New AMD Radeon PRO W6000X Series GPUs Bring Groundbreaking High-Performance AMD RDNA 2 Architecture to Mac Pro

[AleksandarK]

AMD today announced availability of the new AMD Radeon PRO W6000X series GPUs for Mac Pro. The new GPU product line delivers exceptional performance and incredible visual fidelity to power a wide variety of demanding professional applications and workloads, including 3D rendering, 8K video compositing, color correction and more.

Built on groundbreaking AMD RDNA 2 architecture, AMD Infinity Cache and other advanced technologies, the new workstation graphics line-up includes the AMD Radeon PRO W6900X and AMD Radeon PRO W6800X GPUs. Mac Pro users also have the option of choosing the AMD Radeon PRO W6800X Duo graphics card, a dual-GPU configuration that leverages high-speed AMD Infinity Fabric interconnect technology to deliver outstanding levels of compute performance.



"We developed the AMD Radeon PRO W6000X series GPUs to unleash professionals' creativity and help them bring more complex and compute-intensive projects to life, from animating 3D film assets to compositing 8K scenes to game development," said Scott Herkelman, corporate vice president and general manager, Graphics Business Unit at AMD. "The new AMD Radeon PRO W6000X series is packed with remarkable energy efficiency, enhanced compute units and a new visual pipeline, enabling Mac Pro users to do more in less time across a broad range of pro applications."

Key capabilities and features of AMD Radeon PRO W6000X series GPUs include:

• Award-Winning AMD RDNA 2 Architecture - Built on the 7 nm manufacturing process, AMD RDNA 2 architecture offers an array of advanced features elevating professional graphics to new levels of performance and efficiency.
• High-speed GDDR6 memory - Up to 64 GB of GDDR6 memory with up to 512 GB/s bandwidth provides ultra-fast transfer speeds to power data-intensive professional applications.
• AMD Infinity Cache - Up to 256 MB (total) of last-level data cache integrated on the GPU die is designed to reduce latency and power consumption.
• AMD Infinity Fabric - Provides a high-bandwidth, low latency, direct connection between the local AMD GPUs, enabling high speed GPU-to-GPU communications designed to satisfy today's creative workloads.

View at TechPowerUp Main Site

 

[defaultluser]

Exciting DoA Mac shit.

The only planned Apple Silicon systems that will support external GPUs are Mac Pro , and Apple can easily supply that tiny amount of demand themselves (I'm sure each server will come with minimal IGP, to cover cloud computers / webservers )

Apple is Planning to Build Its Own GPUs, Too, But Playing Quiet for Now

Apple may have plans to replace its AMD GPU silicon, not just its Intel CPU cores -- but it hasn't spoken about them to nearly the same extent.

www.extremetech.com

Sorry AMD, you (and all of your external Thunderbolt docks) are out of a job.

 

[dragontamer5788]

I do wonder if Apple's "Metal" API is working fine. AMD put a lot of work into ROCm / HIP, and most of that has gone into clang / llvm, which Apple had a part in creating. So I wonder if the Metal compiler is using those clang bits.

It always seems odd to me that the Apple Pro-series might be the best AMD-GPU compute systems on the market right now. I mean, I know that CDNA exists but there's no largescale manufacturer of ready-to-use computers like Apple (or NVidia DGX). Just seems like a market niche that should be explored.

 

[AnarchoPrimitiv]

Exciting DoA Mac shit.

The only planned Apple Silicon systems that will support external GPUs are Mac Pro , and Apple can easily supply that tiny amount of demand themselves (I'm sure each server will come with minimal IGP, to cover cloud computers / webservers )

Apple is Planning to Build Its Own GPUs, Too, But Playing Quiet for Now

Apple may have plans to replace its AMD GPU silicon, not just its Intel CPU cores -- but it hasn't spoken about them to nearly the same extent.

www.extremetech.com

Sorry AMD, you (and all of your external Thunderbolt docks) are out of a job.

"All of your external Thunderbolt docks"? What?

 

[Steevo]

"All of your external Thunderbolt docks"? What?

Maybe English isn’t the native language of that person.

If Intel can’t make a GPU yet I highly doubt Apple will be able to magically pull one off, and their “CPUs” are slightly better than copy and paste ARM designs, (look at the cache on the M1, and remember it’s power use and actual ability) they get the benefit in tuning their hardware and software together due to a walled garden approach. It’s like saying the APU in the Xbox or PS5 is lightning fast when really it’s the low overhead and closed environment and software tuned for that specific hardware.

Maybe Raja will make something great for them after the smoke gets let out of Intels efforts.

 

[dragontamer5788]

Maybe English isn’t the native language of that person.

