AMD files Patent for hybrid Ray Tracing.

It's nearly impossible to discuss graphics tech in 2019 without bringing up real-time ray tracing. The rendering technique has been popularized by Nvidia, Microsoft, and an increasing number of game developers over the last few months. AMD's stayed pretty quiet about how it plans to support hardware-accelerated ray tracing, but a patent application published on June 27 offered a glimpse at what it's been working on.

AMD filed the patent application with the U.S. Patent and Trademark Office (USPTO) in December 2017. It describes a hybrid system that enables real-time ray tracing using a variety of software and hardware methods rather than relying on just one solution. The company said this approach should allow it to overcome the shortcomings associated with previous attempts to bring ray tracing to the masses.

In the application, AMD said that software-based solutions "are very power intensive and difficult to scale to higher performance levels without expending significant die area." It also said that enabling ray tracing via software "can reduce performance substantially over what is theoretically possible" because they "suffer drastically from the execution divergence of bounded volume hierarchy traversal."

Basically: using software to enable ray tracing on hardware that hasn't been optimized for the rendering technique requires a significant performance sacrifice. Most people don't like it when their hardware is hamstrung by software, even if it's supposed to enable some fancy new graphics, and the inability to handle other processing tasks at the same time can also make the graphics look worse anyway.

AMD didn't think hardware-based ray tracing was the answer either. The company said those solutions "suffer from a lack of programmer flexibility as the ray tracing pipeline is fixed to a given hardware configuration," are "generally fairly area inefficient since they must keep large buffers of ray data to reorder memory transactions to achieve peak performance," and are more complex than other GPUs.

So the company developed its hybrid solution. The setup described in this patent application uses a mix of dedicated hardware and existing shader units working in conjunction with software to enable real-time ray tracing without the drawbacks of the methods described above. Here's the company's explanation for how this system might work, as spotted by "noiserr" on the AMD subreddit:

https://www.tomshardware.com/news/amd-patents-hybrid-ray-tracing-solution,39761.html

AMD didn't think hardware-based ray tracing was the answer either. The company said those solutions "suffer from a lack of programmer flexibility as the ray tracing pipeline is fixed to a given hardware configuration," are "generally fairly area inefficient since they must keep large buffers of ray data to reorder memory transactions to achieve peak performance," and are more complex than other GPUs.

I find this part rather interesting because MS confirmed hardware Ray Tracing but never claimed it was a Hybrid setup.Sony claimed Ray Tracing and didn't claim anything more.

One thing is for sure even AMD recognize that HB RT has a nasty hit to performance.

More info on the link.

Yay more tracing of the rays.

AMD is so far behind when it comes to RT it hardly seems worth all this talk. Watch them get slapped by Intel next year.

come on @tormentos admit it, you wake up in the morning and you do a google search for ray tracing before you even have you're breakfast.

Thats probably the case.

Come on @tdkmillsy admit it,you have an alarm set up to warn you when ever i make a thread so you can come out of your cave like Batman.

Tormy is making a thread quick to the lemmobile..

Hopefully it means much better performance moving forward because ray tracing is crappy atm. Only one that impressed me was battlefield.

Ray tracing is still too experimental to be widespread. It wont really take off until 2025 and by then we might get a mid-gen console refresh with better RT support.

@tormentos: PC Gaming DXR/RTX is part of raster graphics pipeline, hence it's a hybrid. Hollywood's Renderman is raster and raytracing hybrid.

Remember, 5700 XT NAVI 10 has 10.3 billion transisors while TU106 RTX 2070 has 10.8 billion transisors.

TU106 10.8 billion transisors delivered

• Split 64 FP CUDA and 64 INT CUDA cores with Rapid Rack Maths per SM unit (CU unit equlavent).
  • 36 CU equlavent with just CUDA FP cores.
  • Note the reason for Turing's integer units split from floating point units. Hint: raytracing search tree not just BVH
• 288 Tensor cores for raytracing de-noise.
• 36 RT cores which acclerates two functions i.e. interection test and BVH search tree functions
• CUDA Turing cores still has DXR software capability. No PC game has combined shader and RTX resources for dual DXR! PC games are still holding PC's ultra graphics detail raster budgets while adding raytracing. Current PC games are not being design for RT from the start.
• Hardware aysnc compute from Volta.
• 64 ROPS with 4 MB L2 cache
• 9 MB register file storage (register storage is the fastest storage tech)
• Varable shading rate feature.

