AMD: RDNA 3 Speculation, Rumours and Discussion

Status
Not open for further replies.
You mean put the Ray Accelerators outside the CU ? My take is they can beef the RA, and still having them in the CUs. It's still dedicated hardware.

How could they "beef" them up? It's a limited and very cheap implementation of hardware raytracing. They have to redesign their whole raytracing implementation.
 
They already have hardware RT... I guess they will try to accelerate more "stages" in the futur for sure, but saying they don't have hardware RT is wrong.
Ray tracing requires intersection testing and acceleration structure traversal.

AMD only has accelerated intersection testing. It appears likely that AMD has no plans to add new hardware functionality, to make traversal accelerated.

It appears AMD wants to make dynamic branching less of a disaster for SIMD-based architectures. Amongst other things, that will make acceleration structure traversal faster, because less ALU lanes will be turned off due to incoherent processing of rays.

At the same time way more SIMDs will make ray tracing go faster.
 
Thats if we assume AMD manages to go another GPU generation without dedicated hardware support for ray tracing, which i cant imagine they will.
They have dedicated hardware support for raytracing in RDNA2 no matter how you try to twist it.

Edit: @Jawed you can't just arbitrarily decide NVIDIAs spec is the limit for "hardware acceleration" or hardware raytracing. PowerVR accelerated more "stages" before NVIDIA had even their first RT cores out
 
How could they "beef" them up? It's a limited and very cheap implementation of hardware raytracing. They have to redesign their whole raytracing implementation.

By making the RAs do more stuff ? Or faster ? Maybe you're right and they can't do that without rethinking everything. I don't know.
 
They have dedicated hardware support for raytracing in RDNA2 no matter how you try to twist it.

Nothing to do with twisting, theres hardware and hardware support..... I clearly mean like what NV is doing, we have one here claiming amd wont, but i hope he's wrong and that they will.
If they can do it without 'dedicated' hardware then thats fine, it doesnt matter to us pc gamers (which this thread is about anyway). On PC NV is having the dominance in users, if things balance out abit (like the 9700pro did), prices will aswell, aside from nv getting more heat and thus the thrive for even faster/more advanced etc innovations.
 
which hopefully both drives prices down
not happening ever again
You're never getting your green toy for less.
The red toys will be more expensive than the green ones, too.
aside from nv getting more heat and thus the thrive for even faster/more advanced etc innovations.
That's not how any of this works.
All that changes is pricing.
Product roadmaps rarely change due to extrenal pressure.
 
Last edited:
Nothing to do with twisting, theres hardware and hardware support..... I clearly mean like what NV is doing, we have one here claiming amd wont, but i hope he's wrong and that they will.
If they can do it without 'dedicated' hardware then thats fine, it doesnt matter to us pc gamers (which this thread is about anyway). On PC NV is having the dominance in users, if things balance out abit (like the 9700pro did), prices will aswell, aside from nv getting more heat and thus the thrive for even faster/more advanced etc innovations.
So in other words you're at the same time saying NVIDIA doesn't have hardware acceleration for raytracing because PowerVR accelerates more "stages"?
 
#1 is GPU-specific, so it's not coming to PC.
How come?
Didn't the PC GCN GPUs take advantage of e.g. the async optimizations that were widely adopted on the 2013 consoles and their mid-gens? And eventually of triangle culling through compute shaders (which benefited GCN much more than Nvidia's architectures and even made Vega's own NGG obsolete)?


CP2077 runs at less than 30 fps @ 4K with all bells and whistles on (including all the RT gimmicks), so I don't see why anyone would be against getting a GPU that will allow you to play with smooth 120 fps without an IQ degrading crutch
I'm really looking forward to seeing what a Navi 31 or similarly-performing beast can do in real-time rendering, but throwing more watts and compute resources at Cyberpunk 2077 's wholly mediocre performance/visuals ratio on "Ultra RT" is not even within my top-20 list.
Others are already showing significantly better visuals on weaker hardware.
 
Edit: @Jawed you can't just arbitrarily decide NVIDIAs spec is the limit for "hardware acceleration" or hardware raytracing. PowerVR accelerated more "stages" before NVIDIA had even their first RT cores out
If you flesh-out your argument then maybe it's worth discussing.

DXR is the API we have. If you can map the functional parts of DXR to candidates for hardware acceleration, that would be a good start.
 
not happening ever again

I hope your wrong ;)

So in other words you're at the same time saying NVIDIA doesn't have hardware acceleration for raytracing because PowerVR accelerates more "stages"?

Lets say it this way, i hope AMD will compete with NV in the ray tracing department, however they do it doesnt matter.

Cyberpunk 2077 's wholly mediocre performance/visuals

Oof, still havent got over it i see. I think its up there with rift apart, technically superior.

He obviously knows a lot about AMDs roadmap but I wouldn’t put much stock in his Nvidia info. It’s always Armageddon with this guy.

I dont mind him, if he provides valueble AMD information in a civil manner (which i think its untill now), its all good.
 
Ray tracing requires intersection testing and acceleration structure traversal.

AMD only has accelerated intersection testing. It appears likely that AMD has no plans to add new hardware functionality, to make traversal accelerated.

It appears AMD wants to make dynamic branching less of a disaster for SIMD-based architectures. Amongst other things, that will make acceleration structure traversal faster, because less ALU lanes will be turned off due to incoherent processing of rays.

At the same time way more SIMDs will make ray tracing go faster.

Adding the mentioned but optional fixed function traversal block from the older TMU patent to the list of speculation.
Or they do nothing special. Not everybody thinks RT is hyper important.

I thought about more applications of monster GPU power. Layered framebuffers? For some transparency, fixing SS hacks, which would offload a lot RT work?
...all that stuff that seemed completely unpractical just some years ago...
 
I dont care how they do it, aslong as they drive up competition which hopefully both drives prices down aswell as more pressure in the market to innovate and generally even larger leaps in performance.

Hive mind in the PC gaming space is an obstacle to fair competition.
HD 4000, 5000, and 7000, and RX 200 and 400 GPUs yielded nothing for AMD despite having the better valued products (and often times the performance crown). People were more than happy to buy another respin of G92 or a lower-tier Nvidia option rather than a faster and equally priced AMD product.

So Su is leading AMD to stop being the bargain option. Performance will come at a premium. Consumers want to purchase products which they deem as premium or boutique products. Look at Apple or Microsoft's Surface line of products.

Maybe that happens when Apple joins the game dunno.

Apple couldn't care less about entering the gaming GPU market.
 
So Su is leading AMD to stop being the bargain option. Performance will come at a premium. Consumers want to purchase products which they deem as premium or boutique products. Look at Apple or Microsoft's Surface line of products.
We get what we deserve :D
Once understood, i still hope we get this with proper iGPU as reward!
Idea: 1. Beat competition 2. Vaporize it
 
Status
Not open for further replies.
Back
Top