The New Mac Pro is Here!!!

[andrewhake]

Sorry, no, I meant the monitor stand/base sitting at keyboard height. That typically puts the center of the display at or just below eye height for most people, which is where it should be. Problem is that Apple has never incorporated vertical height adjustment into their monitor bases/stands. OK, that's not true, the last and only one to ever include any sort of height adjustment was the first LCD iMac which had the CPU in a base and the display on an articulating arm.

So I typically take my iMacs or Mac Monitors and pull the stand off and use a VESA mount to an arm or other more adjustable system. So I'm glad to see this new pro monitor offer the base separately. It looks like a nice base design, but still needs more vertical adjustment, IMO. Anyway, I'll have to wait and see what it's like.

https://www.apple.com/pro-display-xdr/specs/

The adjustment range with the stand is quite generous. At it's standard height it's much higher than an iMac display midpoint, and at it's highest position, it puts the top of the display higher than an iMac on 10cm high stand.

 

[andrewhake]

I would love to see Apple put some $ into coding resources to facilitate Metal optimization - if for no other reason than to provide competition for nVidia/CUDA.

Cheers - #19

This is exactly what is happening with all the developers mentioned yesterday during the presentation.

 

[andrewhake]

I really think it will be about balancing between GPU options and CPU cores. Based on the specs we have, I would advise people away from the 8-core model. The Xeon W series basically comes in two levels and the 8-core is the top of the lower level. You get slower memory speeds and it does not support the same memory configurations and capacities. There are other differences, but that is the big one. 8-core is still an option for an entry-level configuration or if you need to load up with quad GPUs while CPU and main RAM is not your primary concern.

Without pricing, it will be hard to identify the "sweet spot" but it looks like I'd recommend the 12-core as the starting point and then flesh our your RAM, GPUs and storage as needed. Then, if you still have money left in the budget and can use more cores, spend more money to get them. You will need the top two Xeon CPU options (24 or 28 cores) in order to utilize all the PCIe lanes and features. And yet the 12 thru 28 core options all have a top turbo speed of 4.4GHz and while base clock does drop with more cores, that represents the speed at which it can operate and sustain all cores. In a nutshell, the 12-core can operate all cores at 3.3GHz and burst two cores at up to 4.4GHz while it scales to different speeds in between, depending on how many cores/threads are in use and at what sort of loads. Likewise, the 28-core operates all at 2.5GHz, but can probably still sustain 12 cores at close to 3.3GHz of the 12-core model.

If I had to guess as to which will be the sweet spot for many, I would say the 16-core system. Premiere, Resolve and many others can use 20~30 threads effectively at this point, but the benefit really falls off after that and we also have to take GPUs into consideration as well. I've built some systems with 32+ cores and it actually becomes a thing trying to feed all those cores, all the memory you need, trying to supply adequate GPU resources to support that many cores and yet the clock rates fall with that many cores, so on the flip-side it's hard for lots of cores running at only 2GHz or so to feed a few monster GPUs. IMO, if you need that much processing power, it's not necessarily best to put it all in one box.

Definitely agreed. It has been nice to see that with many of the Intel Xeon offerings recently, as core count goes, up you don't necessarily lose the benefits that the lower core count CPUs offer. If you have mostly highly multi-threaded tasks and also single threaded tasks in your workflow, the higher core count CPUs don't actually suffer that much to their lower core counterparts anymore. So it really becomes more a question of cost. The listed base clocks are the minimum that the CPU can sustain on all cores, not necessarily the speeds it will run at. the Xeon-W 18 core in an iMac pro has a 2.3Ghz base clock but in my experience when using all cored for 3D rendering, it runs at 2.8Ghz-3Ghz.

 

[Blair S. Paulsen]

This is exactly what is happening with all the developers mentioned yesterday during the presentation.

nVidia put resources into not just creating the CUDA hardware, but supporting the infrastructure for coding. If Apple is willing to make a similar investment in fostering a strong development environment for Metal, I'm much more optimistic about getting the performance I'd be paying for on the hardware side. Props to RED for working with Apple to support R3D file handling in Metal, and, by extension, AMD silicon. I'd be curious to know if the new code will work only with GPUs on the PCIe bus, or also in the Afterburner card...