CUDA wave32 payload is broken down into 32 scalar threads within CUDA SM and it's NOT stuck with SIMD model! CUDA is base on many cores scalar architecture.

NAVI 10 (RDNA v1) 10.3 billion transisors delivered

• 40 CU with Rapid Rack Maths. Shared 64 Integer and Floating vector math units per CU.
• Each CU has dual branch units
• Each CU has dual scalar units
• Hardware aysnc compute
• 64 ROPS with 4 MB L2 cache
• ??? MB register file storage (register storage is the fastest storage tech)

AMD's BVH raytacing patent claim in year 2017

AMD adds Interection test acclerated hardware within TMU along with texture filter units.

AMD can't afford BVH RT acclerated hardware with TU106 level transisor budget!

AMD's argument can't be applied on Turing since NVIDIA has logic gate count vs function count ratio superority.

Note that Crytek claims Turing still has performance boost for thier Octree based raytracing.

NAVI's progress is good, but it's inferior to Turing.

Nvidia also share the cache between the texture units and "RT cores".

The context argument between Turing RTX vs Pascal/Volta vs AMD's year 2017 raytracing hybrid patent

AMD patent year 2017 has

• Ray/triangle intersection test hardware. Better than Volta, Pascal and Turing GTX. Volta and Turing GTX's SM has extra integer units, but they are not intersection test hardware.
• Transversing the tree stuctures are done by shaders and texture units. Volta, Pascal and Turing GTX has this level..

AMD patent year 2017's proposed GPU is half a Turing RTX LOL.

I did state AMD could improve heavy branching intersection test functions in my past comments without looking at AMD's year 2017 patents.

@nfamouslegend: It will be mainstream once 7nm Ampere releases. You're right though, AMD may finally get their RT support to be adequate by 2025.

Who really gives a shit! The new consoles will be more powerful and that is all that really matters.

I can’t wait to see my pew pew lasers be extra shiny in a next gen BattleFront game.

GTFO of my threads with your tantrums.

If you don't like it don't be blinded by your Tormy hate and stop entering my threads and hitting reply to derail them.

Now that means RT is new, exclusive to Sony and haven't been done before since AMD filed a patent for it. Well that's what cows think about patents as can be seen in another thread.

@tormentos:

"Claim 19:. A texture processor comprising: a texture address unit connected to a shader; a texture cache connected to the texture address unit; a ray intersection engine connected to the texture address unit, the texture cache and the shader; wherein: the texture address unit is configured to: receive from the shader a texture instruction which includes at least a bounded volume hierarchy (BVH) node pointer and ray data; filter the texture instruction to obtain the ray data; fetch BVH node data from the texture cache based on the BVH node pointer, the ray intersection engine is configured to: receive the ray data and the BVH node data; perform ray-BVH node type intersection testing using the ray data and the BVH node data; and send intersection results based on the ray-BVH node type intersection testing via a texture data return path to the shader."

In the DETAILED DESCRIPTION section:

"A texture processor based ray tracing acceleration method and system are described herein. A fixed function BVH intersection testing and traversal (a common and expensive operation in ray tracers) logic is implemented on texture processors. This enables the performance and power efficiency of the ray tracing to be substantially improved without expanding high area and effort costs."

"In particular, a fixed function ray intersection engine is added in parallel to a texture filter pipeline in a texture processor. This enables the shader unit to issue a texture instruction which contains the ray data (ray origin and ray direction) and a pointer to the BVH node in the BVH tree. The texture processor can fetch the BVH node data from memory and supply both the data from the BVH node and the ray data to the fixed function ray intersection engine.

The ray intersection engine looks at the data for the BVH node and determines whether it needs to do ray-box intersection or ray-triangle intersection testing (Note 1).

The ray intersection engine configures its ALUs or compute units accordingly and passes the ray data and BVH node data through the configured internal ALUs or compute units to calculate the intersection results (Note 2).

Based on the results of the intersection testing, a state machine determines how the shader unit should advance its internal stack (traversal stack) and traverse the BVH tree. (Note 3)

The state machine can be fixed function or programmable. The intersection testing results and/or a list of node pointers which need to be traversed next (in the order they need to be traversed) are returned to the shader unit using the texture data return path. The shader unit reviews the results of the intersection and the indications received to decide how to traverse to the next node in the BVH tree."

