MP 7,1 AMD Big Navi is coming to Mac Pro and iMac Pro

[high heaven]

AMD Navi 21 "Big Navi" GPU stack leaks, budget gamers may have to contend with a Navi 10 Refresh rather than pinning hopes on Navi 23

SKU names and PCIe device IDs variants of AMD's upcoming "Big Navi" Navi 21 GPU across Gaming, Pro, and Apple-exclusive segments have leaked. Navi 21 will be largest die in the Navi 2x family and is expected to feature up to 80 CUs and 5,120 SPs. AMD would also be making available a Navi 10...

www.notebookcheck.net

It seems RDNA 2.0 based GPU will be used for Mac Pro and iMac Pro but why not CDNA based GPU?

 

[casperes1996]

AMD Navi 21 "Big Navi" GPU stack leaks, budget gamers may have to contend with a Navi 10 Refresh rather than pinning hopes on Navi 23

SKU names and PCIe device IDs variants of AMD's upcoming "Big Navi" Navi 21 GPU across Gaming, Pro, and Apple-exclusive segments have leaked. Navi 21 will be largest die in the Navi 2x family and is expected to feature up to 80 CUs and 5,120 SPs. AMD would also be making available a Navi 10...

www.notebookcheck.net

It seems RDNA 2.0 based GPU will be used for Mac Pro and iMac Pro but why not CDNA based GPU?

Because CDNA is meant for datacenter work. Considering Apple’s primary markets for pro devices are things like video editing, animation, 3D modeling and other graphical work, the graphics based architecture makes more sense than the number cruncher. CDNA is more akin to Tesla on Nvidia’s side. Apple would always pick the Quadra, FirePro and now Radeon Pro (RDNA) GPUs, not something like CDNA; which won’t even have video output in most cards it will come in I’m guessing

 

[th0masp]

"AMD could finally become competitive with NVIDIA's flagship GPUs with Navi 2x. Tom from the Moore's Law is Dead YouTube channel cited sources claiming that RDNA2 is only about 10% faster than the NVIDIA GeForce RTX 2080 Ti. However, he also says that these were likely to be mobility cards and desktop RDNA2 should be able to afford much higher performance."

Oh I see, so even a mobile AMD GPU is now going to beat 2018's 2080 Ti. Alright. Considering how well Puny Navi turned out not all that long ago and how AMD's higher powered desktop GPUs seem to generally perform more as space heaters than competitive offerings, I'll just take that with a teeny weeny grain of salt.

 

[Macintosh IIcx]

I think that this article gives a better overview: RDNA 2 for Apple

The interesting part is this:

• Navi 21 XTA (0x731F:50)
• Navi 21 XLA (0x731F:51)
• Navi 21 Pro-XTA (0x731F:11)
• Navi 21 Pro-XLA (0x731F:13)

If this is true, Apple will get four SKUs from the RDNA 2 / Navi 21.

Two Radeon Pro for Mac Pro and maybe iMac Pro, but also some non-pro SKUs. The last part is very interesting, if true. I would think that those GPUs would be too big to fit into the iMac vanilla thermal system, so maybe a redesign of that is on its way or an entirely new Mac desktop might see the light this fall?

 

[bigtomato]

Rumor iMac Pro refreshed this Christmas so sounds about right. Also, what are those current cards in Mac Pro with 2 gpu's on board....

 

[casperes1996]

Rumor iMac Pro refreshed this Christmas so sounds about right. Also, what are those current cards in Mac Pro with 2 gpu's on board....

Those are Vega GPUs. It’s in the name actually, the Vega II Duo. Funnily enough on the point of this post, Vega is closer to CDNA than RDNA - Vega was always really good at computing; it’s shaders were rarely the limiting part of the higher end Vegas - it was more often hindered by other aspects of the pipeline. Ideal for a compute card. But unlike CDNA Vega also was aimed to have a lot of graphic capability

 

[deconstruct60]

Because CDNA is meant for datacenter work. Considering Apple’s primary markets for pro devices are things like video editing, animation, 3D modeling and other graphical work, the graphics based architecture makes more sense than the number cruncher.

Except that is exactly what Apple did for the Pro Vega II. They took the "number cruncher" Vega 20 die. Video editing doesn't technically need 3D graphics for a 2D medium. Converting from one video storage format to another is more computation than .

