New MacPro predictions? Hopes? Fears?

[Antony Newman]

If Apple are making the big move to ARM in 2020/21 ... Why would they remove 32bit x86 compatibility from MacOs this year?

Those utilising tools or Apps that are (still) only available 32bit x86 might Not upgrade to the latest OS

Could this be Apple forcing a level (64 bit only) playing field ready for XCode compilation on Both x86 and ARM.

But why force that this year unless MacOs on ARM is coming out before Q4 2020?

Next question : If a MMP is sold in 2019Q4 on x86, and next year Apple is moving wholesale to ARM - who would want to splash out this year on something that is will be obsoleted the following year?

AJ

 

[M Most]

I also can’t wait to see the new Mac Pro. I think it’s going to be pretty awesome...

Much of what you mention is irrelevant if Apple doesn't re-establish a proper relationship with Nvidia. As long as entities like Red limit their support efforts to Cuda, Apple hardware is at a crippling disadvantage for many of us.

 

[Rakesh Malik]

I think we’ll see a shift to ARM on both the Mac and PC side of things. Intel is working on some pretty compelling ARM products of their own. The tried and true x86 architecture is still relevant, but I think it’s reached a point where it’s carrying around so much legacy bloat that it’s become far too difficult or costly to evolve.

Intel is nearly 2 years late shifting to 10nm process with only a couple CPU products finally shipping late last year. It’s not that they can’t do 10nm... They’ve already shown they can do 7nm on some of their memory products. It’s getting that bloated legacy mess that we know as x86 to scale reliably to 10nm. A daunting challenge to say the least.

That's obviously not the reason. Making 7nm DRAM cells is FAR less complicated than making processors. Nearly every fab launches using simpler to manufacture things like memory, so that's hardly new. Intel's 14, 12, 10, 7nm transition woes aren't related to x86; if they were, AMD would be facing the same problems since the AMD processor line is also x86. AMD is however working with semiconductor manufacturers far larger than Intel, and who have been collaborating for years; Intel's been going mostly solo on that front, and has had some obvious internal problems. Intel hired a very high demand semiconductor manufacturer and put him in charge of Intel's manufacturing in order to get back on track. Intel knows it blew it, and it's trying to get back in gear before the harm is irrevocable -- and this time around, AMD isn't constrained by capacity, because TMSC can outproduce Intel.

We won’t see the shift to ARM, or Apple’s own chips in their systems, until they can truly outperform their x86 counterparts. And that day is rapidly approaching...

I suspect that day is farther off than the Apple and ARM fans believe. With AMD putting such a huge flame under Intel's booty it's very unlikely that even Apple's ARM will be able to catch x86 anytime soon.

Nearly every ARM project that's been aimed at the server and workstation market has ended up capitulating to the x86 juggernaut. The only real contender is the Fujitsu monster which looks very similar to a ThreadRipper.

Things in the GPU front will probably be a lot more interesting though, especially with Intel developing a new GPU... but with this time with an experienced GPU architect at the helm, so hopefully not another Knight's Landing. NVidia needs some competition too, and it's not likely to come from AMD this year, unfortunately.

 

[Rakesh Malik]

If Apple are making the big move to ARM in 2020/21 ... Why would they remove 32bit x86 compatibility from MacOs this year?

Probably just cleaning up legacy garbage; there aren't even any 32-bit x86 processors left, so why bother supporting it? It's a waste of time and disk space. And maintenance effort.

Those utilising tools or Apps that are (still) only available 32bit x86 might Not upgrade to the latest OS

Right, all six of them... there's a mode in x86 that runs 32-bit applications, which really just means using 32-bit memory pointers, but not based on the old legacy 32-bit mess that used to be x86. That's what I'd guess that Apple is abandoning. That stuff could also continue to operate via emulation anyway if it's really necessary.

Next question : If a MMP is sold in 2019Q4 on x86, and next year Apple is moving wholesale to ARM - who would want to splash out this year on something that is will be obsoleted the following year?

That's why Apple is secretive, and why none of this is certain; Apple's not going to confirm/deny it until it's good and ready to launch whatever it's launching.

 

[Jeff Kilgroe]

Much of what you mention is irrelevant if Apple doesn't re-establish a proper relationship with Nvidia. As long as entities like Red limit their support efforts to Cuda, Apple hardware is at a crippling disadvantage for many of us.

I agree with that on some level, especially as it pertains to ‘right now.’ However there is nothing guaranteed that things will stay that way moving to the future. The industry decided CUDA was superior to OpenCL, and rightly so as OpenCL failed to evolve to meet the needs of more computational tasks. Apple is pushing their own Metal platform and with Metal2 and CoreML it is becoming quite capable, also gaining support rapidly. Adobe, Autodesk, SolidWorks and more are all on board. Both AMD as well as Nvidia have pledged full support for Metal from the hardware side. And I know that Nvidia is really trying lately and seem to be wanting to work with Apple again. For something like the R3D SDK, RED would have to build that capability. In order for them to do so, Apple will have to release a compelling hardware option...

