Ray tracing on Apple Silicon?

[Appletoni]

I saw a lot of people including myself wishing to use Unreal Engine on Apple Silicon Mac natively but so far, they are not interested at all. Maybe because of the lawsuit between Epic and Apple? Hmmm... Someone mentioned that Apple Silicon's GPU sucks because it has no capability to run ray tracing features on Unreal engine and therefore, there won't be a native support for AS Macs. Well, not an official statement but lack of ray tracing and unreal engine support is quite disappointing so far. I dont know if Apple is gonna make a GPU for ray tracing since they weren't interested in gaming industry but for 3D, it can be useful.

Ray Tracing with Metal - WWDC19 - Videos - Apple Developer

Metal Performance Shaders (MPS) harness the massive parallelism of the GPU to dramatically accelerate calculations at the heart of modern...

developer.apple.com

Enhance your app with Metal ray tracing - WWDC21 - Videos - Apple Developer

Achieve photorealistic 3D scenes in your apps and games through ray tracing, a core part of the Metal graphics framework and Shading...

developer.apple.com

Explore hybrid rendering with Metal ray tracing - WWDC21 - Videos - Apple Developer

Discover how you can combine ray tracing with your rasterization engine to implement simplified graphics techniques and elevate visuals...

developer.apple.com

Discover ray tracing with Metal - WWDC20 - Videos - Apple Developer

Achieve photorealistic 3D scenes in your apps and games through ray tracing — a core part of the Metal graphics framework and Shading...

developer.apple.com

WWDC18 - Videos - Apple Developer

developer.apple.com

I know that Apple had been working on ray tracing since 2018 but they still didnt add any ray tracing features so far. Some people say Apple Silicon chip does not have a dedicated ray tracing chip or the GPU isn't powerful enough to use ray tracing. Since Both Nvidia and AMD have ray tracing technologies, it's a huge question that Apple need to show or answer. Not just for gaming but also for 3D works.

Do you think Apple will add and support ray tracing with Apple Silicon chip?

I hope that Apple will fix it fast so we can have an amazing Raytracing.
It‘s bad to buy a device, for example a MacBook Pro without M1 Ultra and without an amazing Raytracing.

 

[diamond.g]

I hope that Apple will fix it fast so we can have an amazing Raytracing.
It‘s bad to buy a device, for example a MacBook Pro without M1 Ultra and without an amazing Raytracing.

Apple has Raytracing, it is just done with shaders/CPU instead of being all GPU driven. I don‘t think Apple would ever offer the M1 Ultra in a notebook (thermals and battery life), but I could be wrong.

 

[sunny5]

Apple has Raytracing, it is just done with shaders/CPU instead of being all GPU driven. I don‘t think Apple would ever offer the M1 Ultra in a notebook (thermals and battery life), but I could be wrong.

CPU itself is fine cause it's only 60W. But GPU's power consumption is 120W. Unless Apple want to make a powerful laptop despite the battery issue, it's possible but since they are limiting themselves for 100W, I guess it's not possible.

 

[l0stl0rd]

CPU itself is fine cause it's only 60W. But GPU's power consumption is 120W. Unless Apple want to make a powerful laptop despite the battery issue, it's possible but since they are limiting themselves for 100W, I guess it's not possible.

Yeah I don't think battery life has s on of their selling points.

I thing the 64 Ultra would throttle considering that you can make the M1 Max throttle if you put a lot of load on the GPU and CPU.

I guess if people want better they need to wait for the M2 Max.

 

[Boil]

I thing the 64 Ultra would throttle considering that you can make the M1 Max throttle if you put a lot of load on the GPU and CPU.

That is in the 14" & 16" MBP laptops; I feel the Mac Studio chassis is the perfect solution, it should allow the M1 Max SoC to reach full power without any thermal throttling issues...?

 

[l0stl0rd]

That is in the 14" & 16" MBP laptops; I feel the Mac Studio chassis is the perfect solution, it should allow the M1 Max SoC to reach full power without any thermal throttling issues...?

Yes I guess so, even hoping they will let the GPU run at a bit higher clockrate.

 

[Ethosik]

Ray tracing will come. It will be a standard feature set of all GPUs in the world. After that it will even come to all the iGPUs too.

Right now it is a new feature and very energy intensive. For heat management reasons it is not good to enable RT on a laptop or thin AIO desktop.

In a couple of generations computers will be powerful enough and efficient enough to do RT with a much smaller energy footprint.

Yep, I am surprised people don't know how Apple works at this point. While the haters will say "Apple is slow to introduce feature X", they do it in an organized way. Only a handful of games use RT, and a lot of people can't really tell the difference. Its getting more popular thanks to the new consoles. Apple will include it when its not a major performance hit and more popular.

 

[leman]

Yep, I am surprised people don't know how Apple works at this point. While the haters will say "Apple is slow to introduce feature X", they do it in an organized way. Only a handful of games use RT, and a lot of people can't really tell the difference. Its getting more popular thanks to the new consoles. Apple will include it when its not a major performance hit and more popular.

