AMD RDNA3 Specifications Discussion Thread

AMD losing MS where? Regardless, the performance was aimed higher, but it would have never reached 4090 anyway. It's around 4080/16 speeds (excluding RT) at cheaper price, the chip is bigger when counting all the dies, but chiplets add some overhead and they're using (relatively) cheap older "7nm" node (N7x/N6x) on MCDs, GCD is same "5nm" node (N5x/N4x) but it's possible 4N is customized from denser N4x rather than N5x used by AMD. Not quite sure how that can be seen as a huge failure?
I mean market share, according to the recent Jon Peddie report. If the clock speed was meant to be as high as Bondrewd et al. are reporting, it's a big miss and not the kind of issue AMD needs right now.
 
AMD losing MS where?
AIB share.
It's around 4080/16 speeds
It's plenty faster lol.
Just not as fast as it should've been.
but it's possible 4N is customized from denser N4x rather than N5x used by AMD
They're all samey variants of N5.
Not quite sure how that can be seen as a huge failure?
It's huge failure because the big boy isn't quite the PPA slam dunk the lesser ones shall be soon enough.
 
Nvidia doesn't ship broken hardware.
They ship defective by design software (driver) tho.

The anti mining stuff like licking clocks to lowest when computer being used headless, etc are a pain (and dp dummy plug didn't work, only hdmi dummy plug work to circumvent it)
 
That’s pretty bad, it’s barely faster than the 4080 in raster performance and the performance increase over Navi21 is not what I expected after AMD’s presentation. The RT performance seems somewhat better than I thought it would be, at least.
 
Not bad for 'buggy' stuff. : )
Well yeah in absolute fps numbers it's okay (although looking into benchmark mix is always advisable).
When you look at competition though you're getting a 58B transistors GPU being about equal to 46B in "rasterization" (vs AD103) and to 28.3B in raytracing (vs GA102).

That’s pretty bad, it’s barely faster than the 4080 in raster performance and the performance increase over Navi21 is not what I expected after AMD’s presentation. The RT performance seems somewhat better than I thought it would be, at least.
RT performance is exactly where everyone was expecting it to be: on GA102 levels.
 
What's "broken" hardware? It's functional, the performance just isn't where it was meant to be and it's priced accordingly. Don't like it, don't buy it – but they aren't "screwing" anyone. You don't think NV ships hardware without all sorts of errata, do you?

That being said, Navi31 seemingly being so busted that the performance is nowhere near where the competitor is and where it was projected to be is a huge failure for AMD, not something NV has done in recent years. Made worse by the fact that it's at a time when AMD keeps losing MS and relevance in dGPUs...they really didn't need this.
Their only really saving point is that these should genuinely be relatively affordable to make, so if they can at least sell them, it's still decent money.

Definitely gonna do undue a lot of the positive feelings people had about Radeon GPU's after RDNA2, though.

And while I've got nothing to contribute to the idea of them being 'broken' or not, I think it's undeniable that this is well below expectations, and even well below AMD's own claims by more than an explainable margin. Given the jump to 5nm, 'brand new' architecture, and decent development length, a 35% performance gain is just very hard to explain. This cant have been what AMD was actually aiming for. Something went wrong somewhere.
 
Now that we have a production game using UE5.1, I'd really like to start seeing benchmarks of performance comparison for Nanite and Lumen. With so many developers moving to UE5, I think the reliance on hardware RT isn't going be as high as the industry was moving to previously.
 
Back
Top