "Arnoldness" of Octane, but you might want to give Redshift a try if you want to stay on the Mac until Octane sorts itself out.

It’s encouraging to hear about the positive experience with Redshift. And yes, if you went with 2x W6800X Duos, you’re practically maxed out for 3D rendering. Cheers.

 

[shuto]

Would you mind helping me make up my mind?

I’m on a 16-core Mac Pro (2019) with 384 GB RAM and 1xVega II Duo. I am debating between adding a W6800X Duo or getting an RTX A6000 to use under Bootcamp/ Windows. I can’t get a hold of an RTX 3090 and have no intention of buying a tower PC.

I primarily care about the performance under Octane in C4D.

For my use case scenario, Octane X on Metal has been a gross failure. They’ve overpromised, underdelivered, and took way too long for a barely working release. Speedier or not, adding another MPX module won’t fix the constant crashes. On top of that, the developer is so slow on the progress that I can no longer rely on their word.

On the other hand, CUDA is more mature and would allow me to run deep learning algorithms outside my primary use case scenario.

In your opinion, should I go for:

• Option 1: Add the W6800X Duo to the Vega II Duo and continue struggling with the miserable Octane X experience.
• Option 2: Add the RTX A6000 and double boot to Windows for 3D rendering tasks/ CUDA.
• Option 3: Don’t add anything; wait for the next generation of NVIDIA cards/next-generation Mac Pro.
• Option 4: suggest your own.

Thanks!

thanks for sharing your octaneX experience here. I'm sorry it hasn't been a good one though! I am in a similar situation to lots of people here, in that I prefer to work on Mac, but PCs are much better for certain 3D tasks.

Personally I am on a MacBook Pro which I do 95% of my motion graphics work on. The other 5% is on a PC I built, which sucks, but also has octane on it which doesn't crash often.

Your plan of dual booting into windows might work, but personally I would prefer to have a PC as an extra rendering machine. I def don't think your option 1 makes sense.

 

[Romanesco]

thanks for sharing your octaneX experience here. I'm sorry it hasn't been a good one though! I am in a similar situation to lots of people here, in that I prefer to work on Mac, but PCs are much better for certain 3D tasks.

Personally I am on a MacBook Pro which I do 95% of my motion graphics work on. The other 5% is on a PC I built, which sucks, but also has octane on it which doesn't crash often.

Your plan of dual booting into windows might work, but personally I would prefer to have a PC as an extra rendering machine. I def don't think your option 1 makes sense.

Thanks for the advice and your positive outlook, @shuto. I’m now leaning towards your suggestions myself.

 

[shuto]

Thanks for the advice and your positive outlook, @shuto. I’m now leaning towards your suggestions myself.

is it possible for you to switch your renderer to Redshift? Sounds like its working well for @awkward_eagle. I like octane as well, but sounds like redshift is a lot more reliable, especially on a Mac!

 

[Romanesco]

is it possible for you to switch your renderer to Redshift? Sounds like its working well for @awkward_eagle. I like octane as well, but sounds like redshift is a lot more reliable, especially on a Mac!

It would be another learning curve, but doable. I’m also curious to see the benchmarks of the new M1 Pro/ M1 Max and how that unified memory compares to the Mac Pro (2019) options.

 

[vel0city]

Just ordered a 64GB/2TB 16" Max so will report back on Redshift and Octane running in C4D as soon as it arrives - mid November delivery date for the UK.

 

[deconstruct60]

It would be another learning curve, but doable. I’m also curious to see the benchmarks of the new M1 Pro/ M1 Max and how that unified memory compares to the Mac Pro (2019) options.

apple has some narrow benchmarks on their marketing page versus a i9 , 5600M with HBM memory.

MacBook Pro

MacBook Pro with M3, M3 Pro, and M3 Max chips. Up to 22 hours of battery life. The world’s best laptop display. Now in a new color: Space Black.

www.apple.com

some CPU ones:

Vectorworks (publishing) 1.7x and 1.7x
NASA CFD 3x and 3x

Some of the GPU ones:

Maxon Cinema 4D Redshift 2.5X and 4x
Da Vinci Resolve render 1.4x and 1.9x
Final Cut Pro 8K render 1.7x and 2.9x


Apple is going to cherry pick what they do best though. But if can hit their Accelearte framework sweet spot ( e.g., the NASA code ) there is a serious CPU bump. For the GPU, the dGPU era in Mac laptops is over.

 

[sirio76]

It would be another learning curve, but doable. I’m also curious to see the benchmarks of the new M1 Pro/ M1 Max and how that unified memory compares to the Mac Pro (2019) options.

According to Apple, Redshift benchmark is 4x faster than the Radeon 5600M inside the old 16" Intel MBP.
I don't know if Redshift scale like Octane but if we look at this chart a 4x increase over the 5600M will put the M1 Max GPU above the 6900XT.
Of course is better to wait for some more benchmark or even better real world tests.

 

[Romanesco]

According to Apple, Redshift benchmark is 4x faster than the Radeon 5600M inside the old 16" Intel MBP.
I don't know if Redshift scale like Octane but if we look at this chart a 4x increase over the 5600M will put the M1 Max GPU above the 6900XT.
Of course is better to wait for some more benchmark or even better real world tests.

If that turns out to be accurate, then GPU rendering on Mac is a viable reality again.

First Geekbench scores for the M1‌ Max chip with 10-core CPU and 32-core GPU are stating a single-core score of 1749 and a multi-core score of 11542, but no sign of the Metal/ compute performance. I’m waiting on more reports.

First Geekbench Score Surfaces for MacBook Pro M1 Max With 2x Faster Multi-Core Performance Compared to M1

Just after Apple's event introducing the new MacBook Pro models with M1 Pro and M1 Max chips, the first benchmark for the high-end M1 Max chip...

www.macrumors.com

 

[deconstruct60]

According to Apple, Redshift benchmark is 4x faster than the Radeon 5600M inside the old 16" Intel MBP.

Apple isn't quite doing Apples to Apples there.

footnote 15
"...

1. 
   Testing conducted by Apple in September 2021 using preproduction 16-inch MacBook Pro systems with Apple M1 Max, 10-core CPU, 32-core GPU, and 64GB of RAM, and preproduction 16-inch MacBook Pro systems with Apple M1 Pro, 10-core CPU, 16-core GPU, and 32GB of RAM, as well as production 2.4GHz 8-core Intel Core i9-based 16-inch MacBook Pro systems with Radeon Pro 5600M graphics with 8GB of HBM2 and 64GB of RAM, all configured with 8TB SSD. Prerelease Cinema 4D S25 and prerelease Redshift v3.0.54 tested using a 1.32GB scene. Performance tests are conducted using specific computer systems and reflect the approximate performance of MacBook Pro.

..."

This is redshift stuff can't even get on open market versus the released "historical" scores.