The New Mac Pro is Here!!!

[Bruce Allen]

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.

Oh yeah, totally agree. Also folks don't take software license costs into account - the hardware is often less than that cost per machine! For Avengers: Endgame I got us to upgrade to 48-core systems for the render farm because you're paying Arnold licenses per unit, not per core...

That said, the $6000 option is still kinda a lousy deal if it will be slower than a $4000 PC for most professional tasks... and seriously a 0.25TB SSD - it's the 2019 equivalent of one of those little 4 megabyte compact flash cards that Canon would bundle with their cameras that you throw away instantly haha.

RE: keeping stuff for many years - absolutely. It is awkward timing since PCIe 4.0 is here (have you SEEN the speeds those PCIe 4.0 SSDs are getting - 5GB/sec already!! that's what you want for 8K...) - there's no way I'm spending $10k on a system that won't let me plug in one of those suckers.

Bruce Allen
www.bruceallen.tv

 

[Jarek Zabczynski]

True, but speed isn’t much an issue with r3d. You can edit 8k r3d off single 7200rpm drives if you had to. All my scratch go on an external nas so fast internal ssds were never a huge deal. Yeah you want a fast boot an another internal ssd for application scratch, but everything else is external.

We get 2 internal ssd options on this Mac Pro. And Pegasus already announced a sata drive cage for above the pci slots inside.

 

[andrewhake]

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.

For those concerned about price I would recommend spec-ing out a similar HP Z8 workstation. Just moving to an 8 core Xeon with 3.5Ghz base clock will put you well above $7000.

 

[Brian Timmons]

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..

I don't doubt the friction between Apple and Nvidia but I'm not sure I would put it on the same scale as what happened between Apple and Qualcomm. Consider that Apple put aside their gripes with Qualcomm (including coughing up several billion $$$$) to continue doing business with them as it would effect their bottom line not to.

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 mostly agree. These systems offer a lot of built in power and a considerable amount of headroom for more.
I would even say they surpass some of the most ambitious wish lists I remember reading.
This was meant for a certain type of power user that has clients that can cover the costs of this hardware.
The power requirements alone should make one pause as this system and a 6K monitor or two might be enough to start tripping circuit breakers.

It should be noted how in the last 5 years the iOS/Mac platform now has addressed pro needs in various price points, devices, shapes, sizes, expansion options, etc. The latest iMac/iMac Pros is likely more than enough for many workflows
and that can be expanded with TB3.

My only issue is the "Apple tax" does seem to be a bit higher in the last few releases and they might be careful with pricing themselves out of certain mid range to lower in price points.

There's always room for improvement (Nvidia, perhaps an even lower cost version of the Mac Pro) but I truly feel like they are effectively addressing the wide ranging needs of the Pro community.

Brian Timmons
BRITIM/MEDIA

 

[andrewhake]

#485: WWDC and Interview with the Mac Pro Product Manager

https://www.relay.fm/mpu/485

 

[Brian Timmons]

#485: WWDC and Interview with the Mac Pro Product Manager

https://www.relay.fm/mpu/485

Great find and VERY informative. Thanks.

Brian Timmons
BRITIM/MEDIA

 

[Bruce Allen]

True, but speed isn’t much an issue with r3d. You can edit 8k r3d off single 7200rpm drives if you had to. All my scratch go on an external nas so fast internal ssds were never a huge deal. Yeah you want a fast boot an another internal ssd for application scratch, but everything else is external.

We get 2 internal ssd options on this Mac Pro. And Pegasus already announced a sata drive cage for above the pci slots inside.

I'm guessing you don't deal a lot with 4K VFX shots! Burning through multi-layer 4K and above .exr files or uncompressed DPX files.

When VFX goes to 8K, that's what you'll want a nice PCIe 4.0 SSD for! Just sucks that this Mac Pro won't allow for that.

For those concerned about price I would recommend spec-ing out a similar HP Z8 workstation. Just moving to an 8 core Xeon with 3.5Ghz base clock will put you well above $7000.

Yeah but HP's prices are stupid too.

Puget Systems have good support and aren't cheap but they're $1500 less than that $6000 Mac.

An 8-core Xeon with 4.3ghz base, 256GB SSD, 32GB RAM, 10G ethernet, and an RTX 2070 is $4500:
https://www.pugetsystems.com/nav/peak/tower_single/customize.php

Or more relevantly, here's an 8-core 3.6ghz system, 32GB RAM, 1TB SSD, RTX 2080 for $2780:
https://www.amazon.com/Dell-8930-Special-Tower-Desktop/dp/B07QLNSXV8

Or really this $2000 system with a 3.6ghz 8-core, 32GB RAM and a 512GB SSD and a 1080:
https://www.amazon.com/dp/B06WLMY9N4

Now that's 1/3rd the price of that Mac Pro. And probably both that system and the $6000 Mac Pro will be similarly useless because they'll almost immediately bottleneck on the RAM and lack of SSD space. At least the $2000 system would bottleneck later because at least it has 512GB not 256GB haha!

