AMD RDNA3 Specifications Discussion Thread

[Henry swagger]

https://twitter.com/x/status/1603117873949466624

So amd lied about performance per watt

 

[Bondrewd]

144FPS limit

zoinks

 

[digitalwanderer]

Wouldn't that frame limit hold the 40x0 series way back from what their power draw would be if they could run uncapped? Just another stupid question...

 

[Wesker]

https://twitter.com/x/status/1603146626067046401

Ah jeez, so funny.

Navi 31 is the new R600.
Hot, broken***, and too little too late. Designed to clock high, but massively undershooting its clock speed targets. Also a very misbalanced architecture -- overkill on memory configuration, yet unable to take take advantage of it.

But priced to awkwardly compensate and fit into Nvidia's pricing scheme.

***Remember that R600 had broken MSAA resolve in its ROPs, causing performance to tank whenever AA was enabled in games.

 

[Bondrewd]

and too little too late

?
It's ugh, not even 1Q after NV stuff.

Navi 31 is the new R600

R520 vibes really.
R600 had fundamental uArch problems which weren't really solved until RV770, this one does not.

Also a very misbalanced architecture -- overkill on memory configuration, yet unable to take take advantage of it.

It's fine.

 

[Jawed]

The code quoted implies that A0 of some RDNA 3 chips has no pre-fetch. It doesn't imply that the cards are using A0.

You'd need some other evidence that the cards people have bought are A0.

 

[Bondrewd]

You'd need some other evidence that the cards people have bought are A0

Well it's not *that* hard to find it out, really.
That's the VBIOS that comes with any 7900 anything you buy now (i.e. made by AMD boards).

 

[Jawed]

OK, so this thread is funny:

https://twitter.com/x/status/1603145507488014336

 

[PSman1700]

Guess Sony and MS could look at alternatives for their next machines.

 

[Bondrewd]

Guess Sony and MS could look at alternatives

rofl

 

[BRiT]

Guess Sony and MS could look at alternatives for their next machines.

Why? Console SOCs are entirely different. By the time either look to make next-gen consoles AMD will have 4.5 to 5.5 more years of designs.

 

[Bondrewd]

Console SOCs are entirely different

And no one else has quite this combo of good IP; not even talking about backcompat points.

 

[Remij]

Eeesh!

What the hell is going on in Computopia these days?! It wasn't supposed to be like this!

 

[PSman1700]

Why? Console SOCs are entirely different. By the time either look to make next-gen consoles AMD will have 4.5 to 5.5 more years of designs.

Well then, lets hope AMD got their act together by then.

 

[neckthrough]

Shipping A0 is nothing new for GPUs. They *all* have bugs, but most can be worked around in bringup. AMD just got very very unlucky this time.

 

[trinibwoy]

Wouldn't that frame limit hold the 40x0 series way back from what their power draw would be if they could run uncapped? Just another stupid question...

The uncapped results aren’t much different. Wonder why AND thought it was ok to claim a 50% perf/watt improvement.

 

[techuse]

The uncapped results aren’t much different. Wonder why AND thought it was ok to claim a 50% perf/watt improvement.

There is probably some point on the power curve where it applies.

 

[Qesa]

There is probably some point on the power curve where it applies.

In the footnotes didn't they specify it was 7900 XTX vs 6900 XT with both cards at 300W? 7900 XTX isn't 54% faster at 355W, let alone when limited to 300W

 

[techuse]

In the footnotes didn't they specify it was 7900 XTX vs 6900 XT with both cards at 300W? 7900 XTX isn't 54% faster at 355W, let alone when limited to 300W

I’m sure a game exists where it is true. Did the footnotes specify a title?

 

[homerdog]

We’re they planning on decoupling the clockspeeds from the start? The first thing I thought when I read about that was “something’s wrong”.