Cheers - #19

 

[Jeff Kilgroe]

What would be the benefit of NVidia GPU's over the GPU's Apple will offer in the MacPro's?

More options...

But yes, Nvidia would need to support Metal -- which is a requirement and would be needed for them to provide proper driver support. We would of course want them to optimize it well and fully back the Metal API with their hardware.

As for CUDA, having it available in updated/ current form on the Mac would make it easier for developers to port existing CUDA applications over to the Mac or updated them on the Mac since it's been a while since CUDA has seen a proper update. Although I would think most developers would choose to support Metal as it's also very capable, more so than OpenCL, and would ensure full reach across the Mac platforms.

 

[L. Langer]

I'd be curious to know if the new code will work only with GPUs on the PCIe bus, or also in the Afterburner card...

Cheers - #19

Apple's wording about it is equally odd and confusing. They specifically ONLY state ProRes and ProRes RAW in the specs - and various press write-ups reflect that - and call it a "ProRes Accelerator Card", yet Apple's own press release also makes a point of saying "use native file formats without conversion" as if to reference R3D and others. As it is programmable, it makes sense to use it as an accelerator for R3D and others, as well as Apple's ProRes formats.

 

[DJ Meyer]

I thought the whole point of ProRes Raw was to not need proprietary accelerator cards. :biggrin5:

 

[Misha Engel]

More options...

But yes, Nvidia would need to support Metal -- which is a requirement and would be needed for them to provide proper driver support. We would of course want them to optimize it well and fully back the Metal API with their hardware.

As for CUDA, having it available in updated/ current form on the Mac would make it easier for developers to port existing CUDA applications over to the Mac or updated them on the Mac since it's been a while since CUDA has seen a proper update. Although I would think most developers would choose to support Metal as it's also very capable, more so than OpenCL, and would ensure full reach across the Mac platforms.

Apple is doing this https://www.apple.com/newsroom/2019/06/pro-app-developers-react-to-the-new-mac-pro-and-pro-display-xdr/ all the options they need. When they need more options they just sponsor the developers to make it (same as NVidia did with CUDA, nothing new).

 

[Misha Engel]

I thought the whole point of ProRes Raw was to not need proprietary accelerator cards. :biggrin5:

I think you mistake ProRes(Raw) with Cineform(RAW), 3 streams of 8k CineformRAW with the same bells and whistles and quality will run without an accelerator or GPU-decoding on a Xeon W-3275 (you still need the GPU's for de-bayering, OFX, NR, etc...).

 

[andrewhake]

nVidia put resources into not just creating the CUDA hardware, but supporting the infrastructure for coding. If Apple is willing to make a similar investment in fostering a strong development environment for Metal, I'm much more optimistic about getting the performance I'd be paying for on the hardware side. Props to RED for working with Apple to support R3D file handling in Metal, and, by extension, AMD silicon. I'd be curious to know if the new code will work only with GPUs on the PCIe bus, or also in the Afterburner card...

Cheers - #19

https://home.otoy.com/octane-x-wwdc2019/

Apple has been working with developers directly for a very long time already.

 

[Jarek Zabczynski]

I think right now, before knowing full specs and pricing, I’d be looking at this configuration.

16 core Xeon Processor
96GB RAM (aftermarket)
1x Vega II Dual MPX Module
Afterburner card
2x1TB SSD

Plus adding:
Pegasus dual 3.5mm drive cage, probably SSDs for application scratch.
Blackmagic Decklink card

Nice to know I can start with one MPX module and add another or upgrade. Back to the good ol days!
If we get Nvidia support I can see myself going with 2-4 2080ti cards.

 

[Bruce Allen]

Hmm, this is a kinda cool but also disappointing update.

The individual components in the $5999 machine are pretty lousy so I guess what you're paying for is all the development costs of that proprietary MPX setup.

As someone who has run machines with many GPUs, I appreciate the cooling setup.

I appreciate the power setup less. I guess it's cool that you don't have to plug in external power cables to the GPU (it's powered through a proprietary slot design) but it's also annoying because now there are no power connectors if you need 'em. Apple just cleverly locked you out of plugging a normal (ie good value-for-money) GPU in there and charged a lot of money for the privilege. Uh, OK.