I know, I know, the last two aren't Xeons and it don't have ECC RAM. But guess what? In our industry, you don't need a Xeon with ECC RAM.

This is speaking as a guy who personally owns a 40-core dual Xeon with 384GB ECC RAM. It's great for Houdini sims. ECC RAM still isn't necessary but Xeon is necessary to address that quantity of RAM.

For editing and grading, my 4ghz 8-core standard Core i7 non-Xeon with 128GB of RAM, fast NVIDIA GPUs and many TBs of M.2 storage is a better choice.

Bruce Allen
www.bruceallen.tv

 

[Jarek Zabczynski]

Yeah you’re right. I do sometimes move some stuff to my scratch ssd for AE work.

 

[Misha Engel]

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.

Sponsering developers who don't find pride, added value, or whatever in developing hardware acnostic software.
Most if not all API's are written in a C++, the fastest way to a quick result with GPU API's is/was CUDA, Apple put some sponsor money in Metal and now lots of them are embracing metal.

Few companies take an extra step and use C++ and there are even some who go as far as assembly and make their software hardware and OS acnostic, Blackmagic Design is a good example of one of the few.

FP32 is FP32, memory bandwidth is memory bandwidth no matter what brand you use.

 

[Bruce Allen]

Sponsering developers who don't find pride, added value, or whatever in developing hardware acnostic software.
Most if not all API's are written in a C++, the fastest way to a quick result with GPU API's is/was CUDA, Apple put some sponsor money in Metal and now lots of them are embracing metal.

Few companies take an extra step and use C++ and there are even some who go as far as assembly and make their software hardware and OS acnostic, Blackmagic Design is a good example of one of the few.

FP32 is FP32, memory bandwidth is memory bandwidth no matter what brand you use.

Yeah but that's like saying horsepower is horsepower. Yes, it does have an effect on how fast the car will go... but isn't a perfect way to predict track times!

You'll find that real-world use depends on more than just those dumb numbers, unless you're doing very simple cases.

Intricacies RE how the cache is setup, how the scheduler works, how the pipeline works, how the compilers are built, how well the data structures fit the hardware, throughput vs latency, etc.

It's not trivial - differences here can account for MASSIVE (10x) speedups or slowdowns. This is where NVIDIA has a huge real-world advantage currently - and this won't go away in just a year.

Also, if I were a developer, would I really want to spend a lot of time and money optimizing stuff for this odd case of 4x GCN chips (the core in these Mac Pro GPUs) when AMD has already announced Navi?

Bruce Allen
www.bruceallen.tv

 

[Misha Engel]

Yeah but that's like saying horsepower is horsepower. Yes, it does have an effect on how fast the car will go... but isn't a perfect way to predict track times!

You'll find that real-world use depends on more than just those dumb numbers, unless you're doing very simple cases.

Intricacies RE how the cache is setup, how the scheduler works, how the pipeline works, how the compilers are built, how well the data structures fit the hardware, throughput vs latency, etc.

It's not trivial - differences here can account for MASSIVE (10x) speedups or slowdowns. This is where NVIDIA has a huge real-world advantage currently - and this won't go away in just a year.

Also, if I were a developer, would I really want to spend a lot of time and money optimizing stuff for this odd case of 4x GCN chips (the core in these Mac Pro GPUs) when AMD has already announced Navi?

Bruce Allen
www.bruceallen.tv

Blackmagic Design has proven that it can be done (the combination of FP32 TFlops, memory bandwidth and memory latency is what matters, the programmers make the difference).

Navi is focussed on gaming, like geforce.

RTX Quadro is expensive RTX geforce with (more) ECC memory and driver certification (launch price RTX Quadro 8000 $10,000 selling price today $5,600, https://www.anandtech.com/show/13217/nvidia-announces-turing-powered-quadro-rtx-family). GV100 still sells for $9k+

AMD VEGA and NVidia VOLTA are focussed on compute.

The developers get paid by apple to spend time optimizing stuff, these cost for apple are transfered to the Apple buyers (same as NVidia does and AMD is starting to do) and when they don't cover all the cost they can atleast have fun in Cupertino HQ forcing NVidia to lower their prices.

 

[Bruce Allen]

Blackmagic Design has proven that it can be done (the combination of FP32 TFlops, memory bandwidth and memory latency is what matters, the programmers make the difference).

