AMD RDNA3 Specifications Discussion Thread

While my specualtion of automatically SW rasterizing tiny triangles in the RDNA3 specualtion thread was not serious,
I am indeed disappointed neither NV nor AMD did something like this to address the problem.
Writing a tringle rasterizer on GPU feels just as wrong as writing RT traversal. They should take away the burden from us. :D
 
New slide says the RT unit gets "1.8x" RDNA2 perf at 2.5ghz. But official slide deck says 1.5x. With the split clocks and the mismatch I'm not sure what this means. Is the RT unit on the shader clock or front end? Is it on it's own clock? Still don't even know how much RT performance relies on this block. So now I'm even more confused.

If AMD can fix the bug in time for other RDNA3 cards/rebooting this one they should just reboot the GPU series. The 7XXX moniker is already a mess and now tainted too.

Move it to something sensible, call it RDNA 3XX so there's way less confusion/codes/bs. It's just RDNA, it's just 3 numbers, X(Arch)X(Die)X(Bin). No "RDNA3 RX 7XXX XTX" crap.
 
Last edited:
so the question is how well does the TX OC / what is its real game clock and how far can you push the stock cooler above 300wat tdp.
I kinda want to get a good ray tracing GPU around xmas / witcher 3 rerelease. But not looking forward to spending 2k AUD on a GPU.........
 
If AMD can fix the bug in time they should just reboot the GPU series.
Are you sure there is a bug at all?
It's just a broken telephone, usual suspects had heard somewhere about the 3 GHz frequency, but didn't know anything about RDNA 3's 2 frequencies domains architecture at the time, now they have to speculate about the new revision because otherwise they would look like fools.
 
Are you sure there is a bug at all?
It's just a broken telephone, usual suspects had heard somewhere about the 3 GHz frequency, but didn't know anything about RDNA 3's 2 frequencies domains architecture at the time, now they have to speculate about the new revision because otherwise they would look like fools.
Well, 3 GHz is mentioned as a design target in AMD's own slides. Obviously that doesn't prove that a "bug" is preventing them from hitting that, but it at least suggests that certain architectural choices didn't pan out as expected.
 
If people are worried about future proofing, pretty much all games will be made with consoles notably weaker than these GPUs as a base both from a rasterization perspective and a ray tracing perspective.

So I don't think anyone should be concerned about even the worst case scenarios they are thinking about.

The RT uplift this new RDNA architecture gets may not be to the same level as Nvidia's 4000 series, but it's still a respectable jump from their previous line and it seems like their pricing and marketing properly reflects that reality.

I guess what I'm really asking is, when you get right down to it does this really deserve such back and forth and controversy? If anything id be irritated with AMD not showing certain details at the presentation but I wouldent neccisarily be concerned.

Seeing a shock and awe product like the 4090 come out surely stirred up the market but that came with it's own issues as a product so when you get right down to it, everything becomes a matter of subjectivity based on your personal priority or preference.

I remember hearing a lot about AMD inherent driver support years ago which automatically made people run away from their products, but as far as I know that's for the most part been resolved?

A tentative post musing on the PC market from a console user interested in the workings of the PC market. Tell me if I was off base with anything stated here
 
Well, 3 GHz is mentioned as a design target in AMD's own slides. Obviously that doesn't prove that a "bug" is preventing them from hitting that, but it at least suggests that certain architectural choices didn't pan out as expected.
Well, it is rated at 355 watts and in one slide of the presentation/YT video, they detail how much power (25%!) the decoupled clocks, i.e. downclocking the shaders from 2.5 to 2.3 GHz "game clock" yields for them. Do the math.
 
If people are worried about future proofing, pretty much all games will be made with consoles notably weaker than these GPUs as a base both from a rasterization perspective and a ray tracing perspective.
A tentative post musing on the PC market from a console user interested in the workings of the PC market. Tell me if I was off base with anything stated here

6600XT would net you ballpark console performance (PS5/XSX) in raster and RT. That gpu is probably good enough for most 'casual' gamers, while higher end gamers are not even having something that low-end on the radar to start with. The 'shock & awe' 4090 is again for the enthusiast market not intended for the average gamer, thats either 4060 or even 4070 when talking users intrested in nvidia products, for AMD that'd be 7600XT or something, not 7900XTX...

Regarding RDNA3, AMD most likely felt they needed more time for their RT and ML implementation akin to intel and NV, they could be ironing out things and have it ready with RDNA4.
 
Well, 3 GHz is mentioned as a design target in AMD's own slides. Obviously that doesn't prove that a "bug" is preventing them from hitting that, but it at least suggests that certain architectural choices didn't pan out as expected.
"The bug" is likely a severe increase in power draw which makes it impossible to achieve said clocks on Navi 31.
"Architected to exceed 3GHz" though is about RDNA3 and not Navi 31. So a smaller narrower chip may well manage to do this.
 
Why not put the ROPs on the MCDs as well?
They're cliented to L1 now so that's way inside the chip.
A real possibility for RDNA4 but that's a way different approach to tiled GPU design.
AMD most likely felt they needed more time for their RT and ML implementation akin to intel and NV,
No.
Just no.
RT yes, MFMA no.
they could be ironing out things and have it ready with RDNA4.
No RDNA4 is all about chiplet approach to be even more cost efficient.
"The bug" is likely a severe increase in power draw which makes it impossible to achieve said clocks on Navi 31.
No.
You'll see the issue soon enough.
"Architected to exceed 3GHz" though is about RDNA3 and not Navi 31.
No, it's N31.
They were lowballing 3.15 and high-rolling ~3.35 before shit got awry.
So a smaller narrower chip may well manage to do this
No there's a genuine wall in there.
 
3 GHz is mentioned as a design target in AMD's own slides.
They don't specify the part of the chip for which that was a goal, that's just a vague 3 GHz that is still here in the latest slides. That implies that the goals have been achieved (It would be very weird to mention them otherwise).
Front-end can hit this frequency range and the rest of the chip may simply be power limited, so why would that power limitation be "a bug"?
 
Last edited:
Back
Top