If the GPUs in these MPX modules were super awesome then this would be a pretty nifty upgrade. But they're not. They're really close to the Instinct MI60 and Radeon VII... which are like, kinda mediocre. Not sure I'd want 4 mediocre GPUs. But anyway...

If they update this in a year's time with nVidia cards (or AMD steps up with Navi, or heck, Intel makes an awesome GPU) and Apple makes new MPX modules that don't cost a ton, maybe it'll make sense to buy one.

Just not when these current AMD chips which are pretty out of date in terms of architecture (they're still GCN) are the only option - I'm not too interested in buying 4x kinda-obsolete and inefficient GPUs glued together in an extremely expensive proprietary config.

I kinda suspect that a 32-core AMD system with two nVidia Titan RTX cards might just beat it and be MUCH less expensive.

But for the future in 2020 or 2021 with some actual current-gen cards from nVidia or AMD or even Intel it'd be pretty cool. Also I'd guess Intel will be forced to respond to AMD so you might get cheaper faster better CPU options too.

Of course those will have PCIe 4.0 haha.

This thing doesn't support PCIe 4.0 does it?

There you go - I found out how Apple will get all the people who bought the 1st gen new Mac Pro to upgrade! PCIe 4.0!

AMD with somewhere between 32 and 64 cores (I will evaluate price / performance when it's out) plus PCIe 4.0 and nVIDIA will probably be the way for me next year, but I'm glad the Mac Pro is out there. Missed the mark for but that's cool, and they could still be a contender if they do a good update next year.

Bruce Allen
www.bruceallen.tv

 

[Jarek Zabczynski]

There are power connections available inside for standard cards.
I do hope they bring Nvidia support. Technically nothing stopping them.

 

[Chad Lancaster]

 

[Bruce Allen]

There are power connections available inside for standard cards.
I do hope they bring Nvidia support. Technically nothing stopping them.

Oh nice! Yeah if they bring nVIDIA support (or AMD or Intel get more competitive) this could be interesting.

No interest in overpaying for that 4x-medicore-AMD GPU setup but that can be fixed in a later update.

I wonder how much the 6K display will be when Apple starts having refurb models BTW?

Bruce Allen
www.bruceallen.tv

 

[Misha Engel]

Apple won't bring NVidia support, when AMD goes bankrupt, Apple will build a GPU themselves, buy a third party GPU or go bankrupt themselves, they rather die than use NVidia.
Those Vega 2's are competetive with the quadro GV100 when looking at the same power usage, compute perfomance, memory and memory bandwidth.

Vega 2 base clock 1400 MHz, 4096 cores ~ 11.47 fp32 Tflops and 1,024 GB/s memory bandwidth, 32 GB HBM2 $ ???
GV100 base clock 1132 MHz, 5120 cores ~ 11.59 fp32 Tflops and 868.4 GB/s memory bandwidth, 32 GB HBM2 $9,400 https://www.amazon.com/NVIDIA-Quadro-GV100-Volta-Graphics/dp/B07JBQ4DBV/ref=sr_1_1?crid=32OFNQCM8R50E&keywords=nvidia+gv100&qid=1559758706&s=gateway&sprefix=nvidia+GV%2Caps%2C233&sr=8-1

RTX Quadro's have a medio core memory bandwidth upto only 672 GB/s and high latencies compared to HBM2 because of GDDR6, hence the crappy performance with Noise Reduction in combination with 8k.R3D.
They are very good for playing Battlefield, etc.. with all those pixel shaders.

 

[DJ Meyer]

Apple won't bring NVidia support, when AMD goes bankrupt, Apple will build a GPU themselves, buy a third party GPU or go bankrupt themselves, they rather die than use NVidia.
Those Vega 2's are competetive with the quadro GV100 when looking at the same power usage, compute perfomance, memory and memory bandwidth.