CDNA is more akin to Tesla on Nvidia’s side.

CDNA descends from Vega 20 chip which Apple already using. Unless AMD completely rips out the the whole DisplayPort output system it is probably a candidate for the top end MPX module. Even more so if pragmatically AMD is packaging two dies into one package ( making it more like a MPX Vega Duo straight from the AMD package factory, so even less work for Apple to contract for a Duo card. )

Apple would always pick the Quadra, FirePro and now Radeon Pro (RDNA) GPUs, not something like CDNA; which won’t even have video output in most cards it will come in I’m guessing

There has been a report that the CDNA almost completely stripped the video out to throw more transistors at computation engines. If it is just plain missing then Apple won't select it. But if it is a matter the AMD's cards just don't attach the display PHYS adapter chips+socket to the edge of the card , then Apple largely wasn't going to do the same thing anyway ( Thunderbolt 3 (or 4) versus mini-DP ports (or a plain Type-C ).

The CDNA card won't fit all the way down the "bottom" of the MPX line up. The largest gap the Mac Pro has right now in term of customer discontent and "friction" (grumbling) is the 580X. They absolutely don't need a 500mm^2 chip to solve that issue.

Apple relatively just got the W5700X out the door. If there was a 'big Navi' with HBM that would more likely be the Apple one. If there is on then it probably will be last "out of the gate" in AMD's roll out plan. Apple could take "Big Navi" as an option, but that doesn't necessarily point toward "soon" (2020).

 

[casperes1996]

Except that is exactly what Apple did for the Pro Vega II. They took the "number cruncher" Vega 20 die. Video editing doesn't technically need 3D graphics for a 2D medium. Converting from one video storage format to another is more computation than .



CDNA descends from Vega 20 chip which Apple already using. Unless AMD completely rips out the the whole DisplayPort output system it is probably a candidate for the top end MPX module. Even more so if pragmatically AMD is packaging two dies into one package ( making it more like a MPX Vega Duo straight from the AMD package factory, so even less work for Apple to contract for a Duo card. )



There has been a report that the CDNA almost completely stripped the video out to throw more transistors at computation engines. If it is just plain missing then Apple won't select it. But if it is a matter the AMD's cards just don't attach the display PHYS adapter chips+socket to the edge of the card , then Apple largely wasn't going to do the same thing anyway ( Thunderbolt 3 (or 4) versus mini-DP ports (or a plain Type-C ).

The CDNA card won't fit all the way down the "bottom" of the MPX line up. The largest gap the Mac Pro has right now in term of customer discontent and "friction" (grumbling) is the 580X. They absolutely don't need a 500mm^2 chip to solve that issue.

Apple relatively just got the W5700X out the door. If there was a 'big Navi' with HBM that would more likely be the Apple one. If there is on then it probably will be last "out of the gate" in AMD's roll out plan. Apple could take "Big Navi" as an option, but that doesn't necessarily point toward "soon" (2020).

Yes, Vega is a compute powerhouse, and it is true that for most video tasks, 3D geometry and such isn't required (though for VFX work and 3D modelling it can be); But Vega was still a GPU that had all the 3D graphics power still. CDNA descends from Vega, but as I understand it, has all display output and geometry removed entirely. - I may very well be mistaken though and it could be as you say just the physical layer that Apple would replace with their Thunderbolt port anyway; Guess time will tell

 

[eflx]

Uh, the Vega II in the Mac Pro is using the CDNA architecture already. What do you mean?

Oops! Clearly got my AMD acronyms mixed up

 

[casperes1996]

Uh, the Vega II in the Mac Pro is using the CDNA architecture already. What do you mean?

No; It's using a second generation Vega. CDNA is descendant from Vega, but it is not the same

 

[deconstruct60]

Yes, Vega is a compute powerhouse,
...CDNA descends from Vega, but as I understand it, has all display output and geometry removed entirely. - I may very well be mistaken though

There is a reduction.

"... Just how different these architectures are (and over time, will be) remains to be seen. AMD has briefly mentioned that CDNA is going to have less “graphics-bits”, so it’s likely that these parts will have limited (if any) graphics capabilities, making them quite dissimilar from RDNA GPUs in some ways. ..."

AMD Unveils CDNA GPU Architecture: A Dedicated GPU Architecture for Data Centers

www.anandtech.com

I can't find direct quotes from AMD or a transcript to get direct feel for how they are bounding "less". But the "have less" other folks have presented as ...