AMD's year 2017 RT patent doesn't completely detach shader resource with tree transversal workloads.

NAVI CU has increased texture filter processor width by 2X over GCN CU. AMD is preparing TMUs to be burden down with additional workloads.

TA = Texture Address Unit. Read/write instructions from compute shaders. Shaders could contain embeded data playload beside shader instructions.

TCP = Texture Cache Processor Unit, read/write data from on-chip cache.

It's logical to conclude placing "Interection Engine" in the TMU.

---

Note 1. Almost NVIDIA RT core equalvent. AMD recycled TCP unit for the fetch operation. NVIDIA RT has it's own fetch operation unit.

Deciding (branching) between ray-box intersection or ray-triangle intersection testing is similar to NVIDIA RT's 1st stage.

NVIDIA RT's can advance it's search tree node via independent loops and repeated fetch operations.

Note 2. From ray intersection engine, AMD is not sure between internal ALU or compute unit to calculate the intersection results.

Note 3. Advancing it's search tree node via shaders.

NVIDIA RT's can advance it's search tree node via independent loops and repeated fetch operations.

BVH raytracing is a standard in the next "RDNA 2". Custom argument for PS5 actually damages cow's argument.

Wonder how important RT will be in the end as a marketing tool. Resolution and framerate are more familiar battlegrounds and I suspect people will return to them.

Can you put a simple number on RT so that it's easy to compare between different platforms? People love to compare numbers because then they can know which is better without knowing anything about what it is. To be relevant I think RT needs a number associated with it so people/companies can claim victory when theirs is bigger.

That's true. I don't know if they're pulling a wait-and-see or what. I'm only paying attention to AMD's processors anyway.

@BenjaminBanklin: It's not wait and see, they're just playing from behind like usual. AMD is at least 2 years behind Nvidia at this point.

Last I checked this is an open forum. Seriously get over yourself and for the love of god quit crying. BTW that was not a tantrum just the truth.

No in fact i never once claimed that here,sony say ray tracing and nothing more,i do believe is hardware based and not just shader.

But i also believe that Ray tracing is the new buzzword this gen alone side SSD.

I don't see games on console having Ray Tracing without lowering assets or resolution to lower than 1080p.

Stronger GPU on PC have a huge impact by it,so i don't see how sony even with HB RT will not suffer the same fate,i would love to be wrong on this one and that both can do it,but i am not sold with any of the 2 console makers kwoning how hard HB RT hit performance on the RTX2080Ti which is a amn $1,000+ GPU.

Yes it was and you always do the same,if you don't care ab out RT what the f** are you doing in this thread.

Common sense use it.

Spreading facts that both consoles will be more powerful then what we have now so no sense in arguing the little things. It was not that hard to read into my post, but of course you have to go full cry mode because you didn't like being told the truth.

Pretty sure that's what you and your fellow cows are doing in the 'Sony filed a patent thread', hence it's never been done before.

@ajstyles: hah best reply this thread so fare!

NO quote me.

The main point with variable shading rate (VSR) feature is to reduce raster shading resolution while preserving geomety edges to native resolution, hence conserving shader resource for additional raytracing workloads.

Current PC games like Metro Exodus, Tomb Raider and BF5 doesnt use VSR and rapid pack maths (via DirectML access) features, hence it's wasting Turing's other new shader resource conservation hardware features.

Microsoft has DirectML with Windows 10 1903, hence it's late for current PC games. Games that used VSR (NV side) or RPM used Vulkan API with vendor specfic extensions or vendor spefic Direct3D API superset kitbash (AMD side with RPM).

I just did. The entire thread is you and your fellow cows DCing and saying it's different because patent, works on the system level, all sorts of crap. It's like saying show me the water is wet when you're in the water.

No.

1-Sony never said they have a hybrid system.

2-I think Ray Tracing on consoles will have some nasty effect,much like trying to puch 60FPS in some games on xbox one X,which force the xbox one X and even the Pro to bend knees greatly.

I had few post on this thread and non is hyping this from AMD.

