AMD to Redesign Ray Tracing Hardware on RDNA 4

[btarunr]

AMD's next generation RDNA 4 graphics architecture is expected to feature a completely new ray tracing engine, Kepler L2, a reliable source with GPU leaks, claims. Currently, AMD uses a component called Ray Accelerator, which performs the most compute-intensive portion of the ray intersection and testing pipeline, while AMD's approach to ray tracing on a hardware level still relies greatly on the shader engines. The company had debuted the ray accelerator with RDNA 2, its first architecture to meet DirectX 12 Ultimate specs, and improved the component with RDNA 3, by optimizing certain aspects of its ray testing, to bring about a 50% improvement in ray intersection performance over RDNA 2.

The way Kepler L2 puts it, RDNA 4 will feature a fundamentally transformed ray tracing hardware solution from the ones on RDNA 2 and RDNA 3. This could probably delegate more of the ray tracing workflow onto fixed-function hardware, unburdening the shader engines further. AMD is expected to debut RDNA 4 with its next line of discrete Radeon RX GPUs in the second half of 2024. Given the chatter about a power-packed event by AMD at Computex, with the company expected to unveil "Zen 5" CPU microarchitecture on both server and client processors; we might expect some talk on RDNA 4, too.



View at TechPowerUp Main Site | Source

 

[Onasi]

I mean, one would assume so, seeing as how RT is kinda a weak-spot for AMD for now. Still, I would not hold my breath expecting miracles. Overhyping leads to inevitable disappointment when reality hits.

 

[Minus Infinity]

AMD had patent approved last year for BVH acceleration in hardware that would greatly speed up RTing. From online Nvidia info:

"Bounding Volume Hierarchy (BVH) is a popular ray tracing acceleration technique that uses a tree-based “acceleration structure” that contains multiple hierarchically-arranged bounding boxes (bounding volumes) that encompass or surround different amounts of scene geometry or primitives. Testing each ray against every primitive intersection in the scene is inefficient and computationally expensive, and BVH is one of many techniques and optimizations that can be used to accelerate it. The BVH can be organized in different types of tree structures and each ray only needs to be tested against the BVH using a depth-first tree traversal process instead of against every primitive in the scene. Prior to rendering a scene for the first time, a BVH structure must be created (called BVH building) from source geometry. The next frame will require either a new BVH build operation or a BVH refitting based on scene changes."

So AMD hopefully might be adding hardware to greatly facilitate BVH. You can search for patent US2023/0206543A1 by AMD for more details.

 

[64K]

Hoping this turns out well. I'm wondering how the Leather Jacket Guy will be pricing 5xxx GPUs without more competition from AMD.

tbh even those that would never buy a Radeon product should root for more success from AMD in RT performance.

 

[Flyordie]

One of the many reasons I am going to try and hold onto my Vega64 for one more generation if I can.

 

[Quicks]

Was thinking of switching to Nvidia because of superior DLSS and RTS. But if AMD can do the same I might just stick around for a bit longer.

 

[Onasi]

Hoping this turns out well. I'm wondering how the Leather Jacket Guy will be pricing 5xxx GPUs without more competition from AMD.

tbh even those that would never buy a Radeon product should root for more success from AMD in RT performance.

That’s a wrong question to ask. It’s AMD that prices its cards relative to what NVidia asks for. They tried the price war strategy in the past. It did not work. Overall, NVidia will ask for what they think the market can bear and AMD will be forced to price similarly to what they did before- NVidia, but slightly cheaper.

 

[ZoneDymo]

What I like always, is working with what you got, dealing with the cards dealt.
Bit of a sidestep, but I loved top gear a lot more then grand tour and I think a lot of that has to do with budget restraints they used to have, it asks for creativity which leads to great results.

With AMD, I dig that they are stuck with the consoles, I like that those are not fantastic which leads to innovation in optimization, how can we get some solid out of something that isnt quite that capable?
With an RTX4090 you basically just throw vast amounts of horsepower at it, not terrible but imo not as interesting or impressive.

It makes sense that they have to go with the times and the way the (gaming) world leads, but I love knowing the consoles will still be around for a long time forcing innovation through intelligent design.

 

[AusWolf]

If I wasn't dead tired after work, I could probably find the ancient post where I said something like "we don't need more raw power, we need better RT engines for a smaller performance hit". It seems AMD listened.

 

[Dristun]

Happy for AMD fans that are going to finally check out how cool path tracing looks in CP77 and Alan Wake 2 but I regret to inform them that the performance was complete garbage even on 4090 and they'll have to join everyone else in dreaming of some distant future cards that will be able to run stuff like this properly.

 

[ZoneDymo]

Happy for AMD fans that are going to finally check out how cool path tracing looks in CP77 and Alan Wake 2 but I regret to inform them that the performance was complete garbage even on 4090 and they'll have to join everyone else in dreaming of some distant future cards that will be able to run stuff like this properly.

DF recently did a video of a dev hack mod thing for Dragons Dogma 2.
This allowed for path tracing to be active during the game, explained was that that would be used to measure baked lighting against.

Anywho, there is no denoiser for it so its a grainy mess but obviously in the future, we could simply up the bounce count to a point where a denoiser would not be needed.
But like you said, that is future music.

 

[Vya Domus]

Happy for AMD fans that are going to finally check out how cool path tracing looks in CP77

Yeah they can finally check out how the game looks almost indistinguishable from the regular RT, a real treat.

 

[Daven]

Happy for AMD fans that are going to finally check out how cool path tracing looks in CP77 and Alan Wake 2 but I regret to inform them that the performance was complete garbage even on 4090 and they'll have to join everyone else in dreaming of some distant future cards that will be able to run stuff like this properly.