In that specific SIMPLE case that suits AMD. I actually wrote a paint program once and wrote color correction, blurs, etc. Try something more complex, like facial recognition or machine learning denoising and you'll see very different performance and NVIDIA will quite likely pull far ahead.

I actually have a degree in computer science and have written GPU accelerated code so please trust me I know what I'm talking about.

RTX Quadro is expensive RTX geforce with (more) ECC memory and driver certification (launch price RTX Quadro 8000 $10,000 selling price today $5,600, https://www.anandtech.com/show/13217/nvidia-announces-turing-powered-quadro-rtx-family). GV100 still sells for $9k+

You don't need a GV100 or a Quadro 8000. Not sure what you're talking about.

Navi is focussed on gaming, like geforce.

No, that is a stupid statement. They will very, VERY likely make professional and compute derivatives that use Navi.

AMD VEGA and NVidia VOLTA are focussed on compute.

That's just how they brand it - the fundamental architecture of the GPU processor in VEGA is the same between RADEON VII and their "compute" INSTINCT MI60. It's all GCN.

You're just reading a bunch of marketing stuff and believing it and repeating it without actually understanding it.

Bruce Allen
www.bruceallen.tv

 

[David Collard]

That's just how they brand it - the fundamental architecture of the GPU processor in VEGA is the same between RADEON VII and their "compute" INSTINCT MI60. It's all GCN.

You're just reading a bunch of marketing stuff and believing it and repeating it without actually understanding it.

Thanks, Bruce.

 

[Misha Engel]

In that specific SIMPLE case that suits AMD. I actually wrote a paint program once and wrote color correction, blurs, etc. Try something more complex, like facial recognition or machine learning denoising and you'll see very different performance and NVIDIA will quite likely pull far ahead.

I actually have a degree in computer science and have written GPU accelerated code so please trust me I know what I'm talking about.

With what kind of software did your write that program, was is CPU or GPU focussed.
Facial recognition, machine learning, denoising will show that a $700 AMD card can be faster than a $ 2500 NVidia card, Blackmagic Design just showed it with DR16.
8k.R3D, TNR + 3 OpenFX effects in DR15 on a Radeon VII is upto 40(..50)% faster than with a RTX Titan X or RTX Quadro 8000 (the RTX has the FP32, but lacks the memory bandwidth and low latencies).

You don't need a GV100 or a Quadro 8000. Not sure what you're talking about.

When you want fast memory and low latency you need GV100 or VEGA2, when you need a lot of memory for VFX and you don't have HBCC, you need GPU's with a lot of VRAM, RTX Quadro 8000 has 48 GB of VRAM.
VEGA 2 has only upto 32 GB but can swap to system memory.

No, that is a stupid statement. They will very, VERY likely make professional and compute derivatives that use Navi.

Currently VEGA is focussed on compute and NAVI on gaming, not my statement but straight from AMD's CEO Lisa SU, no statements about next gen.

That's just how they brand it - the fundamental architecture of the GPU processor in VEGA is the same between RADEON VII and their "compute" INSTINCT MI60. It's all GCN.

You're just reading a bunch of marketing stuff and believing it and repeating it without actually understanding it.

Bruce Allen
www.bruceallen.tv

Radeon VII is an instinct MI50 with video outputs, some vans and some options turned of. Radeon VII is a good compute card and a mediocore gaming card (Price/perf).
Apple read the same marketing stuff and decided to use VEGA GPU's for most of their systems (with some polaris chip being the exception).

 

[Jeff Kilgroe]

One thing I want to bring up here is Metal in relation to CUDA. Apple did phenomenal work with Metal 2 and I truly can't think of anything that can be done with CUDA that can not also be done with Metal. Adapting from one to the other is actually not bad at all, it just takes a developer fluent in both APIs. Some may consider it heresy, but there are even some things about Metal that I think are more powerful or capable than counterpart operations in CUDA. I have not done that much with CUDA in the past year, but I've been diving pretty deep into Metal.

The ultimate downfall of CUDA will be its proprietary nature, unless Nvidia chooses to open it up. The industry wanted OpenCL to be the open standard, but it never could mature to the point of keeping step with CUDA. Metal is great in that it's hardware agnostic, yet still proprietarily Apple. Microsoft is also trying to pull the Windows world away from CUDA for compute and over to their new compute libraries/APIs in DirectX. In the end I think we're going to see DirectX on Windows and Metal on Mac/iPad. I still fully expect Apple to be making their own GPUs at some point in the next few years. Nvidia offering GPUs for the new Mac Pro could make sense in the short term in regards to CUDA support, but long term it would only make sense if they could offer hardware with a price and performance advantage. Given that CUDA has not been kept up on the Mac and many apps on Mac have had to rely on OpenCL or have transitioned to Metal, I'm not so sure that developers would put the effort into CUDA on the Mac platform. Especially since it would be a third party GPU solution and not something Apple would offer directly with a system purchase. II don't know if Apple would offer an Nvidia GPU as a BTO option... they probably would if Nvidia chose to offer one. The two companies are not getting along due to other reasons and competition for IP and autonomy tech. So who knows....

