Navi 21: AMD’s Flagship Radeon RX 6000 Series Die Dwarfs Navi 10 in Alleged Size Comparison

[Tsing]

Image: AMD



Coreteks has shared an alleged rendition of AMD’s new Navi 21 die. The chip, which will power red team’s flagship Radeon RX 6000 Series graphics card (Radeon RX 6900 XT?), looks rather massive next to the previous generation’s Navi 10 (Radeon RX 5600/5700/XT Series) die.



Assuming that the approximation is correct, the name “Big Navi” makes more sense than ever. Coreteks claims that the Navi 21 die is around 536 mm2 (29 mm tall by 18.55 mm wide) with 26.3 billion transistors. In comparison, the Navi 10 die only measures 251 mm2...

Continue reading...

 

[Armenius]

It supposedly has 5120 cores, double that of the 5700 XT, so it's no surprise that the die is is also about double the size.

 

[Uvilla]

So no mcm yet huh? Im sure its coming.

 

[LeRoy_Blanchard]

Does double the size also mean double or more the performance though?

 

[Armenius]

So no mcm yet huh? Im sure its coming.

Have not heard anything about MCM recently from either AMD or NVIDIA in a while despite people hyping it up. I would think that implementing it would be a lot harder to do with a GPU than it is with a CPU, so I don't think we'll be seeing it anytime soon.

 

[Auer]

Does double the size also mean double or more the performance though?

That's what I want to know too.

Also....RT..? DLSS equivalent...?

 

[LazyGamer]

That's what I want to know too.

Also....RT..? DLSS equivalent...?

RTX voice? Intel has their answer coming, apparently. How about content-creation driver issues?

Wife's box has occasional crashiness with Adobe stuff... and has an RX460. I'm tempted to throw the 1050Ti I use for Plex transcode streaming in there just to see.

 

[LazyGamer]

Have not heard anything about MCM recently from either AMD or NVIDIA in a while despite people hyping it up. I would think that implementing it would be a lot harder to do with a GPU than it is with a CPU, so I don't think we'll be seeing it anytime soon.

I expect so. If the ROI was there, they'd likely already be using it, but relative to CPUs they're looking at an order of magnitude more throughput for GPUs.

It might make sense if they could do the MCM on an interposer a la HBM, but interposer assembly has been its own challenge.