Lets talk iGPU

[Danny82]

Not being hopeful of iGPU reaching the level of dGPU as leak have shown that Big Sur does have Navi31 driver.. so I believe mbp16, iMac and mac pro will still be getting and dGPU..

So let's talk iGPU, this will only compete again iGPU and not dGPU.. Apple have just licenced Imagination again and Imagination new X1 chip seems to be able to run at 2tflops.. if apple could do its magic and make it run at 2.5tflops for iGPU I will be really happy.. and seems that in some ways it could also be able to run low level ray tracing.. awesome.. next year Imagination will be coming out a hardware base ray tracing and if Apple is able to put off an iGPU running hardware base ray tracing by end of next year.. I think many would be happy, as many have been living the intel iris life anyway

 

[cmaier]

I think apple’s will run much faster than imagination’s. They didn’t license a particular design, just intellectual property rights in general, so they can do whatever design they want, just like the CPU.

 

[Janichsan]

Not being hopeful of iGPU reaching the level of dGPU as leak have shown that Big Sur does have Navi31 driver.. so I believe mbp16, iMac and mac pro will still be getting and dGPU..

You do remember that Big Sur will still be running on as of yet unreleased Intel Macs? Deducing anything about the capabilities of Apple's own GPUs from the presence of drivers in Big Sur is pretty much pointless.

 

[Danny82]

You do remember that Big Sur will still be running on as of yet unreleased Intel Macs? Deducing anything about the capabilities of Apple's own GPUs from the presence of drivers in Big Sur is pretty much pointless.

Agree that the navi will and maybe support the intel macs.. but cant imagine Apple suddenly releasing a dGPU as great as the one from Nvidia and AMD.. and definitely not an iGPU that competes with dGPU.. would probably take them a few more years.. of course that being said, it would be nice if they can come out their own dGPU so soon

Anyhow, if the iGPU is 2x more powerful than the one on crappy intel will be most welcome

 

[UltimateSyn]

This has been discussed in a few different threads. Apple GPUs will be integrated into the SoC. No dGPUs or separate VRAM.

"And to know if a GPU needs to be treated as integrated or discrete, use the isLowPower API. Note that for Apple GPUs isLowPower returns False, which means that you should treat these GPUs in a similar way as discrete GPUs. This is because the performance characteristics of Apple GPUs are in line with discrete ones, not the integrated ones. Despite the property name though, Apple GPUs are also way, way more power-effficient than both integrated and discrete GPUs."

Bring your Metal app to Apple silicon Macs - WWDC20 - Videos - Apple Developer

Meet the Tile Based Deferred Rendering (TBDR) GPU architecture for Apple silicon Macs — the heart of your Metal app or game's graphics...

developer.apple.com

It sounds like Apple GPUs are just going to be part of the SoC with the CPU. Think iPad chips with more cores, clocked up with the greater thermal headroom, and now on 5nm die. They're going to scream. No eGPU or dGPU needed.

"Intel-based Macs contain a multi-core CPU and many have a discrete GPU ... Machines with a discrete GPU have separate memory for the CPU and GPU. Now, the new Apple Silicon Macs combine all these components into a single system on a chip, or SoC. Building everything into one chip gives the system a unified memory architecture. This means that the CPU and GPU are working over the same memory."

Explore the new system architecture of Apple silicon Macs - WWDC20 - Videos - Apple Developer

Discover how Macs with Apple silicon will deliver modern advantages using Apple's System-on-Chip (SoC) architecture. Leveraging a unified...

developer.apple.com

 

[Janichsan]

but cant imagine Apple suddenly releasing a dGPU as great as the one from Nvidia and AMD.. and definitely not an iGPU that competes with dGPU..

Neither can I, but Apple clearly thinks they can: when querying the GPU on code-level, the Apple GPUs will report as being as powerful as dGPUs.

 

[Boil]

So inbound 500W APUzilla for forthcoming arm64-based Mac Cube Pro...?!?

32 P cores / 16 E cores / 64 GPU cores / 32GB HBM2e UMA

 

[Danny82]

Neither can I, but Apple clearly thinks they can: when querying the GPU on code-level, the Apple GPUs will report as being as powerful as dGPUs.

Awesome, thanks for the information.. so looking forward to the release of the new arm macbook now

 

[leman]

Agree that the navi will and maybe support the intel macs.. but cant imagine Apple suddenly releasing a dGPU as great as the one from Nvidia and AMD.. and definitely not an iGPU that competes with dGPU..

I doubt they will release a dGPU. But their I GPUs are competitive with dGPUs. iPad Pro running at 10W or so is more or less the same as an Nvidia 1050. Apple GPUs have completely different architecture and operate differently. They are less reliant on memory bandwidth. And besides, there is nothing preventing an iGPU from having a fast memory interface. Depends what kind of system memory you choose.

P.S. Apple Metal supports ray tracing.

 

[Janichsan]

P.S. Apple Metal supports ray tracing.

Yeah, but Nvidia GPUs have dedicated cores for raytracing. Apple's - as far as we know - don't.

 

[AVonGauss]

Deleted.