If Intel can’t make a GPU yet I highly doubt Apple will be able to magically pull one off, and their “CPUs” are slightly better than copy and paste ARM designs, (look at the cache on the M1, and remember it’s power use and actual ability) they get the benefit in tuning their hardware and software together due to a walled garden approach. It’s like saying the APU in the Xbox or PS5 is lightning fast when really it’s the low overhead and closed environment and software tuned for that specific hardware.

Maybe Raja will make something great for them after the smoke gets let out of Intels efforts.

Intel is having issues as a company right now. Hopefully they figure things out, but... I don't think Intel having issues necessarily means that Apple will have issues.

 

[zlobby]

Intel is having issues as a company right now. Hopefully they figure things out, but... I don't think Intel having issues necessarily means that Apple will have issues.

Oh, Apple have lots of issues too. However, somehow hundreds of billions in cash reserve tend to make them not giving a flying frack.

 

[Fouquin]

If Intel can’t make a GPU yet I highly doubt Apple will be able to magically pull one off

The TBDR GPU in M1 is an in-house design. It uses some fundamentals of PowerVR AXT but is otherwise unique.

 

[Darmok N Jalad]

Maybe English isn’t the native language of that person.

If Intel can’t make a GPU yet I highly doubt Apple will be able to magically pull one off, and their “CPUs” are slightly better than copy and paste ARM designs, (look at the cache on the M1, and remember it’s power use and actual ability) they get the benefit in tuning their hardware and software together due to a walled garden approach. It’s like saying the APU in the Xbox or PS5 is lightning fast when really it’s the low overhead and closed environment and software tuned for that specific hardware.

Maybe Raja will make something great for them after the smoke gets let out of Intels efforts.

I dunno, the M1's GPU is pretty decent at running my DXO PureRAW files. It can run a 20MP file in around 25 seconds, which is parity with the R9 380 in my old Mac Pro. I know, I know, the R9 380 is old, but the M1 is running in a passively cooled MBA chassis, and is sharing resources with the CPU. Macs aren't for gaming, at least not the gaming you get on PCs. They haven't been gaming machines for a long time. However, for compute tasks, the GPU does surprisingly well. They also have the Afterburner Accelerator card for the newest Mac Pro, which can handle 3 8K RAW streams or 12 4K streams simultaneously, in real time. Apple has cash and then some, but they won't threaten any market outside of their own.

 

[Valantar]

Kind of surprised to see these using RDNA tbh. Would have expected CDNA to be a better fit - though maybe that lacks parts of the graphics pipeline that this needs?

Exciting DoA Mac shit.

The only planned Apple Silicon systems that will support external GPUs are Mac Pro , and Apple can easily supply that tiny amount of demand themselves (I'm sure each server will come with minimal IGP, to cover cloud computers / webservers )

... but this is for the Mac Pro. Specifically. It literally only fits in the Mac Pro. And it definitely won't work as an eGPU unless Apple makes a bespoke chassis with its weird extended PCIe slot. And the Apple Silicon MP isn't likely to arrive for another year (and who knows what expansion card support it might have?).

But even without that, this is likely to perform well, and do its job nicely for the couple of years until the people needing/wanting something like this are likely to upgrade anyway. And it'll live on just fine when they sell their systems afterwards.

 

[medi01]

Anything Apple is ground breaking in some way.
I recall when they were waving slow power hungry IBM chips it was also super cool, in ways.

 

[defaultluser]

Kind of surprised to see these using RDNA tbh. Would have expected CDNA to be a better fit - though maybe that lacks parts of the graphics pipeline that this needs?


... but this is for the Mac Pro. Specifically. It literally only fits in the Mac Pro. And it definitely won't work as an eGPU unless Apple makes a bespoke chassis with its weird extended PCIe slot. And the Apple Silicon MP isn't likely to arrive for another year (and who knows what expansion card support it might have?).

But even without that, this is likely to perform well, and do its job nicely for the couple of years until the people needing/wanting something like this are likely to upgrade anyway. And it'll live on just fine when they sell their systems afterwards.

How long do you expect Apple to continue offering AMD GPUs when they can make their own (for much higher margin?)

They already killed off external gpus on m1 (so, 1/3 of the use case for Thunderbolt 3 is already gone), so how much further do you expect they will take things with these M2 Mac Pros?

Will they kill off eGPU universally? or Will apple take it even further (with an Apple-exclusive discrete GPU)? The initial outlook from m1 doesn't make it look good for either one of these features surviving the ARM transition.

 

[Valantar]

How long do you expect Apple to continue offering AMD GPUs when they can make their own (for much higher margin?)

They already killed off external gpus on m1 (so, 1/3 of the use case for Thunderbolt 3 is already gone), so how much further do you expect they will take things with these M2 Mac Pros?

Will they kill off eGPU universally? or Will apple take it even further (with an Apple-exclusive discrete GPU)? The initial outlook from m1 doesn't make it look good for either one of these features surviving the ARM transition.

