The New Mac Pro is Here!!!

[DJ Meyer]

A dozen of the first batch here.
A few years ago I was reading on overheating issues and did not experience it till a few months ago with a D700.
That's one of 12 and after an OS re-install, the unit is serving fine as a graphics design station.
We shall see how long the other one will last.
Its hosting Avid and Pro Tools hardware.
I am hoping it dies, maybe I can replace with the new model

We haven't been so lucky. 4 of 8 D700 models have been in the shop at least once for GPU replacements.

 

[Jeff Kilgroe]

I melted the D700 GPUs twice on the same 12-core trashcan Mac. My other two were 8-core units and in one system I swapped the 8-core for a 10-core CPU just because. Never had a problem with either of those. But they were short-lived for me as I sold my last one a year ago when the iMac Pro could replace it.

The trashcan Mac Pro was awesome for a lot of uses. Excellent node system for clusters and they stacked laying horizontally like pop cans with the right hardware in equipment racks. With the TB2 ports we had multi-port 10G interconnects right out of the box. At times I think Apple didn’t understand the markets they were trying to serve and yet made something awesome for a market they tried to abandon when they killed the Xserve. All things considered, the system was a nice idea but flawed execution and they missed their target markets needs. If they offered it with i5 and i7 CPUs and singular decent desktop-class GPU, all at a much lower price point, they could have had a winner...

 

[Antony Newman]

What would be interesting is if Apple don’t limit on the number of CPU cores recognised by MacOs.

If 64 cores are ‘allowed’, and the Afterburner is just a PCIe device ... what are the chances that a PCIe4 Hackintosh with quadruple Navi + ABurner will make an appearance?

AJ

 

[Blair S. Paulsen]

Seriously doubt you'd be able to run an Afterburner accelerator card on anything but an actual Mac.

Extremely unlikely that Apple will do anything of the kind, but selling a lower spec version of the tcMP 6,1 to fill the space between the MacMini and the mMP 7,1 workstation would be great for many users. They already have the infrastructure in place and NRE recovered to keep producing the cylinders at a reasonable cost per unit. Nah, makes too much sense...

Cheers - #19

 

[Misha Engel]

Why would you use a vendor locked FPGA(Afterburner accelerator card) when you can do everything with off the shelf GPU's and CPU's.

 

[Blair S. Paulsen]

Why would you use a vendor locked FPGA(Afterburner accelerator card) when you can do everything with off the shelf GPU's and CPU's.

You could be totally right Misha. That said, I hope you don't mind if I wait until the actual card is in the wild before concluding what it can or cannot do vs off the shelf components...

Cheers - #19

 

[Misha Engel]

You could be totally right Misha. That said, I hope you don't mind if I wait until the actual card is in the wild before concluding what it can or cannot do vs off the shelf components...

Cheers - #19

With 4x RTX titanX cards, you can edit 4x (8k.R3D 5:1 24 fps) streams in full quality and in real-time with GPU decode(when are NLE's going to support this?), what more do you want?
Add one of those 64 core EPYC cpu's + 512 GB of mem to it and you can handle 2+ 8k.R3D streams in realtime when GPU decoding doesn't work (for around half the price ~20k).
Most if not all software mentioned during WDCC is already optimized for CUDA or OpenCL, so no benefit there.

 

[Antony Newman]

Apple need all the big frameworks and (3D intensive) software to be ported into Metal, and no doubt have a $$$ war chest to make that a reality over the next couple of years.

For as long as what has been CUDA only software is not ‘production proven’ after porting - Nvidia have a locked in professional market, and so I doubt that even if Apple offered Nvidia money, that Nvidia would offer Metal drivers.

When Intel roll out their Xe GPU, something tells me that will be offer a Metal drivers (and not CUDA), and will likely displace Nvidia as at the top end performance wise (for whichever market they are taking aim). Would Nvidia finally release Metal drivers I wonder?

Perhaps the question right now shouldn’t be if you want a >MacPro or not, but rather do you think that MacOs will flourish with FPGA accelerated functionality that could enable greater application performance than GPU alone?

Do you want to use software from Blackmagic where GPU software bottlenecks will benefit from hardware acceleration explicitly built into the OS?

AJ

 

[DJ Meyer]

Apple need all the big frameworks and (3D intensive) software to be ported into Metal, and no doubt have a $$$ war chest to make that a reality over the next couple of years.

For as long as what has been CUDA only software is not ‘production proven’ after porting - Nvidia have a locked in professional market, and so I doubt that even if Apple offered Nvidia money, that Nvidia would offer Metal drivers.

When Intel roll out their Xe GPU, something tells me that will be offer a Metal drivers (and not CUDA), and will likely displace Nvidia as at the top end performance wise (for whichever market they are taking aim). Would Nvidia finally release Metal drivers I wonder?