 

[Misha Engel]

NVidia needs some competition too, and it's not likely to come from AMD this year, unfortunately.

Yep, AMD Radeon VII(15 fps) is only 50% faster than a RTX Titan X(10 fps), in Resolve with 8k .R3D 9:1 25 fps, with Basic + 4 Power Window + 3 OpenFX + TNR, on the otherhand nobody in the industry uses Davinci Resolve, OpenFX and TNR.
https://www.pugetsystems.com/labs/articles/DaVinci-Resolve-15-AMD-Radeon-VII-16GB-Performance-1382/

NVidia can't use their memory tricks anymore, it's a security risk https://nvidia.custhelp.com/app/answers/detail/a_id/4772

Apple cares about ProResRAW and ProRes in the content world, the rest is irrelavent to them.

 

[Rakesh Malik]

Yep, AMD Radeon VII(15 fps) is only 50% faster than a RTX Titan X(10 fps), in Resolve with 8k .R3D 9:1 25 fps, with Basic + 4 Power Window + 3 OpenFX + TNR, on the otherhand nobody in the industry uses Davinci Resolve, OpenFX and TNR.
https://www.pugetsystems.com/labs/articles/DaVinci-Resolve-15-AMD-Radeon-VII-16GB-Performance-1382/

In light of how small a slice of the GPU market that is, it's largely irrelevant as far as its affect on NVidia and its pricing.

I rather like the Radeon VII and I'm glad that I have one; I plan on attaching it to a mac mini that I'll dedicate to Resolve and the student Baselight shortly. But I'm transitioning most of my grading to Mistika and Scratch, both of which are a LOT faster on NVidia hardware than on even a Radeon VII, and the same holds for most of the higher end OFX like the Boris suite. NVidia's edge isn't computing power, it's the fact that it's working with 3rd parties to get their software optimized around features the Turing line has that the Vega line doesn't, and AMD is way behind on that front.

Honestly I don't see AMD making a dent in NVidia's market this year, but hopefully as it gets healthier it will ramp up its GPU R&D and rejoin the race in earnest.

Apple cares about ProResRAW and ProRes in the content world, the rest is irrelavent to them.

That's even more myopic than usual for Apple.

 

[Misha Engel]

In light of how small a slice of the GPU market that is, it's largely irrelevant as far as its affect on NVidia and its pricing.

I rather like the Radeon VII and I'm glad that I have one; I plan on attaching it to a mac mini that I'll dedicate to Resolve and the student Baselight shortly. But I'm transitioning most of my grading to Mistika and Scratch, both of which are a LOT faster on NVidia hardware than on even a Radeon VII, and the same holds for most of the higher end OFX like the Boris suite. NVidia's edge isn't computing power, it's the fact that it's working with 3rd parties to get their software optimized around features the Turing line has that the Vega line doesn't, and AMD is way behind on that front.

Honestly I don't see AMD making a dent in NVidia's market this year, but hopefully as it gets healthier it will ramp up its GPU R&D and rejoin the race in earnest.



That's even more myopic than usual for Apple.

I'm okay with NVidia raising the price even more, that's up to NVidia. I look at price/performance for content creation and currently AMD gives us the best price/performance ratio.

BorisFX GPU requirements: Graphics Cards must support OpenGL 2.1 or higher.
https://borisfx.com/support/supported-graphics-cards/

Can you name some of the features optimize for Turing in the content creation area?

EA is joining the Khronos Group(Vulkan), Davinci is supporting VCE h.264/265, Apple is AMD and Intel only and actively working against NVidia.
Next gen consoles will again use AMD.

The more negative Jensen is about AMD the better AMD goes.

Navi is on the horizon 7/7/2019.

The one thing AMD is really behind is HPC adoptation and their number of pixel shaders for games.

Long story short, AMD has an upto a 500% better price/perf ratio with 8K footage (16 GB or more GPU's).

 

[M Most]

Long story short, AMD has an upto a 500% better price/perf ratio with 8K footage (16 GB or more GPU's).

I don't know how you can say that on RedUser when the new Red SDK only supports additional GPU functionality on Nvidia GPUs due to it's Cuda exclusivity...