The VEGA GPUs spec'd for the new Mac Pro look pretty good. I'm sure they will be expensive, but probably (hopefully) not Quadro expensive.

 

[Jarek Zabczynski]

What can we guess one Vega II Dual MPX module will cost?

 

[Mike Halper]

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.

For those concerned about price I would recommend spec-ing out a similar HP Z8 workstation. Just moving to an 8 core Xeon with 3.5Ghz base clock will put you well above $7000.

I'm guessing you don't deal a lot with 4K VFX shots! Burning through multi-layer 4K and above .exr files or uncompressed DPX files.

When VFX goes to 8K, that's what you'll want a nice PCIe 4.0 SSD for! Just sucks that this Mac Pro won't allow for that.



Yeah but HP's prices are stupid too.

Puget Systems have good support and aren't cheap but they're $1500 less than that $6000 Mac.

An 8-core Xeon with 4.3ghz base, 256GB SSD, 32GB RAM, 10G ethernet, and an RTX 2070 is $4500:
https://www.pugetsystems.com/nav/peak/tower_single/customize.php

Or more relevantly, here's an 8-core 3.6ghz system, 32GB RAM, 1TB SSD, RTX 2080 for $2780:
https://www.amazon.com/Dell-8930-Special-Tower-Desktop/dp/B07QLNSXV8

Or really this $2000 system with a 3.6ghz 8-core, 32GB RAM and a 512GB SSD and a 1080:
https://www.amazon.com/dp/B06WLMY9N4

Now that's 1/3rd the price of that Mac Pro. And probably both that system and the $6000 Mac Pro will be similarly useless because they'll almost immediately bottleneck on the RAM and lack of SSD space. At least the $2000 system would bottleneck later because at least it has 512GB not 256GB haha!

I know, I know, the last two aren't Xeons and it don't have ECC RAM. But guess what? In our industry, you don't need a Xeon with ECC RAM.

This is speaking as a guy who personally owns a 40-core dual Xeon with 384GB ECC RAM. It's great for Houdini sims. ECC RAM still isn't necessary but Xeon is necessary to address that quantity of RAM.

For editing and grading, my 4ghz 8-core standard Core i7 non-Xeon with 128GB of RAM, fast NVIDIA GPUs and many TBs of M.2 storage is a better choice.

Bruce Allen
www.bruceallen.tv

I completely agree with Bruce. You could easily build a $2000 computer that you can safely overclock and have similar performance to the $6000 entry level Mac Pro. In the past the entry level Mac Pro was always $3000 or less. I understand that Apple’s hardware is built better than most others, and I’m willing to pay extra for that. But not $4000 extra. This entry Mac Pro should be $3000-$3500.


Xeon and ECC memory is a complete waste for most of our industry. Nothing needs it. Regular desktop CPUs would be fine. Apple needs to have a desktop computer that’s between the Mac Mini and this Mac Pro. IMac Pro doesn’t qualify because it’s still overpriced for the performance and we need to be able use our own monitors with a computer that’s more powerful than the Mac Mini.

 

[Misha Engel]

What can we guess one Vega II Dual MPX module will cost?

Atleast RTX Quadro 6000 x2 + 30% apple tax ~ $12k, when it's lower I would be surprised.

 

[Andy_Johnson]

I completely agree with Bruce. You could easily build a $2000 computer that you can safely overclock and have similar performance to the $6000 entry level Mac Pro. In the past the entry level Mac Pro was always $3000 or less. I understand that Apple’s hardware is built better than most others, and I’m willing to pay extra for that. But not $4000 extra. This entry Mac Pro should be $3000-$3500.

That's not quite true. Yes you can overclock, but considering apples-to-apples comparisons (no pun intended) since we want equal hardware longevity, a comparable build would be around $3200.
Linus covered that here: https://youtu.be/PPiEpSMkzoo?t=724
The rest of that video after the time I linked, he also covers the past Mac Pro prices - we are all in agreement on the fact that it's marked way up haha.

I agree you can save money by building it, but not as low as $2000 if you expect the same quality/longevity.

 

[Blair S. Paulsen]

Can we please stop comparing the announced base model against pimped PCs - it's pointless. The more relevant comparo would be in the $8-12K range. I'd imagine the Apple Tax would still be substantial, but not the "obvious ripoff" narrative one can construct with 256GB flash, a modest CPU and a "placeholder" GPU.

Cheers - #19