You think? You're NOT Sony.

https://wccftech.com/gran-turismo-polyphony-raytracing/

Gran Turismo developer Polyphony Digital revealed they’ve created their own in-house real-time ray tracing tech

This in-house "ray tracing" indicates non-DXR/RTX bounding volume hierarchy ray-tracing method.

https://www.overclock3d.net/news/software/gran_turismo_developer_polyphony_digital_are_working_on_real-time_ray_tracing_tech/1

Polyphony Digital's "raytracing" is based on path-tracing which are one of "cheap rays" methods

At SIGGRAPH Asia 2018, Polyphony Digital confirmed that they were experimenting with real-time ray tracing, developing their ownpath tracing solution to offer accurate reflections that offer a notable graphical leap over the pre-baked reflections that are used in their recent games.

Sony is again playing lose with raytracing from "cheap rays" methods.

This is Sony's direct response to Microsoft DXR/NVIDIA RTX Turing marketing.

--------

For floating point handling, IBM/Sony didn't care about the full IEEE-754 rounding modes and NaN. IBM/Sony couldn't be bothered with keeping floating point handling and results to be the same between PPE's Altivec and SPE i.e. Sony's mindset is good enough and fuk accurancy. Different rounding modes between PPE's Altivec and SPE yields slightly different floating point results, hence BS "fusion" results.

@tormentos said:

@tdkmillsy said:

come on @tormentos admit it, you wake up in the morning and you do a google search for ray tracing before you even have you're breakfast.

Come on @tdkmillsy admit it,you have an alarm set up to warn you when ever i make a thread so you can come out of your cave like Batman.

Tormy is making a thread quick to the lemmobile..

I like that response, this forums isn't important enough to use alarms, but funny all the same.

nice one ;)

I wasn't talking about RT.

All of this talk about how great RTX Ray tracing is, while in the real world you need a $1200 graphics card to get 60FPS most of the time at 1080p.

Not in Metro Exodus.

https://www.gamersnexus.net/guides/3440-metro-exodus-rtx-benchmark-built-in-vs-game-tests

4K gameplay 52.9 fps, 48.5 fps and 40.4 fps results are in FreeSync's 60 to 40 hz range. Variable Shading Rate and CUDA rapid pack math wasn't used to conserve shader/TMU/memory bandwdith resources.

Very high raster graphics settings has demished visual quality returns.

Even though they had Radeon rays way be fore Nvidia had RTX......... lmao...

They probably have more experience with RT then the others.

NVIDIA has OptiX Ray Tracing Engine before RTX.

https://www.geeks3d.com/20100121/nvidia-optix-ray-tracing-sdk-available-with-geforce-support/

Year 2010.

@scatteh316: No they don't, even if they did it's funny you think that matters at this point.

What's funny is you trying to downplay things....

Why wouldn't it matter?

@scatteh316: I'm not downplaying anything, they're behind, period.

@Pedro: Real time RT in games vs dev tool... Who cares if they were first (which they weren't) when they're behind today? He made a irrelevant and incorrect point.

Real time in games especially on PC will be derived form the DXR API which available to both companies. The point I am gathering from Scatteh is that they have tech in ray tracing and is not so far behind as you are indicating(Nvidia is just more popular). Ray traced tanked on the current RTX cards. While it started the conversation it sure didn't nailed. Both companies will be offering more viable solutions in the future and the difference in performance when both offer a solution is to be seen.

Not sure how You inferred that from him when all he said is AMD was first (they weren't) and they probably have more experience (they don't).

Wishful thinking, AMD has shown nothing to garner such optimism.

How is it wishful thinking that AMD is going to offer a viable solution when next gen consoles will be featuring hardware Ray tracing and using AMD? Now you just not making any sense.

@Pedro: Lol, I'm making perfect sense. Did they ever confirm HWRT? I don't think they have. Even if they did how do you know it will be up to snuff? Lot of optimism for a company that has been historically lack luster for a decade.

Microsoft confirmed Hardware Ray Tracing for their next system and their next system is powered by AMD. Not sure where is the confusion. "Lot of optimism" I wasn't expressing any optimism, just stating what has been disclosed.

@Pedro: Cool, well looks like AMD is still behind then.

AMD claims hardware accelerated raytracing on the next "RDNA 2" in year 2020.

Micrsoft also claims hardware accelerated raytracing for Xbox Scarlet in year 2020

AMD's 2017 patent is missing seperate tree node transvesal and fetch hardware. NAVI 10 with 40 CU which contains 160 TMUs, hence 160 "interect engines" for RDNA 2 varaint LOL.

Now look what you have done? See above.