"... Data centers and HPCs using Radeon Instinct accelerators have no use for the GPU's actual graphics rendering capabilities. And so, at a silicon level, AMD is removing the raster graphics hardware, the display and multimedia engines, and other associated components that otherwise take up significant amounts of die area. In their place, AMD is adding fixed-function tensor compute hardware, similar to the tensor cores on certain NVIDIA GPUs. ... "

AMD Announces the CDNA and CDNA2 Compute GPU Architectures

AMD at its 2020 Financial Analyst Day event unveiled its upcoming CDNA GPU-based compute accelerator architecture. CDNA will complement the company's graphics-oriented RDNA architecture. While RDNA powers the company's Radeon Pro and Radeon RX client- and enterprise graphics products, CDNA will...

www.techpowerup.com

That isn't really a "have less" as much as chuck the whole raster system down through the display managers out the window. If it is a total swap on transistor budget to tensor cores then CDNA will have problems in the workstation market (and with Apple as they want some GUI for someone to sit in front of. ).

If AMD is chasing eye-popping high tensor computational ops per second then "have less" may be something AMD was doing not to show all their cards ( or let Nvidia know than know how much they knew what Ampere intended to do; same thing 'bet the farm' on tensor subsystem. )

In trying to chase down some Vega die layouts perhaps the multimedia stuff is soaking up more space than I remember it did.

and it could be as you say just the physical layer that Apple would replace with their Thunderbolt port anyway; Guess time will tell

If the raster+display got pruned down to just 1-2 DP streams I can see Apple passing on that that too. At that point perhaps barely handling one XDR display. I thought it would get capped around 4, but perhaps AMD did 'bet the farm' with zero.

 

[deconstruct60]

I think that this article gives a better overview: RDNA 2 for Apple

The interesting part is this:

• Navi 21 XTA (0x731F:50)
• Navi 21 XLA (0x731F:51)
• Navi 21 Pro-XTA (0x731F:11)
• Navi 21 Pro-XLA (0x731F:13)

If this is true, Apple will get four SKUs from the RDNA 2 / Navi 21.

Four SKU of what is probably the exact same actual GPU die (and the gyrations are more so playing around with memory capacity and active CUs ) . Or maybe two dies ( if Pro means HBM versus GDDR6 )

But these overviews of mapping these down onto the 5700 variants don't make much sense. This chip is substantively larger. It isn't going in at anywhere near those price points. At best this might imitate a similar layering structure at a higher price range, but some of this is couched as a "replacement for". That's probably not true.

Two Radeon Pro for Mac Pro and maybe iMac Pro, but also some non-pro SKUs. The last part is very interesting, if true. I would think that those GPUs would be too big to fit into the iMac vanilla thermal system, so maybe a redesign of that is on its way or an entirely new Mac desktop might see the light this fall?

If "Pro" takes on its usual Apple euphemism of late of "even more expensive" then this may be just iMac Pro and Mac Pro. Where the iMac Pro gets 16GB of HBM ( or 8GB GDDR6 ) and the "more Pro" (more expensive) Mac Pro gets 32GB of HBM ( or 16GB GDDR6).

I think this is a bit large to go into a iMac ( growing the GPU footprint on the same size of the board as the PowerSupply. That is somewhat capped in the iMac Pro). I'm a bit skeptical that this is going anywhere near a regular iMac. ( and some "gamer" Mac desktop ... even more skeptical. )

 

[high heaven]

Uh, the Vega II in the Mac Pro is using the CDNA architecture already. What do you mean?

CDNA based GPU is not even announced yet. Vega II is GCN based GPU.

 

[MisterAndrew]

That's interesting. 4 new Apple GPUs. I wonder if they'll give the Mac Pro a mild refresh later this year as an 8,1 or just release new graphics cards.

 

[eflx]

Oooppss lol my mistake, mixed up GCN

 

[high heaven]

Because CDNA is meant for datacenter work. Considering Apple’s primary markets for pro devices are things like video editing, animation, 3D modeling and other graphical work, the graphics based architecture makes more sense than the number cruncher. CDNA is more akin to Tesla on Nvidia’s side. Apple would always pick the Quadra, FirePro and now Radeon Pro (RDNA) GPUs, not something like CDNA; which won’t even have video output in most cards it will come in I’m guessing

it seems you missed some points.

