Why is Apple slow to implement Hardware based Ray-Tracing?

[Xiao_Xi]

I don't know how Nvidia does it. Do they use limited-precision approximate tests?

So far, I have not been able to find it. The white papers don't explain it.

NVIDIA Ampere Architecture Whitepaper Available Now

Peek under the hood of the new NVIDIA Ampere GA10x architecture, which powers GeForce RTX 30 Series graphics cards.

www.nvidia.com

NVIDIA Turing Architecture Whitepaper Available Now For Download

Get an in-depth explanation of all the tech that’ll enhance and accelerate games on GeForce RTX Turing-architecture graphics cards.

www.nvidia.com

Real-Time Path Tracing and Beyond | Research

In this keynote presented at High Performance Graphics 2022, Petrik Clarberg shares an update on real-time path tracing and the next steps for real-time graphics research. Abstract

research.nvidia.com

 

[leman]

It looks like SER isn't used "automatically", you have to code for it.

Improve Shader Performance and In-Game Frame Rates with Shader Execution Reordering | NVIDIA Technical Blog

Learn about Shader Execution Reordering (SER), a performance optimization that unlocks the potential for better ray and memory coherency in ray tracing shaders.

developer.nvidia.com

Since only 3 cards support it, it will probably take a while for it to be added to programs.

Thanks for the link! I’ve had a look at it and I think there is a fundamental difference between how Nvidia describes raytracing and how Apple patents describe it. It appears that Nvidia hardware treats RT intersection tests similar to texture reads - a shader asks for an intersection result and obtains it some time later. These new APIs allow the shader to call a special type of synchronization barrier that will reorder threads based on ray hits to improve hardware utilization. However, if I understand Apple patents correctly, in their model invoking ray tests will basically terminate the shader program and after the RT tests have been performed, a new shader program will be assembled and launched with the intersection results. It’s a very different approach, and I don’t dare to even speculate about merits and disadvantages of either.

Unpopular opinion: Ray Tracing is over rated and unless you have a RTX 4090 class GPU, then you probably don't have it enabled in the game.

I have a rather anemic GPU that technically supports it, but is rather ineffective. I used it on a handful of cards and while it did add visual depth and better looking images, its not a night and day difference and playing games the scenes change enough that you won't miss it

I don't see why Apple has to implement it, particularly when gaming is not (at least as of yet) a major priority for them. I'd much rather see them improve other aspects of the GPU and CPU

You are absolutely right that current RT implementations require high-end power hungry hardware and even than the usability is limited. But a company that can bring useable RT to the mainstream devices might get a competitive advantage.

For Apple specifically one area of interest is production rendering. They have been pushing hard in this area with both hardware and software. If they can deliver a solution that closes the gap to Nvidia at a lower power consumption it would be a huge win for them.

 

[sunny5]

AMD, Nvidia and Intel have it on their GPUs and now ARM itself is coming out with a HW RT GPU next year.
Hopefully A16 has it otherwise Apple is going to lag behind.

HW RT is really useful in Blender and games.

Forget it, Nvidia and CUDA itself is an industry standard for many kinds of 3D stuff. And ray tracing is more for gaming and Mac is far from that. Also, who even render 3D works with Mac while Nvidia dominate 3D markets by 90%?

As long as Nvidia dominates both AI and 3D markets, there is no hope for Mac. It's just too slow.

 

[Xiao_Xi]

However, if I understand Apple patents correctly, in their model invoking ray tests will basically terminate the shader program and after the RT tests have been performed, a new shader program will be assembled and launched with the intersection results.

This is very similar to Any-Hit shader and intersection shader.

 

[satcomer]

All you switchers seem not to understand Apple years ago went with Mental support!

 

[leman]

This is very similar to Any-Hit shader and intersection shader.

It's important to clearly distinguish between the software model and the hardware model. I think we already touched upon that when we briefly discussed graphical pipelines.

 

[l0stl0rd]

Maybe they don't see demand for real time ray tracing currently (and I don't think RTX will be a thing this generation... it has been poorly implemented and in most cases its not worth the performance impact). If gaming interest starts to increase I can see they pushing in that direction. They are probably working in making it efficient enough to bring it to the iPhone and scale up to the Mac.

yup this says it all

 

[maflynn]

yup this says it all

Yup, that was the point I was making. Its eye candy that people turn off, as they prefer higher FPS. Only those people who sold their kidney to afford a 4090 can keep RT on as that card will have fast FPS with RT.

I'm not a huge gamer, but I balance visual detail with game performance. I'd rather have medium detail if that produces good frames per second. I don't think that mentality is an outlier.