Perhaps the question right now shouldn’t be if you want a >MacPro or not, but rather do you think that MacOs will flourish with FPGA accelerated functionality that could enable greater application performance than GPU alone?

Do you want to use software from Blackmagic where GPU software bottlenecks will benefit from hardware acceleration explicitly built into the OS?

AJ

Apple clearly has the $$$ to make it happen. They now claim to have the will, but we will see. Press releases are easy. Sustaining organizational attention on the tiny (for Apple, as a % of their business) pro market will not be.

This Mac Pro is a great first step but there is a long way to go.

 

[Misha Engel]

Apple Shifts Mac Pro Production To China From The US.

Nice 25% price increase(Trump China tax) for the consumer and the workforce of Flex Ltd in the United States are out of work.

Just love the big tech companies...

 

[Bruce Allen]

Perhaps the question right now shouldn’t be if you want a >MacPro or not, but rather do you think that MacOs will flourish with FPGA accelerated functionality that could enable greater application performance than GPU alone?

Well, currently MacOS is behind in FPGAs vs Windows because you can get FPGA cards for PCs and Windows right now

They have even been used in the past in our industry (eg Quantel iO - I have worked on many projects that finished on that system, which was impressive at the time but ultimately lost out to GPU-based approaches).

And wasn't the REDROCKET card FPGA-based too?

If they were really a TON better for our industry, we'd be already using them more, I'd guess.

We'll see what happens with AI, Intel AgileX, etc.

I suspect the Apple one is one of the existing solutions with a little custom sauce, we'll see!

Bruce Allen
www.bruceallen.tv

 

[Jeff Kilgroe]

Yes, RED Rocket was FPGA. Rocket-X was ASIC. The RED One camera was built on FPGAs, subsequent DSMC[2] platforms are ASIC.

The FPGA in the upcoming Apple Afterburner card are still a mystery somewhat. We don’t know the extend of its capabilities or how adaptable it will be. Like with ASIC GPU solutions, FPGAs can be state of the art and very powerful when they are released, but soon lose out to newer FPGA or ASIC solutions. Such is the nature of tech... Having the card in the new Mac Pro could be a good thing if Apple prices it right and supports developers to make it great. If Apple handles this right, it’s not an alternative to the GPU options they’re offering, but complementary. No reason it can’t work in harmony with multiple GPUs and CPU cores to further accelerate a variety of tasks. OTOH, if they give us this FPGA card to just accelerate ProRes and perhaps a couple other video codecs and related processing and don’t utilize it fully in conjunction with their GPU offerings then we would have to say “WTF, Apple?”

The main reason our industry isn’t using a ton of FPGA solutions and has adopted GPUs is because GPU tech is driving forward at a fast and steady pace. GPUs are volume sellers for gaming, ML/AI, and computational tasks these days with a huge market. Much easier to adapt our software and tools to use these commodity GPUs than it is to adopt a proprietary FPGA system. If Apple’s Afterburner card is 10% faster than a GTX 2080 at processing ProRes RAW or R3D and clocks in at $1000 or less and can be leveraged along with the GPUs, then it will be a great deal... for now. When the next round of GPUs arrive and up the performance by another 25% while maintaining roughly the same pricing, then we have to question if having the FPGA is enough extra in the mix to bother with it. 2 or 3 years down the road, it won’t make sense when GPUs have doubled their computational performance vs now.

...Just like we saw with the RED Rocket and Rocket-X. Obsoleted by GPUs. And the OG Rocket was surpassed by CPU tech a while ago.

 

[Blair S. Paulsen]

Assuming Apple is aware of the situation Jeff describes, there are two possibilities:

1) They figure a lot of companies in our industry will get one, even if it's not truly necessary, which makes them money.
2) There's more to it than we know at this juncture. In particular, I'd be curious if it might enable more responsive multi-format timelines and better speed ramping.

IAC, I'll wait until they manifest before calling them redundant and overpriced...

Cheers - #19

 

[DC Chavez]

"Jony Ive’s departure marks the end of the hardware era at Apple"

His final piece of design is a Cheese Grater: https://www.cnbc.com/2019/06/28/jony-ives-departure-marks-the-end-of-the-hardware-era-at-apple.html

 

[DJ Meyer]

"Jony Ive’s departure marks the end of the hardware era at Apple"

His final piece of design is a Cheese Grater: https://www.cnbc.com/2019/06/28/jony-ives-departure-marks-the-end-of-the-hardware-era-at-apple.html

Perhaps Tim Cook overruled him on the Mac Pro and rumored 16" pro laptop. So offended by the Macs being defiled with usable ports and PCI-E slots, he resigned. :tongue_smilie:

 

[Antony Newman]

They lost Ive but gained Filippo (ARMs Lead coherent mesh network / Lead lower power CPU / Lead high power desktop CPU architectural designer...).

