AMD: RDNA 3 Speculation, Rumours and Discussion

[fehu]

I think that he's probably referring to Infinity Cache given that everyone was expecting at least 256MB

 

[trinibwoy]

I think that he's probably referring to Infinity Cache given that everyone was expecting at least 256MB

Hmmm I doubt he’s making excuses for RDNA 3 already. Probably talking about Ada’s massive girth.

 

[Nebuchadnezzar]

Probably a hint towards frequencies vs unit count.

 

[DegustatoR]

AMD Radeon RX 7900XT PCB layout leaks out with Navi 31 GPU and three 8-pin power connectors - VideoCardz.com

Alleged AMD Radeon RX 7900XT PCB Igor Wallossek is sharing a first look at the board design for the unreleased RDNA3 graphics card. AMD RDNA3 GPU, Source: AMD Igor does not want to engage into fight with leakers sharing various specs for AMD or NVIDIA GPUs. Those have been all over the place...

videocardz.com

 

[arandomguy]

Is he only responsible for GPU marketing? Seems like he tweets about CPUs/Ryzen as well.

 

[troyan]

Probably a hint towards frequencies vs unit count.

nVidia has both: 50% higher clock rates and 67% more compute units...

 

[Nebuchadnezzar]

nVidia has both: 50% higher clock rates and 67% more compute units...

They'll still be at around 30% clock disadvantage if things go as rumoured.

 

[TopSpoiler]

However, this also limits the maximum board power of the card to 450 watts, whereby the actual TBP should be far below that.

Why doesn't Igor count additional 75 watts from the PCIe slot?

 

[Nebuchadnezzar]

Why doesn't Igor count additional 75 watts from the PCIe slot?

Those high-end cards don't actually draw much if anything from the PCIe slot, it's around 10W.

 

[TopSpoiler]

Those high-end cards don't actually draw much if anything from the PCIe slot, it's around 10W.

No, it's not. For example, RTX 3090 with two 8-pins draws 50~60 watts from the slot.

source: https://www.thefpsreview.com/2022/0...tx-3090-founders-edition-video-card-review/3/

 

[Nebuchadnezzar]

Because those cards are under-provisioned / too high TDP, almost anything else is 10-15W. Slot power tends to be last in load distribution.

 

[TopSpoiler]

Slot power tends to be last in load distribution.

Yes. So there's no reason to not count slot power for estimating the possible max TGP/TBP.

 

[Nebuchadnezzar]

Yes. So there's no reason to not count slot power for estimating the possible max TGP/TBP.

Except for the fact that he has the actual schematics and it's immediately visible if the slot is load bearing or not, I suspect he didn't make a mistake.

 

[Frenetic Pony]

Probably a hint towards frequencies vs unit count.

Nah, it's the gigantic form factor, released literally the same day just after the leaked pics of Mr Behemoth. It's a dumb choice of words though for real.

 

[Jawed]

Right. Rendering a frame is an intricate dance where the the bottleneck shifts every few milliseconds. Even if dual-issue isn’t possible that on its own wouldn’t explain the lack of scaling in games.

In light of about 3x the FLOPS, it's bizarre.

Most passes are bandwidth or occupancy limited anyway.

I'm expecting good stuff on those points from NVidia.

 

[Leoneazzurro5]

Almost 4 GHz? This is the latest rumor. That would also mean 90+TF.

https://twitter.com/x/status/1571834170900623360

 

[pjbliverpool]

That would align with that earlier comment of "bigger doesn't mean better'.

 

[trinibwoy]

That would also be a lot bigger clock speed gap than RDNA 2 vs Ampere.

 

[fehu]

How much is it credible?

 

[Bondrewd]

How much is it credible?

+- sorta there.

That would also mean 90+TF.

Fast but not quite fast enough.