Back to Apple, I said this before, adding ray tracing does nothing for Apple, and the majority of its customers. I would rather see Apple use its R&D budget on other improvements that we all could take advantage and enjoy.

 

[fuchsdh]

Yup, that was the point I was making. Its eye candy that people turn off, as they prefer higher FPS. Only those people who sold their kidney to afford a 4090 can keep RT on as that card will have fast FPS with RT.

I'm not a huge gamer, but I balance visual detail with game performance. I'd rather have medium detail if that produces good frames per second. I don't think that mentality is an outlier.

Back to Apple, I said this before, adding ray tracing does nothing for Apple, and the majority of its customers. I would rather see Apple use its R&D budget on other improvements that we all could take advantage and enjoy.

Raytracing looks really cool, but outside stuff like Doom and Minecraft where the raytracing is the visual candy, it's often a fairly minor improvement that definitely tanks your FPS. People like raytracing, but it's not the main bell and whistle they will prioritize in games.

Apple should get into the game, since raytracing is the future, but outside of their mobile GPUs (where they're mostly focused on things besides games, and the gaming performance is a nice benefit) they've never really led from the front for that kind of tech.

 

[l0stl0rd]

Yup, that was the point I was making. Its eye candy that people turn off, as they prefer higher FPS. Only those people who sold their kidney to afford a 4090 can keep RT on as that card will have fast FPS with RT.

I'm not a huge gamer, but I balance visual detail with game performance. I'd rather have medium detail if that produces good frames per second. I don't think that mentality is an outlier.

Back to Apple, I said this before, adding ray tracing does nothing for Apple, and the majority of its customers. I would rather see Apple use its R&D budget on other improvements that we all could take advantage and enjoy.

Yes they will only add it if they think it is necessary or enough “Pros” request it. Like they video encoders, it made sense in their mind so it got added.

Personally I hope they add it because of their investment in Blender but not sure if that will be enough.

How ever why does Resident Evil Village show greyed out RT options?

Perhaps RT was supported to come to the chips and it did not make it in time 🤷‍♂️

Perhaps RT will be in the M2 Pro/Max?

 

[Xiao_Xi]

It's important to clearly distinguish between the software model and the hardware model.

You're right. The Nvidia Turing GPU architecture white paper (page 36) explains:

The RT Cores in Turing can process all the BVH traversal and ray-triangle intersection testing, saving the SM from spending the thousands of instruction slots per ray, which could be an enormous amount of instructions for an entire scene. The RT Core includes two specialized units. The first unit does bounding box tests, and the second unit does ray-triangle intersection tests. The SM only has to launch a ray probe, and the RT core does the BVH traversal and ray-triangle tests, and return a hit or no hit to the SM. The SM is largely freed up to do other graphics or compute work.

 

[singhs.apps]

Why do we assume raytracing is just for games ? And since gamers prefer higher fps over visual quality , therefore raytracing should be a lower priority for Apple ?

Considering games that support Ray tracing would generally run slower on macs anyway compared to PCs, even with raytracing switched off ? So that Linus poll is an irrelevant example in the land of Macs (considering, again, that Macs have NEVER been a first choice for gamers, and Apple never pushed it’s Mac offerings as such ?)

Switching off ray tracing … well a far more comprehensive data would be available to the game hardware manufacturers, game engines, game publishers etc. Yet AMD and Intel are trying to push their solutions at the hardware level to catch-up with Nvidia…. Why ? Just marketing?

 

[theorist9]

I think Apple realizes this is an important group for the Mac. Generally the scientists use Macs, but their clusters/GPUs are on Linux so this isn't a huge deal at the moment. I'm sure Apple would like that to change and Metal compute shaders would be well suited since they are almost the same as C++14. From what I understand (I don't have CUDA experience) some projects actually target both CUDA and Metal using a subset of C++ using the same code with a shim layer. Certainly for CPU bound tasks, the M-series processors must look pretty compelling to scientific workloads.

I'm not sure if Apple is interested in creating MacOS AS computing clusters. They worked with a couple of US universities (UCLA and Virginia Tech) to create PowerMac clusters about 20 years ago, but I don't believe anything more was done after that.

They also had a clustering application called XGrid, but it was deprecated around 2012.

 

[Tagbert]

It appears that Apple has recently had plans to add hardware raytracing to their GPUs though that effort last year ran into some snags and had to be delayed. Perhaps that will come in one of the next rounds of GPUs?

Report: Apple axed plans for next-generation GPU in iPhone 14 Pro after rare engineering blunder - 9to5Mac

A report from The Information highlights employee retention issues inside Apple’s silicon group, with some employees sourced citing extremely long...