 

[Jeff Kilgroe]

Yep, AMD Radeon VII(15 fps) is only 50% faster than a RTX Titan X(10 fps), in Resolve with 8k .R3D 9:1 25 fps, with Basic + 4 Power Window + 3 OpenFX + TNR, on the otherhand nobody in the industry uses Davinci Resolve, OpenFX and TNR.
https://www.pugetsystems.com/labs/articles/DaVinci-Resolve-15-AMD-Radeon-VII-16GB-Performance-1382/

Yes... BUT it flips the other way in Resolve when using the builds supporting the Nvidia R3D enhancements. The thing is, AMD could also work with RED to accomplish much the same thing. But as was mentioned above, AMD just isn’t at that stage yet in terms of integrating so closely with developers. GPU hardware has never been where AMD falls short, it’s always been their software and developer support. Their drivers/ software have improved dramatically as of late.

Back on the Apple thing above... Apple could decide to offer Nvidia GPUs with the Mac Pro, or even an open GPU setup where we could use off the shelf GPUs. At that point we would see Nvidia bring CUDA back current on the Mac platform.

 

[Misha Engel]

I don't know how you can say that on RedUser when the new Red SDK only supports additional GPU functionality on Nvidia GPUs due to it's Cuda exclusivity...

You still have to process the de-bayer, OpenFX effects and TNR, no matter if you use CPU or GPU .R3D decoding.
As long as your not decode limited, the other limits start to kick in: De-bayer, OpenFX, TNR, etc..

In Resolve (full-res premium) with 8k .R3D 9:1 25 fps, with Basic + 4 Power Window + 3 OpenFX + TNR (CPU is fast enough to decode the R3D in realtime i9-7960 and up or TR1950 and up)

Radeon VII: $700/15fps ~ $ 47/fps
RTX Titan X: $2500/10fps ~ $ 250/fps
Quadro RTX 5000: $2300/8fps ~ $ 287/fps

For 8k 7:1 25fps you'll need at least a 16core 4.1 GHz cpu to decode in realtime at the moment (overclocked tr2950x or i9-7960x and up).
For 8k 5:1 25fps you'll need at least a 16core 4.5 GHz cpu to decode in realtime at the moment (i9-7960x and up).

Looking forward to the R9-Ryzen series of cpu's

 

[Misha Engel]

Yes... BUT it flips the other way in Resolve when using the builds supporting the Nvidia R3D enhancements. The thing is, AMD could also work with RED to accomplish much the same thing. But as was mentioned above, AMD just isn’t at that stage yet in terms of integrating so closely with developers. GPU hardware has never been where AMD falls short, it’s always been their software and developer support. Their drivers/ software have improved dramatically as of late.

Back on the Apple thing above... Apple could decide to offer Nvidia GPUs with the Mac Pro, or even an open GPU setup where we could use off the shelf GPUs. At that point we would see Nvidia bring CUDA back current on the Mac platform.

What are those Nvidia R3D enhancements?

Apple can do whatever they want and at the moment they don't support NVidia in their latest OS (Radeon VII is supported in 10.15 beta).

 

[M Most]

What are those Nvidia R3D enhancements?

Decompression and debayer on the GPU. Using even a 1080Ti, each GPU gets you about 15fps of performance at full debayer, and they scale almost linearly. So by putting 2 of them into a "cheap" PC (the CPU almost doesn't matter) you can get better than real time performance from Red 8K files at full debayer, if that's important to you.

Apple can do whatever they want and at the moment they don't support NVidia in their latest OS (Radeon VII is supported in 10.15 beta).

Which is why the Mac, at this point in time, is a really shitty platform for Red if you want full debayer or close to it. It's fine for a lot of things, but processing Red files at high quality isn't one of them. The sad part is that this could be instantly addressed if Apple wanted to.

 

[Misha Engel]

Decompression and debayer on the GPU. Using even a 1080Ti, each GPU gets you about 15fps of performance at full debayer, and they scale almost linearly. So by putting 2 of them into a "cheap" PC (the CPU almost doesn't matter) you can get better than real time performance from Red 8K files at full debayer, if that's important to you.

AMD took another route:
(for decoding)1x R9-38xx ~ $ 500..600 ~125..135Watt(real) + (for de-bayer and effects) 1 Radeon VII $700 ~300 Watt(real) ~ $ 1,300 and ~ 435 Watt for Full-Res Premium 8k 5:1 25 fps playback in realtime.

NVidia + Intel route:
CPU i9-9900k ~$ 500 ~ 135 Watt(real) + (for decoding, de-bayering and effects) 1 GTX2080ti $1,200 280 Watt(real) ~ $ 1,700 and ~ 415 Watt for Full-Res Premium 8k 5:1 25 fps playback in realtime.

I don't know when the NLE's will support the NVidia decoding, the estimates for the R9-38xx are 7/7/2019.

Which is why the Mac, at this point in time, is a really shitty platform for Red if you want full debayer or close to it. It's fine for a lot of things, but processing Red files at high quality isn't one of them. The sad part is that this could be instantly addressed if Apple wanted to.

Apple's current hardware is kind of weak for realtime editing with 8K at the moment, we will have to wait an see what they come up with. In the meantime Linux and Windows 10 works fine with almost every kind of computer hardware.