Until they start selling Apple Silicon Mac Pros at the very least. So for at least another year - but likely longer, IMO.

"Killed off" is weird phrasing. "Haven't enabled", possibly "haven't enabled yet". Given their love for Thunderbolt and the vast install base for various TB AIC enclosures with professional mac users, I would be surprised if support wasn't added down the line - but that would of course mean that third party vendors would need to be willing (and able) to create drivers for ARM MacOS.

It's entirely possible that Apple will make a discrete GPU-like compute/rendering/whatever accelerator, absolutely. They have a very good GPU arch, after all. But until we see that, I see no reason to expect that to be the only solution. That would definitely piss off a lot of their pro users, at least. Not that that's something Apple hasn't done before, but they're not very consistent.

 

[claes]

Apple’s silicon isn’t ready for professionals yet. While we’ll likely see an AS Mac Pro in the next year, the Intel ones will see support for quite awhile. From what I’ve read Audio software is being rewritten for AS but it’ll be awhile for video and scientific software. M1 is just too buggy at this point (aside from being an ultraportable SoC that’s not designed for expansion — it’s basically a proof of concept).

 

[Steevo]

The TBDR GPU in M1 is an in-house design. It uses some fundamentals of PowerVR AXT but is otherwise unique.

Vega is twice as fast as the “new” Intel on die graphics which were supposed to be great, but aged really quick, just like Apple hardware will when placed next to actual high power hardware.

General compute will always be slower at some tasks, but at least you don’t have to buy a new machine for new features to work.

 

[Fouquin]

Vega is twice as fast as the “new” Intel on die graphics which were supposed to be great, but aged really quick, just like Apple hardware will when placed next to actual high power hardware.

General compute will always be slower at some tasks, but at least you don’t have to buy a new machine for new features to work.

Intel's GPU has literally nothing to do with Apple's. For one, Apple's GPU isn't even a raster engine. Secondly, they don't run any of the same APIs. There is no parallel you can draw between them to judge any performance at any point.

 

[medi01]

GROUNDBREAKING *

* for Apple

 

[Aquinus]

Kind of surprised to see these using RDNA tbh. Would have expected CDNA to be a better fit - though maybe that lacks parts of the graphics pipeline that this needs?

Apple already has drivers for RDNA based chips and is fully supported with Metal on x86. I suspect that when CDNA comes around, that'll be when we'll start seeing drivers for these GPUs on Apple's ARM-based chips. Honestly the 5600M in my MBP works pretty well for a 50w mobile part.

 

[Valantar]

Apple already has drivers for RDNA based chips and is fully supported with Metal on x86. I suspect that when CDNA comes around, that'll be when we'll start seeing drivers for these GPUs on Apple's ARM-based chips. Honestly the 5600M in my MBP works pretty well for a 50w mobile part.

Apple's implementations of AMD GPUs have been pretty good - that HBM-equipped Vega 12 they had was really cool. And they're good at tuning them for efficiency, unlike what AMD has been doing on the desktop in recent years (guess that's what comes from a non-competitive market, higher margins, and only selling 'premium' products). CDNA has come around though - the Instinct MI100 launched late last year. But that's a pure HPC/accelerator card with a gargantuan die, and it doesn't have a real graphics pipeline at all, so it might be as simple as there not being any CDNA designs suitable for applications like this. Though I guess if they wanted to they could support those as accelerators with a separate GPU. Can't really imagine the market for that is huge though.

 

[Aquinus]

Apple's implementations of AMD GPUs have been pretty good - that HBM-equipped Vega 12 they had was really cool. And they're good at tuning them for efficiency, unlike what AMD has been doing on the desktop in recent years (guess that's what comes from a non-competitive market, higher margins, and only selling 'premium' products). CDNA has come around though - the Instinct MI100 launched late last year. But that's a pure HPC/accelerator card with a gargantuan die, and it doesn't have a real graphics pipeline at all, so it might be as simple as there not being any CDNA designs suitable for applications like this. Though I guess if they wanted to they could support those as accelerators with a separate GPU. Can't really imagine the market for that is huge though.

I suspect that Apple would prefer to keep everything in house if they can. If we see movement on this, it would probably be with an ARM-based Mac Pro. If the support is there, I'd expect there to be eGPU support to come shortly after for certain GPUs. If Apple doesn't support that level of upgradability with the Mac Pro lineup, then I doubt we'll ever see support for it. The real question is if Apple is up to the task with regard to making really powerful GPUs. Clearly they're not bad at making iGPUs. I'd still take the Radeon Pro 5600m in my MBP over the GPU in the M1 chip any day though.

Personally, I'd like to see Apple continue to work with AMD on GPU support. I just don't know if the market is there for Apple to continue pursuing it.