High-performance raytracing is mostly about reordering memory access to properly utilize massive memory and computational parallelism of modern GPUs. Nvidia has some magic sauce here (there were investigations in how their hardware works that concluded that RT cores are probably tightly integrated with texture units). AMD doesn’t - their “hardware RT” boils down to trivial fixed-function ray/box intersection, which is also probably why their RT is much slower in practice. Apple wants to have very programmable RT so they probably need a more general memory coalescing solution. But I’m sure they are working on it and I hope it can be shipped with M2 series.

 

[diamond.g]

High-performance raytracing is mostly about reordering memory access to properly utilize massive memory and computational parallelism of modern GPUs. Nvidia has some magic sauce here (there were investigations in how their hardware works that concluded that RT cores are probably tightly integrated with texture units). AMD doesn’t - their “hardware RT” boils down to trivial fixed-function ray/box intersection, which is also probably why their RT is much slower in practice. Apple wants to have very programmable RT so they probably need a more general memory coalescing solution. But I’m sure they are working on it and I hope it can be shipped with M2 series.

Not yet, or at least not the base M2. Lets see if they release a M2 Pro/Max in Q4 or if they will wait till Q4 of 2023 to update them.

 

[dugbug]

metal 3 brings improvements to raytracing performance for Apple Silicon. I haven't watched the session yet but it was talked about in the state of the union

Edit from Metal 3 improvements:

New Ray Tracing features​

The latest advancements in Metal Ray Tracing mean less GPU time is spent building acceleration structures, work like culling can move to the GPU to reduce CPU overhead, and both intersection and shading can be optimized with direct access to primitive data.

Metal Overview - Apple Developer

Metal powers hardware-accelerated graphics on Apple platforms by providing a low-overhead API, rich shading language, tight integration between graphics and compute, and an unparalleled suite of GPU profiling and debugging tools.

developer.apple.com

As mentioned, I have not yet watched any metal 3 sessions yet so thats all I know.

 

[l0stl0rd]

From their presentation, I guess there is hope.

 

[diamond.g]

View attachment 2015657
From their presentation, I guess there is hope.

Makes me wonder if the RT wasn't GPU driven before.

 

[deconstruct60]

CPU itself is fine cause it's only 60W. But GPU's power consumption is 120W. Unless Apple want to make a powerful laptop despite the battery issue, it's possible but since they are limiting themselves for 100W, I guess it's not possible.

The power consumption won't totally disappear, but it will get mitigated over time. So the "Max" class die will get batter at it when Apple goes to TSMC N3. They will likely crank up the 'horsepower' of the P cores and GPU cores while also growing the shared on die memory workspace space larger (more megabytes ) . Crank up the number of GPU cores.

All of that will make their Metal 3 based raytracing stuff rolling out now to just preform better. So will be able to do more with 80W than can do now. ( Metal 3 will bring some of that to the 'old' M1 Max ).

The sizable fab shrink after that would be perhaps be a time where some dedicated fixed function logic kicks in. Apple's ambitions in the AR/VR space would likely be were they probably have more intermediate term future need for fixed function logic rather than the laptops.

Metal 3 is suppose to have support for GPU hosted ray tracing hardware. But likely more so for the minimal support that AMD added to RNDA2 that Apple is "leaving on the floor unused" now. Again not completely skewed fixed function logic support. Currently Apple is shipping hardware whee ray trace logic isn't being used. Need to fix that first before moving on to potentially their own stuff later.

 

[deconstruct60]

View attachment 2015657
From their presentation, I guess there is hope.

The AMD RDNA2 GPUs that Apple is already shipping have unused ray trace logic. That is most likely just 'catch up' to hardware already in the field.

I wouldn't read lots of new RNDA3 AMD GPUs into that.

Metal 3 covers Apple GPUs , newer Intel iGPUs , and newer AMD GPUs. ( they are leaving older 3rd party GPUs behind but are still making contributions to the ones that are left . )

 

[l0stl0rd]

The AMD RDNA2 GPUs that Apple is already shipping have unused ray trace logic. That is most likely just 'catch up' to hardware already in the field.

I wouldn't read lots of new RNDA3 AMD GPUs into that.

Metal 3 covers Apple GPUs , newer Intel iGPUs , and newer AMD GPUs. ( they are leaving older 3rd party GPUs behind but are still making contributions to the ones that are left . )

Yes perhaps I guess we don’t know for sure till the M2 Pro/Max show.
Should they not have any RT hardware we will not see it till the M3 I bet.

It could also mean that their Metal Raytracing was running on the CPU and is now running on the GPU🤷‍♂️

 

[jmho]

There are different parts of Raytracing: the low level part of comparing millions of mathematical rays with various mathematical objects. Apple was definitely already doing this on the GPU.

There are higher level things like building and traversing the bounding volume hierarchy and various acceleration structures (that are used to help cull this massive list of ray intersections) that Apple potentially wasn't doing on the GPU, but will now in Metal 3.

 

[throAU]

There are different parts of Raytracing: the low level part of comparing millions of mathematical rays with various mathematical objects. Apple was definitely already doing this on the GPU.