 

[M Most]

AMD took another route:
(for decoding)1x R9-38xx ~ $ 500..600 ~125..135Watt(real) + (for de-bayer and effects) 1 Radeon VII $700 ~300 Watt(real) ~ $ 1,300 and ~ 435 Watt for Full-Res Premium 8k 5:1 25 fps playback in realtime.

NVidia + Intel route:
CPU i9-9900k ~$ 500 ~ 135 Watt(real) + (for decoding, de-bayering and effects) 1 GTX2080ti $1,200 280 Watt(real) ~ $ 1,700 and ~ 415 Watt for Full-Res Premium 8k 5:1 25 fps playback in realtime.

You're missing my point. Red has embraced Cuda and has not to this point given any solid indication if and when they will support other graphics frameworks. So AMD doing things based on a brute force approach to current file formats and the current Red SDK is a bit counter to what Red themselves are doing, and will by definition not be as effective or scalable going forward. One can come up with whatever financial justification one wants, but that is the fact. Nvidia is supported directly by Red and other approaches are not, at least to this point in time.

Apple's current hardware is kind of weak for realtime editing with 8K at the moment, we will have to wait an see what they come up with. In the meantime Linux and Windows 10 works fine with almost every kind of computer hardware.

That is true, but the title of this thread is "New MacPro predictions? Hopes? Fears?.", not "Mac vs. PC." So, while self evident, use of other OS's is rather irrelevant to this discussion...

 

[PatrickFaith]

I really like my AMD CPU, but if it doesn't come with nvidia rtx it's a non-starter for me. Waited way to long for it to have anything questionable as a pro system. Also if there is any cpu/GPU heat throttling because of design look over functionality, then it's just joke time.

 

[Blair S. Paulsen]

(snip) The sad part is that this could be instantly addressed if Apple wanted to.

Ding, ding, ding. Full engagement between Apple and nVidia could yield "right sized" solutions for many in our industry. Machine rooms full of servers (or cloud compute resources) are the obvious path for shops needing to cycle through many TBs of assets, render out long form with TNR/composites/etc. There are also plenty of folks for whom a hot laptop or iMacPro with fast storage over TB3 is all they need.

What a lot of RedUsers would like is an under $10K solution that eats R3Ds for lunch, doesn't throttle unless you pound on it and can run most crucial applications well. For some of us, a Macintosh with nVidia GPU(s) that doesn't make a lot of noise or heat would be an excellent solution. Will Apple deliver? Are there enough of us for them to care?

Cheers - #19

 

[Dermot Shane]

The ACES DCTL set has a complex and multi nested structure, therefore exceeding the boundaries of what OpenCL and Metal can process. This also applies to GPU kernels in the OFX plugins. Only CUDA and C (for CPU processing) can handle such a structure.

simply too limiting to have only a choice of Metal or CL in 2019

 

[PatrickFaith]

Ding, ding, ding. Full engagement between Apple and nVidia could yield "right sized" solutions for many in our industry. Machine rooms full of servers (or cloud compute resources) are the obvious path for shops needing to cycle through many TBs of assets, render out long form with TNR/composites/etc. There are also plenty of folks for whom a hot laptop or iMacPro with fast storage over TB3 is all they need.

What a lot of RedUsers would like is an under $10K solution that eats R3Ds for lunch, doesn't throttle unless you pound on it and can run most crucial applications well. For some of us, a Macintosh with nVidia GPU(s) that doesn't make a lot of noise or heat would be an excellent solution. Will Apple deliver? Are there enough of us for them to care?

Cheers - #19

Ding^4. Also for a long time it was just the colorist needing the serious gpu, then the raw ingest people as people are dropping the rocket cards ... now editors on raw ... now autodesk is moving their serious rendering to the gpu (is still in beta).

Along with the compositing software starting to use the gpu more, the entire post side needs nvidia gpu's. So I "hope" Apple is doing their research on the number of professionals that need the nvidia gpu. Also as more people use tensorflow, which right now is a huge amount of people outside of post, tensorflow really needs nvidia gpu for those doing image convolution - which I think will start to impact the industry late 2020. Here's a little video on autodesk moving it's rendering over to gpu as an example of most of production is starting to require serious gpu power: https://lesterbanks.com/2019/03/hands-on-arnold-gpu/

I have been moving to Linux, but man it's a pain in the butt running a small Linux post shop without a dedicated Linux guru - which isn't needed for a Mac shop.

 

[Michael Tiemann]

I have been moving to Linux, but man it's a pain in the butt running a small Linux post shop without a dedicated Linux guru - which isn't needed for a Mac shop.

You don't need a special guru for a Mac shop becuase Misha Engle will tell you FOR FREE how much Apple sucks.