RDNA 2.0 aimed only for gaming purposes. You might say what about RDNA based GPU like RX 5500M? That's because the first-gen RDNA architecture is still using GCN codes. RDNA 2.0 is a gaming architecture based on what AMD said.

Later this year, AMD will divide all GPU into either RDNA 2.0 or CDNA. I highly doubt that Apple is interested in gaming architecture.

 

[casperes1996]

it seems you missed some points.

RDNA 2.0 aimed only for gaming purposes. You might say what about RDNA based GPU like RX 5500M? That's because the first-gen RDNA architecture is still using GCN codes. RDNA 2.0 is a gaming architecture based on what AMD said.

Later this year, AMD will divide all GPU into either RDNA 2.0 or CDNA. I highly doubt that Apple is interested in gaming architecture.

What does "gamin architecture" mean to you? Most GPU designs throughout the past two decades have been "gaming architectures". It just so happens that there's a general purpose aspect to that, and that a lot of what you need for games also applies to other fields. RDNA 2 will have a media block for video and shader hardware that will run FP32 well, which is the essence of GPGPU.
RDNA 2 may be first and foremost a "gaming architecture", but a "gaming architecture" is also a general use architecture. CDNA will be datacentre oriented with anywhere between all and some of the general video hardware removed, perhaps even unable to accelerate display output at all.

Until we know more about the architectures in detail it's hard to say much more than that, but thinking that RDNA 2 will exclusively be good for games is false reasoning. It's a general purpose GPU.

 

[high heaven]

What does "gamin architecture" mean to you? Most GPU designs throughout the past two decades have been "gaming architectures". It just so happens that there's a general purpose aspect to that, and that a lot of what you need for games also applies to other fields. RDNA 2 will have a media block for video and shader hardware that will run FP32 well, which is the essence of GPGPU.
RDNA 2 may be first and foremost a "gaming architecture", but a "gaming architecture" is also a general use architecture. CDNA will be datacentre oriented with anywhere between all and some of the general video hardware removed, perhaps even unable to accelerate display output at all.

Until we know more about the architectures in detail it's hard to say much more than that, but thinking that RDNA 2 will exclusively be good for games is false reasoning. It's a general purpose GPU.

You are miss understanding.

AMD has been using GCN or general-purpose architecture for several years. That's why gaming graphic cards were also used for workstation or data center. Later this year, all GPUs will be divided into RDNA and CDNA.

And how do you even know CDNA isn't for workstation GPU?

 

[casperes1996]

You are miss understanding.

AMD has been using GCN or general-purpose architecture for several years. That's why gaming graphic cards were also used for workstation or data center. Later this year, all GPUs will be divided into RDNA and CDNA.

View attachment 918667

And how do you even know CDNA isn't for workstation GPU?

I think we’ll just have to wait and see what happens here.
If we look at your Venn Diagram there You presented you can see that there’s some overlap in the middle between RDNA and CDNA. Now if that overlap really goes both ways, There may well be CDNA workstation cards, but the majority of workstation workloads have more in common with what a gaming card needs to do than what a datacenter card needs to do, though high TFLOPs is of course essential.
AMD’s marketing there made me laugh a bit though, since TFLOPs stands for “Terra Floating Operations Per Second” and they wrote TFLOPS/second.... Per second per second? Hehe.

But really what’s most applicable to what Apple usually wants from a GPU is fairly close to the middle of that Venn Diagram.

I don’t know any more details about the specific hardware architecture choices than anyone else. But I do know what’s required from the hardware for certain tasks - I’m not a GPU programming specialist, but I have worked with Metal code and a tiny bit of OpenCL

 

[theorist9]

Because CDNA is meant for datacenter work. Considering Apple’s primary markets for pro devices are things like video editing, animation, 3D modeling and other graphical work, the graphics based architecture makes more sense than the number cruncher. CDNA is more akin to Tesla on Nvidia’s side. Apple would always pick the Quadra, FirePro and now Radeon Pro (RDNA) GPUs, not something like CDNA; which won’t even have video output in most cards it will come in I’m guessing

Agreed, the principal market seems to be for graphical work. However, back when I was a grad student, the previous-gen Mac Pros were often used in physics and physical chemistry research for calculation work that could be done locally, and for prototyping programs for the cluster. I don't know whether the new Mac Pro has priced itself out of that use.