Vega 2 base clock 1400 MHz, 4096 cores ~ 11.47 fp32 Tflops and 1,024 GB/s memory bandwidth, 32 GB HBM2 $ ???
GV100 base clock 1132 MHz, 5120 cores ~ 11.59 fp32 Tflops and 868.4 GB/s memory bandwidth, 32 GB HBM2 $9,400 https://www.amazon.com/NVIDIA-Quadro-GV100-Volta-Graphics/dp/B07JBQ4DBV/ref=sr_1_1?crid=32OFNQCM8R50E&keywords=nvidia+gv100&qid=1559758706&s=gateway&sprefix=nvidia+GV%2Caps%2C233&sr=8-1

RTX Quadro's have a medio core memory bandwidth upto only 672 GB/s and high latencies compared to HBM2 because of GDDR6, hence the crappy performance with Noise Reduction in combination with 8k.R3D.
They are very good for playing Battlefield, etc.. with all those pixel shaders.

Misha your paper benchmarks constantly, relentlessly telling us how AMD cards can keep up with nVidia's best in resolve always makes me wonder why they do so badly then in other applications.

Perhaps with Apple's money behind Metal they will be able to pull even in more areas but I find it hard to believe they'll make up the gap in something like 3d rendering. RTX tech in Octane and Redshift is showing enourmous gains.

 

[Bruce Allen]

Apple won't bring NVidia support, when AMD goes bankrupt, Apple will build a GPU themselves, buy a third party GPU or go bankrupt themselves, they rather die than use NVidia.
Those Vega 2's are competetive with the quadro GV100 when looking at the same power usage, compute perfomance, memory and memory bandwidth.

Vega 2 base clock 1400 MHz, 4096 cores ~ 11.47 fp32 Tflops and 1,024 GB/s memory bandwidth, 32 GB HBM2 $ ???
GV100 base clock 1132 MHz, 5120 cores ~ 11.59 fp32 Tflops and 868.4 GB/s memory bandwidth, 32 GB HBM2 $9,400 https://www.amazon.com/NVIDIA-Quadro-GV100-Volta-Graphics/dp/B07JBQ4DBV/ref=sr_1_1?crid=32OFNQCM8R50E&keywords=nvidia+gv100&qid=1559758706&s=gateway&sprefix=nvidia+GV%2Caps%2C233&sr=8-1

RTX Quadro's have a medio core memory bandwidth upto only 672 GB/s and high latencies compared to HBM2 because of GDDR6, hence the crappy performance with Noise Reduction in combination with 8k.R3D.
They are very good for playing Battlefield, etc.. with all those pixel shaders.

"Compute performance" is not a simple number like that at all. Well, maybe if all you're doing is just simple denoising with Neat Video on 8K footage all day.

And those pixel shaders you deride as being good for games are also good for things like rendering using Redshift, V-Ray GPU and Arnold GPU. Coupled with RTX for ray intersection calculations. Some things need pixel shaders, some need ray intersections, others need integer performance! Also deep learning - there is so much coming in terms of AI techniques for video.

Coders depend on an ecosystem to wring the best out of the hardware. Right now, CUDA is massively, massively ahead in this.

Don't get me wrong - I want competition and prefer open standards, but NVIDIA is way ahead, currently.

So right now I'm just saying buying 4x of those AMD GPUs is a massive waste of money unless you've researched and found they are competitive in those specific tasks you need (eg Neat Video).

I personally bet that in other things (deep learning, Redshift etc) they will get crushed by much cheaper systems. And in either event, for bandwidth surely you want PCIe 4.0 for future proofing? I get that most tasks are not bandwidth-constrained but it sucks this Mac doesn't support PCIe 4.0 whereas the new AMD Ryzen systems do.

Bruce Allen
www.bruceallen.tv

 

[Andy_Johnson]

The individual components in the $5999 machine are pretty lousy so I guess what you're paying for is all the development costs of that proprietary MPX setup.

Everyone has their hardware opinions, but even if you aren't an Apple fan, you have to admit that this was a *giant* step in the right direction for professionals. I'm not sure where you get the "lousy" specs from but there are some great videos from Linus Tech Tips on the Mac Pro and even he's mostly impressed.

 

[Jarek Zabczynski]

I seriously don’t get the bitching about the price. The Mac Pro was never for the average user. If it’s too expensive for you, odds are you really don’t need one.

I look at it this way. I anticipate spending about 10k on mine. For something that I will use almost every day for a good 5 years to make money, that cost is pretty irrelevant. My current system is going on 7 years now. The price doesn’t phase me.