 

[PortoMavericks]

This has been discussed in a few different threads. Apple GPUs will be integrated into the SoC. No dGPUs or separate VRAM.

Craig said something like the system having an unified memory architecture. If that’s the case, it’ll suck for gaming with DDR4 for the GPU.

 

[Zackmd1]

Craig said something like the system having an unified memory architecture. If that’s the case, it’ll suck for gaming with DDR4 for the GPU.

What makes you think Apple Silicone Macs will have ddr4? LPDDR5 exists which will be the likely choice for mobile devices. Desktops will likely come with dGPU options.

 

[Joelist]

Just remember that the GPU in the A12X was already at 7 cores and has the same performance as an Xbox One S. Apple has already said that we have not seen the Apple Silicon series for Macs yet, and also they will have more thermal headroom to work with than in an iPad. I expect something with much faster cores than we are currently seeing and more of them. Also it will have a VERY advanced memory architecture and probably a lot of RAM so all the SOC components can blaze away without needing to use paging except in extreme scenarios.

 

[Rastafabi]

My guess is the Mac Pro will still use separate graphics processors, however most if not all other devices will use Apple's iGPU equivalent.

I don't. I think by the time apple will release a new Mac Pro apple's GPU will be advanced enough to do the work to output to several 6K/8K… monitors. Those graphics might not have the same compute power like for instance an Vega VII has, but they won't need to. Even today the Mac Pro does not have all those graphics performance to do graphics but rather it is the current goto standard for efficient large scale computing. However graphics card are already progressing past that. NVIDIA Tesla comes to mind as well as those countless mining cards which do not even over monitor ports. It's better to think about those GPUs apple offers in the Mac Pro today as compute units. Apple already did showcase what kind of compute cards they can offer with the Apple Afterburner card. Thus I'm certain that within the Apple environment GPUs as we now them are redundant once the Apple Silicon transition has been completed.

 

[macsplusmacs]

Maybe apple and NVidia and apple kiss and make up and work on co-developing drivers for a new line of GPUs (for all platforms) by the time the 2nd gen ARM macs.

Maybe Radeon deal was only for intel based macs.

Or, and I hope its this...

Maybe apple can fab up entire sheets of GPUs that connect to the SOC via some super bandwidth highway and we get the best of both worlds.

 

[leman]

Maybe apple and NVidia and apple kiss and make up and work on co-developing drivers for a new line of GPUs (for all platforms) by the time the 2nd gen ARM macs.

There is no way that Apple would allow Nvidia to have CUDA on Mac. Apple is investing heavily to have their own GPGPU ecosystem and CUDA is poison to that.

I am a big supporter of the open standards in computing, but in the meantime, I am leaning more and more towards platform-specific APIs for GPUs and ML accelerators. The open standard simply cannot reasonably account for all the differences and special features of different chips. It's best to have vendor-specific low-level APIs and then higher-level unified frameworks that would provide the platform-independent layers (if one needs).

 

[macsplusmacs]

There is no way that Apple would allow Nvidia to have CUDA on Mac. Apple is investing heavily to have their own GPGPU ecosystem and CUDA is poison to that.

Agreed. Apple would write the Metal drivers and keep feature parity outside of CUDA feature parity in check.

 

[pldelisle]

There is no way that Apple would allow Nvidia to have CUDA on Mac. Apple is investing heavily to have their own GPGPU ecosystem and CUDA is poison to that.

I am a big supporter of the open standards in computing, but in the meantime, I am leaning more and more towards platform-specific APIs for GPUs and ML accelerators. The open standard simply cannot reasonably account for all the differences and special features of different chips. It's best to have vendor-specific low-level APIs and then higher-level unified frameworks that would provide the platform-independent layers (if one needs).

You nailed it. As a software engineer specialized in parallel/distributed computing and ML, I totally agree with this.

 

[PortoMavericks]

What makes you think Apple Silicone Macs will have ddr4? LPDDR5 exists which will be the likely choice for mobile devices. Desktops will likely come with dGPU options.

LPDDR5 is equally slow for VRAM application.

 

[pldelisle]

LPDDR5 is equally slow for VRAM application.

If you rely less on memory, it’s no big deal.

 

[PortoMavericks]

If you rely less on memory, it’s no big deal.

I don't think the gaming industry will spend time optimizing for the Apple Silicon on complex games.

They're lazy, they want a lot of bandwidth and capacity. You'll see just that on Series X and PS5.

 

[pldelisle]

I don't think the gaming industry will spend time optimizing for the Apple Silicon on complex games.

They're lazy, they want a lot of bandwidth and capacity. You'll see just that on Series X and PS5.

They don’t have too. Metal and the chip’s architecture itself do it.

 

[macsplusmacs]

Unity > Checkbox > Apple Silicon > Optimized

 

[Zackmd1]

LPDDR5 is equally slow for VRAM application.

It is better then what the intel and AMD iGPUs use which is usually LPDDR4... It is never going to be as good as dedicated memory yes but the key is it doesn't have to be. Speaking of that, the dev kit already beats the ryzen 4500U in graphics performance and that is with an A12Z in a highly un-optimized test.