 

[casperes1996]

Agreed, the principal market seems to be for graphical work. However, back when I was a grad student, the previous-gen Mac Pros were often used in physics and physical chemistry research for calculation work that could be done locally, and for prototyping programs for the cluster. I don't know whether the new Mac Pro has priced itself out of that use.

Well it would probably be more rare to see them in those use cases than it used to be, but I think that with academic discount it's still a possibility.

And of course, my doubts about a CDNA GPU being offered in any of the standard configurations doesn't mean you wouldn't be able to put one into the Mac Pro; I very much think it will be supported in macOS. But it would probably be more of a secondary GPU if it would be a thing, since my understanding is still that most if not all CDNA GPUs will come without video output capability which Apple wouldn't go for at all. And to my knowledge there's no iGPU in any of the Xeons in the Mac Pro to at least drive basic output

 

[OkiRun]

Well it would probably be more rare to see them in those use cases than it used to be, but I think that with academic discount it's still a possibility.

And of course, my doubts about a CDNA GPU being offered in any of the standard configurations doesn't mean you wouldn't be able to put one into the Mac Pro; I very much think it will be supported in macOS. But it would probably be more of a secondary GPU if it would be a thing, since my understanding is still that most if not all CDNA GPUs will come without video output capability which Apple wouldn't go for at all. And to my knowledge there's no iGPU in any of the Xeons in the Mac Pro to at least drive basic output

Do you think Apple could begin to pull out specialty MPX Modules in the future? The Afterburner Card is an example for video editing Pro Res. Could they have an accelerator MPX for Gaming, and an accelerator MPX for Number Crunching?

 

[theorist9]

Well it would probably be more rare to see them in those use cases than it used to be, but I think that with academic discount it's still a possibility.

And of course, my doubts about a CDNA GPU being offered in any of the standard configurations doesn't mean you wouldn't be able to put one into the Mac Pro; I very much think it will be supported in macOS. But it would probably be more of a secondary GPU if it would be a thing, since my understanding is still that most if not all CDNA GPUs will come without video output capability which Apple wouldn't go for at all. And to my knowledge there's no iGPU in any of the Xeons in the Mac Pro to at least drive basic output

Also agreed. But the way I'd frame it is that the Mac Pro, because of its multiple-GPU capability, is particularly well-suited to accommodate speciality (secondary) cards that lack a video out.

Given this, I think the main barrier to making it appealing for number-crunching applications isn't that Apple would refuse to offer non-video CDNA GPU options per se; I don't see any reason they'd be reluctant to offer that as a secondary GPU. Indeed, they've at least acknowledged scientific use of the machine in their marketing (the published benchmarks mention its Matlab performance for "simulation of dynamical systems"); and offering a CDNA GPU would bolster that. Rather, the barrier is the limitation that Apple would place on the nature of those cards -- specifically that Apple (at least currently) refuses to partner with NVIDIA (or maybe it's mutual--I don't know).

 

[mattspace]

I highly doubt that Apple is interested in gaming architecture.

...aside from the whole "Apple pro GPUs are just gaming cards with extra VRAM and a "Pro" sticker" thing.

 

[deconstruct60]

That's interesting. 4 new Apple GPUs. I wonder if they'll give the Mac Pro a mild refresh later this year as an 8,1 or just release new graphics cards.

There is nothing substantive to do a "mild refresh" too. Bumping the SSD capacity to a capacity can already buy is a bit of a stretch to label 'mild'. ( Yes Apple is engaged in more than suspect behavior of labeling the Mac Mini as 'new' when it is not. The last time they tried to do that for the Mac Pro some folks turned them into the FTC. ) A new GPU card? Pragmatically they already did that with the W5700X. A new module isn't a new system; again 'mild' is an overstatement. The system already could take a new module.

CPU wise there is nothing. There is some chance of a new CPU in this class later in the year but all of the alternatives require a new logic board. That would probably span past "mild" ( unless anchored on the case externals as being the primary aspect of the system. ). New CPU more likely heading to the iMac Pro first. ( decent chance either Intel cranks out a Xeon W-2300 (ice lake ) or switch to AMD later in year ). Probably 2021 (or later) for something for the Mac Pro as an overall system upgrade.

Navi + HBM + CPU with PCI-e v4 would be a substantive bump for the iMac Pro.