Yeah, I’m totally psyched to stare at light reflecting off street water for hours on end rather than actually playing the game.

 

[Dristun]

Yeah, I’m totally psyched to stare at light reflecting off street water for hours on end rather than actually playing the game.

It's cooler to play the game *and* have cool lights reflecting off street water! And in 5 years we might be even able to do that without ugly upscalers.

 

[dgianstefani]

Yeah they can finally check out how the game looks almost indistinguishable from the regular RT, a real treat.

Ah, and you have tried this with your AMD card?

Cool, wasn't aware it could run PT.

It's cooler to play the game *and* have cool lights reflecting off street water! And in 5 years we might be even able to do that without ugly upscalers.

I don't find DLSS/DLAA ugly.

DF recently did a video of a dev hack mod thing for Dragons Dogma 2.
This allowed for path tracing to be active during the game, explained was that that would be used to measure baked lighting against.

Anywho, there is no denoiser for it so its a grainy mess but obviously in the future, we could simply up the bounce count to a point where a denoiser would not be needed.
But like you said, that is future music.

PT is certainly an interesting tech.

If I wasn't dead tired after work, I could probably find the ancient post where I said something like "we don't need more raw power, we need better RT engines for a smaller performance hit". It seems AMD listened.

Two generations too late for good upscaling or RT so only targeting mid range. Fingers crossed for RDNA 5 if it's called that.

I wonder if Battlemage will have a higher end card than RDNA 4. Intel's upscaler and RT tech is already ahead.

 

[john_]

They better be tremendously improving that RT performance.
If RDNA3 was just RDNA2+double the RT, AMD would had much better success than what it managed with the current form of RDNA3.

 

[dgianstefani]

AMD had patent approved last year for BVH acceleration in hardware that would greatly speed up RTing. From online Nvidia info:

"Bounding Volume Hierarchy (BVH) is a popular ray tracing acceleration technique that uses a tree-based “acceleration structure” that contains multiple hierarchically-arranged bounding boxes (bounding volumes) that encompass or surround different amounts of scene geometry or primitives. Testing each ray against every primitive intersection in the scene is inefficient and computationally expensive, and BVH is one of many techniques and optimizations that can be used to accelerate it. The BVH can be organized in different types of tree structures and each ray only needs to be tested against the BVH using a depth-first tree traversal process instead of against every primitive in the scene. Prior to rendering a scene for the first time, a BVH structure must be created (called BVH building) from source geometry. The next frame will require either a new BVH build operation or a BVH refitting based on scene changes."

So AMD hopefully might be adding hardware to greatly facilitate BVH. You can search for patent US2023/0206543A1 by AMD for more details.

This is one of the things needed. Actual full hardware acceleration rather than a half hearted approach.

 

[dgianstefani]

They better be tremendously improving that RT performance.
If RDNA3 was just RDNA2+double the RT, AMD would had much better success than what it managed with the current form of RDNA3.

Yep. The "all we need is raster" approach was doomed even with RDNA 2, the fact they persisted as long as RDNA 3 was amazing and explains the "nosedive" AMD GPU sales made according to their earnings report.

They better be tremendously improving that RT performance.
If RDNA3 was just RDNA2+double the RT, AMD would had much better success than what it managed with the current form of RDNA3.

I'm also hoping the AI trained FSR isn't more of the same, and actually delivers good IQ.

 

[Vya Domus]

If RDNA3 was just RDNA2+double the RT, AMD would had much better success

If that was the case Nvidia would have still been faster in RT and nothing would have changed, be sure that when AMD will get close to Nvidia in RT they'll move the goal post to something else.

 

[dgianstefani]

If that was the case Nvidia would have still been faster in RT and nothing would have changed, be sure that when AMD will get close to Nvidia in RT they'll change the goal post to something else.

My dude the goal posts are set by consumers. Instead of being salty AMD is behind, perhaps be hopeful they release competitive products in the DGPU scene, rather than "NVIDIA, but last gen, and without many of the software/hardware perks, at 5-10% cheaper".

Aiming for second place isn't aggressive or ambitious enough, especially now that there's three players in the scene.

 

[Hyderz]

Would love to see amd pull a 9700pro level performance on nvidia for rdna4 but I’m not too hopeful

 

[Vya Domus]

My dude the goal posts are set by consumers.

Literally no one ever asked for things like path tracing to be added to games.

 

[dgianstefani]

That’s a wrong question to ask. It’s AMD that prices its cards relative to what NVidia asks for. They tried the price war strategy in the past. It did not work. Overall, NVidia will ask for what they think the market can bear and AMD will be forced to price similarly to what they did before- NVidia, but slightly cheaper.

It would be nice if AMD would take the lead in ambition/aggression/innovation with graphics. Instead of just responding to whatever NVIDIA does, but a generation or two behind.

Maybe a price war strategy would work if the products were roughly equivalent and thus competitive, but they're not.

Being forced to price your cards according to competition - 10%, rather than dictating prices based on performance and value of the featureset is weak and without initiative.

 

[64K]

Would love to see amd pull a 9700pro level performance on nvidia for rdna4 but I’m not too hopeful

Would be nice. It's been a while since they gave Intel a spanking. I would enjoy the ensuing tech drama on the net.

 

[Pumper]

If they manage to greatly improve RT performance (at least 50%) and FSR image quality (by using "ai" upscaling), AMD will get way more gamers buying their cards, as most are going to nvidia just for DLSS/RT advantage.