What is puzzling : Apple has gone to great lengths to leverage the AMD Infinity Fabric to speed up GPU - GPU xfers. Why didn’t they switch to AMD for the CPU - and also leverage IF for CPU <-> GPU?

Not sure staying ‘loyal’ to Intel for the CPU makes sense ... especially if there are plans afoot to migrate to ARM in the next few years.

AJ

 

[Blair S. Paulsen]

(snip) Not sure staying ‘loyal’ to Intel for the CPU makes sense ... especially if there are plans afoot to migrate to ARM in the next few years. AJ

OTOH, if ARM is coming soon, why develop a whole new code base for AMD at all. Apple is already committed to supporting Intel CPUs for another 10 years (assuming they don't change policy). Do they really want to wake up in a couple years supporting Intel, AMD and ARM simultaneously? There's also the limited Thunderbolt support on AMD compared to Intel. Bottom line, to avoid being even later to the party - it had to be Intel.

Cheers - #19

 

[L. Langer]

They lost Ive but gained Filippo (ARMs Lead coherent mesh network / Lead lower power CPU / Lead high power desktop CPU architectural designer...).

What is puzzling : Apple has gone to great lengths to leverage the AMD Infinity Fabric to speed up GPU - GPU xfers. Why didn’t they switch to AMD for the CPU - and also leverage IF for CPU <-> GPU?

Not sure staying ‘loyal’ to Intel for the CPU makes sense ... especially if there are plans afoot to migrate to ARM in the next few years.

AJ

In short, it makes business sense to stay away from AMD CPUs. Intel is now contracting with Samsung for replication of their CPUs, so they can shift all of the Apple Gets First Dibs CPU business to Samsung's China fabs while using their own for their regular market business, which makes Mac production very easy, as Samsung's China fabs are literally up/down the road from the Mac assembly lines. In order to use AMD, Apple would have to switch to AMD CPUs (APU's too!) across the entire Mac line and they would have to cut another replication deal with TSMC, as AMD is fabless and relies on TSMC for their chip production. Given that, there simply may not be available fab space at TSMC to accommodate business from Apple at this time. Also, almost all of TSMC's fabs are in Taiwan and not China, so that also could factor into it, as the China and Taiwan politics come into play and Apple must always remain on good terms with the Chinese government.

 

[Misha Engel]

In short, it makes business sense to stay away from AMD CPUs. Intel is now contracting with Samsung for replication of their CPUs, so they can shift all of the Apple Gets First Dibs CPU business to Samsung's China fabs while using their own for their regular market business, which makes Mac production very easy, as Samsung's China fabs are literally up/down the road from the Mac assembly lines. In order to use AMD, Apple would have to switch to AMD CPUs (APU's too!) across the entire Mac line and they would have to cut another replication deal with TSMC, as AMD is fabless and relies on TSMC for their chip production. Given that, there simply may not be available fab space at TSMC to accommodate business from Apple at this time. Also, almost all of TSMC's fabs are in Taiwan and not China, so that also could factor into it, as the China and Taiwan politics come into play and Apple must always remain on good terms with the Chinese government.

All high-end Apple semiconductors are made by TSMC. Apple is also fabless like AMD and intel is a generation behind TSMC. Apple will assemble the Mac Pro in CHINA.
The reason Apple choose intel for the CPU is a business decision, not a technical one (AMD is way ahead of intel with X86-64 CPU's).

 

[Mike Halper]

In short, it makes business sense to stay away from AMD CPUs. Intel is now contracting with Samsung for replication of their CPUs, so they can shift all of the Apple Gets First Dibs CPU business to Samsung's China fabs while using their own for their regular market business, which makes Mac production very easy, as Samsung's China fabs are literally up/down the road from the Mac assembly lines. In order to use AMD, Apple would have to switch to AMD CPUs (APU's too!) across the entire Mac line and they would have to cut another replication deal with TSMC, as AMD is fabless and relies on TSMC for their chip production. Given that, there simply may not be available fab space at TSMC to accommodate business from Apple at this time. Also, almost all of TSMC's fabs are in Taiwan and not China, so that also could factor into it, as the China and Taiwan politics come into play and Apple must always remain on good terms with the Chinese government.

It would not be difficult for Apple to switch to AMD. They could flip a switch and it would be done. AMD Ryzen Hackintoshes run with a simple kernel patch. Some other optimizations and changes would be needed, but it wouldn’t be that difficult. And I don’t see why Apple couldn’t have the kernel adjusted to work for both Intel and AMD at the same time, or to just have to different versions of the OS like they did during the PowerPC->Intel transition. It would mostly be invisible to the end user. Thunderbolt would even be working since it works on a couple AMD-based motherboards. It doesn’t make sense for Apple to do that right now though when it’s pretty certain that the switch to ARM is really where they’re headed.