 

[Valantar]

I suspect that Apple would prefer to keep everything in house if they can. If we see movement on this, it would probably be with an ARM-based Mac Pro. If the support is there, I'd expect there to be eGPU support to come shortly after for certain GPUs. If Apple doesn't support that level of upgradability with the Mac Pro lineup, then I doubt we'll ever see support for it. The real question is if Apple is up to the task with regard to making really powerful GPUs. Clearly they're not bad at making iGPUs. I'd still take the Radeon Pro 5600m in my MBP over the GPU in the M1 chip any day though.

Personally, I'd like to see Apple continue to work with AMD on GPU support. I just don't know if the market is there for Apple to continue pursuing it.

Makes me wonder how much support is actually needed though. I mean, they can't lock down third party driver development entirely - that would tank peripheral sales - and they can't really control implementations of APIs like OpenCL (which they explicitly support AFAIK). So while they might never support these cards as GPUs, i.e. allow them to render real-time 3D graphics and the like (which would require Metal API access), I don't quite see how they could realistically stop them from working as accelerators in various applications. Still, I guess we won't know until the Apple Silicon Mac Pro arrives.

 

[Aquinus]

Makes me wonder how much support is actually needed though. I mean, they can't lock down third party driver development entirely - that would tank peripheral sales - and they can't really control implementations of APIs like OpenCL (which they explicitly support AFAIK). So while they might never support these cards as GPUs, i.e. allow them to render real-time 3D graphics and the like (which would require Metal API access), I don't quite see how they could realistically stop them from working as accelerators in various applications. Still, I guess we won't know until the Apple Silicon Mac Pro arrives.

Yeah, they can't stop them, but it tends to help when Apple supports your efforts. I believe the main issue right now is the architecture differences between x86 and ARM, so it's not as simple as just recompiling the driver for ARM. At least in the Linux kernel, there tends to be preprocessor directives for things that use something architecture specific. I don't know how much AMD has tuned their drivers for x86, it could be a lot, or that the current state is just incredibly underperforming compared to where it needs to be.

 

[Darmok N Jalad]

Yeah, they can't stop them, but it tends to help when Apple supports your efforts. I believe the main issue right now is the architecture differences between x86 and ARM, so it's not as simple as just recompiling the driver for ARM. At least in the Linux kernel, there tends to be preprocessor directives for things that use something architecture specific. I don't know how much AMD has tuned their drivers for x86, it could be a lot, or that the current state is just incredibly underperforming compared to where it needs to be.

If I remember right, PowerPC also required specific Apple-only GPUs. It wasn't until Intel-Macs that you could start using off-the-shelf GPUs, but they had to be GPU architectures supported by MacOS drivers, and the card had to have a standard EFI implementation. For example, most RX560s would work just fine in the MacPro, but there were certain manufacturer cards that wouldn't work at all. I bet a lot of work needs to be done to get GPUs working on ARM-Macs, as evident by the Mac Pro and 16" MBP not switching to AS yet. Both products have upgraded graphics, better than what's in the M1. We'll either get dGPU support, or whatever Apple has planned next is going to have a much faster Apple GPU. I wonder if Apple will actually make separate cards of their own for this market. With the state of the GPU market, it would actually be a great time to do it. They could charge Apple-like prices for upgrades, and it would probably still look like a relative bargain.

 

[Aquinus]

If I remember right, PowerPC also required specific Apple-only GPUs. It wasn't until Intel-Macs that you could start using off-the-shelf GPUs, but they had to be GPU architectures supported by MacOS drivers, and the card had to have a standard EFI implementation. For example, most RX560s would work just fine in the MacPro, but there were certain manufacturer cards that wouldn't work at all. I bet a lot of work needs to be done to get GPUs working on ARM-Macs, as evident by the Mac Pro and 16" MBP not switching to AS yet. Both products have upgraded graphics, better than what's in the M1. We'll either get dGPU support, or whatever Apple has planned next is going to have a much faster Apple GPU. I wonder if Apple will actually make separate cards of their own for this market. With the state of the GPU market, it would actually be a great time to do it. They could charge Apple-like prices for upgrades, and it would probably still look like a relative bargain.

I'd personally like it if Apple continued to collaborate with AMD. I think I've mentioned it before, but the Radeon Pro 5600m is the nicest mobile GPU I've ever used. It's the only GPU on the MBP lineup that can drive two 5k displays silently (thanks to the HBM2 memory,) and it's good enough for gaming (and by mobile standards, it's pretty darn good.) For me, that's the bar that's set for Apple with the upcoming 16" MBP with an AS chip. The M1 is almost on par with the performance of the CPU, but the GPU is a long way away from being at the same level.

If Apple releases an updated Mac Mini with the new AS chip they release for the 16", I might consider it. It needs to support at least 32GB of memory though.