There are higher level things like building and traversing the bounding volume hierarchy and various acceleration structures (that are used to help cull this massive list of ray intersections) that Apple potentially wasn't doing on the GPU, but will now in Metal 3.

For current implementations, to make it run fast enough there's also denoising and upscaling, which the Apple processor's neural engine capabilities can potentially be utilised for.

 

[Xiao_Xi]

There are different parts of Raytracing: the low level part of comparing millions of mathematical rays with various mathematical objects. Apple was definitely already doing this on the GPU.

For current implementations, to make it run fast enough there's also denoising and upscaling, which the Apple processor's neural engine capabilities can potentially be utilised for.

Does any software use it?

 

[jmho]

Does any software use it?

I don't think anyone is using the neural engine for de-noising and upscaling just yet, although maybe that's part of the new MetalFX library, or something that could be added to it in the future.

I also have no idea if anyone is actually using the Metal APIs for raytracing. I believe Blender is using its own custom compute kernels, which will likely be just as fast as the metal APIs until we get hardware.

 

[throAU]

Does any software use it?

No not that i am aware of yet, but that doesn't mean it can't be done.

Nvidia are using their AI cores on the RTX cards for upscaling (and denoising i believe?), Apple are likely able to do the same or at least similar.

 

[leman]

I don't think anyone is using the neural engine for de-noising and upscaling just yet, although maybe that's part of the new MetalFX library, or something that could be added to it in the future.

Looking at the API design, it's flexible enough that it could be moved to some other coprocessor in the future. That's the advantage of wrapping the APIs into a black box instead of letting the devs do things themselves I suppose.

 

[lcubed]

I don't think anyone is using the neural engine for de-noising and upscaling just yet, although maybe that's part of the new MetalFX library, or something that could be added to it in the future.

DXO's photo lab 5 deep prime uses the ANE for noise reduction. works amazingly well

 

[deconstruct60]

Yes perhaps I guess we don’t know for sure till the M2 Pro/Max show.
Should they not have any RT hardware we will not see it till the M3 I bet.

The M2 Pro/Max hardware isn't really the point. The Metal API really didn't have a way to semantically 'talk' about having GPU ray tracing capabilites at all. You couldn't say that the GPU did or did not have GPU ray tracing abilities or that you wanted use or not use them. If can't "talk" about it then can't do it. Whether the hardware has it or not immateral if can't express it in the code.

It could also mean that their Metal Raytracing was running on the CPU and is now running on the GPU🤷‍♂️

Apple enabled that back in 2019 with indirect command buffers. That isn't new. But it is more robust now.

The actual reveal in the Metal Ray tracing session was rather limited. Basically get to say

" Allow you to encode GPU work independently on the GPU

... icbDescriptor.supportRayTracing = true ... "

Maximize your Metal ray tracing performance - WWDC22 - Videos - Apple Developer

Learn how to simplify your ray tracing code and increase performance with the power of Metal 3. We'll explore the GPU debugging and...

developer.apple.com

Properly implemented performance shader compiler support toolchain for AMD RDNA2 should attempt now where using an optimized library call to use the hardware on the GPU. If not the compute still gets done on the GPU. All the indirect command buffer work is bing done on the GPU. Really up the opaque , "black box" of the Apple/AMD compilers and library code to gets some traction here with the actually fixed function calls or not. And if there is not much work put into the "black box" layer then there won't be much additional new traction here.

Before couldn't really even say that an indirect buffer had Ray Tracing Support or not. It is at least a distinguishable attribute now. And it is an indicator that Apple is probably going to spend time and effort optimizing it. But not necessarily means Apple hardware is immediately eminent.

There is zero mention of non Apple GPUs in the whole session. All the speed up metrics are with the current Apple Silicon GPU . So most of these ray tracing upgrades in this session are probably largely driven by lessons learned over first two years of "indirect command buffers" being fed back into an incrementally better Metal 3.

The improvements outlined here probably do work better on Apple's largest GPUs even if there is no specific hardware ray intersection fixed function logic. Probably works well on AMD GPU's with big enough Infinity Cache also. It doesn't look like a DirectX 12 DXR 'killer' addition to Metal or Nvidia RT software/hardware 'Killer" move. Just incrementally better than Metal than last iteration.

 

[diamond.g]

Do we think Apple will use AI for denoising like Nvidia is (with DLSS 3.5)? Seems like a free IQ win.

 

[Xiao_Xi]

A user on the Blender forum did an experiment comparing the speed that ray tracing hardware provides to Intel, AMD and Nvidia GPUs in Blender. You can't draw many conclusions from it, but the charts are very interesting. For instance:

Cycles AMD HIP device feedback

I’ve been interested in comparing the Ray Accelerators in the AMD GPUs to the Ray Tracing Units in Intel GPUs to the Ray Tracing Cores in Nvidia GPUs. And now that we have HIPRT, OptiX, and Embree GPU in Cycles, we can do that. And with Cycles we can extract some extra debug information to help...

devtalk.blender.org

I wonder what those charts would look like when the Apple GPU had ray tracing hardware.