9to5mac.com

 

[Xiao_Xi]

By the way, Imagination has a new GPU IP.

IMG DXT GPU - Imagination

IMG DXT is an evolution of IMG CXT bringing its revolutionary Photon architecture, featuring Level 4 RTLS hardware ray tracing, to all mobile GPU IPs.

www.imaginationtech.com

Imagination has modified its own classificion to stay ahead of the competition.

 

[Flint Ironstag]

Well, ray tracing has arrived. Who's had a chance to benchmark the new machines?

 

[fathergll]

Wait until M3. M2 is basically just a slight refresh of M1.

I am not saying it will come with M3, but this will be the next big change in architecture so there is a chance.

Sounds like good advice.

 

[vigilant]

It's important to clearly distinguish between the software model and the hardware model. I think we already touched upon that when we briefly discussed graphical pipelines.

Wouldn’t the Dynamic Cache of the GPU pose to be a benefit for the graphics pipeline? Making more efficient use of the cache should leave room for larger results to hit the graphics pipeline more efficiently.

I’m not GPU expert. I mean, I understand them, but when you are talking about the points of intersection during that part of the conversation, it occurred to me that that could very well be a big part of why they didn’t do it till now.

 

[BootLoxes]

Sounds like good advice.

The best part about my advice is that it has a 50% chance of being right

 

[leman]

Wouldn’t the Dynamic Cache of the GPU pose to be a benefit for the graphics pipeline? Making more efficient use of the cache should leave room for larger results to hit the graphics pipeline more efficiently.

If your shaders are complex enough, sure. I think this will affect raytracing the most, since it relies on recursive function calls and dynamic dispatch.

Not quite sure what you mean by "leave room for larger results to hit the graphics pipeline more efficiently"?

 

[vigilant]

If your shaders are complex enough, sure. I think this will affect raytracing the most, since it relies on recursive function calls and dynamic dispatch.

Not quite sure what you mean by "leave room for larger results to hit the graphics pipeline more efficiently"?

The Dynamic Cache could allow for better ray tracing compared to many of the first iterations of ray tracing. I was referring to the recursive functions and dynamic dispatch.

 

[diamond.g]

Apple first gen RT is going against Intel's first gen, AMD's second gen and Nvidia's third gen. And it will not get a pass if it isn't as performant even though it is only their first try. Or at least they would not get a pass in non-Apple leaning locations.

If it is as performant, that will look really good (Arc is in this boat sorta as it came out right before the other two updated their RT architectures).

For Blender (rendering) it is all about the speed of Path Tracing, and any trick they can do behind the scenes to speed it up (I assume Nvidia uses brute force here).

Based on IMGTech's "Ray Tracing Levels" do we think Apple is at level 4?

 

[Xiao_Xi]

I assume Nvidia uses brute force here

Nvidia's latest generation of GPUs use Shader Execution Reordering.

Nvidia's Shader Execution Reordering is a big deal for ray tracing efficiency

Nvidia's Shader Execution Reordering tech is a deceptively large advancement in performance for ray tracing.

www.pcinvasion.com

Shader Execution Reordering: Nvidia Tackles Divergence

In a previous article, we covered Cyberpunk 2077’s “Overdrive” raytracing mode. Rasterization and raytracing are both parallelizable, but raytracing tends to be less predictable. …

chipsandcheese.com

 

[diamond.g]

Nvidia's latest generation of GPUs use Shader Execution Reordering.

Nvidia's Shader Execution Reordering is a big deal for ray tracing efficiency

Nvidia's Shader Execution Reordering tech is a deceptively large advancement in performance for ray tracing.

www.pcinvasion.com

Shader Execution Reordering: Nvidia Tackles Divergence

In a previous article, we covered Cyberpunk 2077’s “Overdrive” raytracing mode. Rasterization and raytracing are both parallelizable, but raytracing tends to be less predictable. …

chipsandcheese.com

I saw that but was under the impression that Blender (and other renderers that are not UE) don't actually use those APIs.

 

[diamond.g]

Found a Blender devtalk that speaks to SER:

2023-09-05 Render & Cycles Meeting

Attendees Brecht Van Lommel (Blender) Thomas Dinges (Blender) Xavier Hallade (Intel) Nikita Sirgienko (Intel) Patrick Mours (NVIDIA) Brian Savery (AMD) Christophe Hery (Meta) Michael Jones (Apple) Notes Hydra MaterialX export pull request was submitted, Brecht will review. This exports...

devtalk.blender.org

Looks like Optix 8 supports SER but Blender doesn't use that version. They do say that Cycles has it own reordering system so they are not sure if there would be a performance improvement.