AMD: Navi Speculation, Rumours and Discussion [2019-2020]

Video for those, like me, that had no idea what this was referencing.

board shot at ~4:50

 

Oh sorry, with all the boom this cause I though everyone knew already

 

Well, isn't that 5 or 6 so far we've descovered, relating to RDNA now... ?

Dr Su is a Gamer. She was also given (ie: She gave herself) the oportunity of a lifetime, to designed a dGPU gaming chip, that is 100% towards Modern day Gaming. And in that pursuit, She split the division and architectures to do so (CDNA/RDNA).

I suspect, that rdna has been done a while, but AMD themselves were waiting to leverage it's "abilities" using a more mature "Infinity Fabric".. and make use of some really forward thinking metal, to make a smrter front end, to pump the back-end & using higher bandwidth, etc.

The XSX is the base of Gaming right now (essentially & looking forward) and a discrete graphics card using the same architecture and scaled up.... will surely be 2x or 3x graphically more powerful.

How RDNA handles the latest game engines is going to be most interesting to me.... because I think that is where Dr Su put her efforts.

 

I wonder how this plays with UE5 Nanite given that Nanite is accomplished mostly through compute. Is it even possible to use dedicated geometry hardware to accelerate something like Nanite?

 

Now that we have a picture of the PS5 die I think it's safe to say that if PS5 and XSX have Infinity Cache, it has not added huge chunks to their die areas.

It may be that Infinity Cache is simply not appropriate in a console chip. Or maybe they are both just too old in their design to inherit Infinity Cache. In the same way as both are too old to use Zen 3.

 

Zen3 is rumored to have a rather large IPC improvement.... infinity cache something to do with that?

 

Well If it supposedly helps improving the memory bandwidth, having that on a relatively slow ddr4 would bring bigger benefits, isn't it?

 

Or Infinity cache is the L1 shared by all CUs patent thing. That only adds 0,09 mm2 per CU.

 

Zen 3 effectively doubles the L3 available for a single core. 32MB is pretty massive.

 

That could be. Caches are the hot thing in increasing perf/watt/cycle.

 

Another rumor by RGT, the 8MB of L3 in PS5 CPU is unified.

 

Or another iteration of Fiji Vega based Radeon Pro SSG: https://www.extremetech.com/extreme...ew-ssg-a-gpu-with-1tb-of-ssd-storage-attached

 

My feeling is that if AMD is marketing something with Infinity in the name, it would be somehow interacting with the Infinity Fabric, which that shared L1 concept would not do. Then again, AMD called Zen's L3 gamecache or something, so marketing could slather the name onto anything.

 

Given current buzzword trends they ought to call it AI Cache.

 

What graphics card will AMD use to demo Zen 3 gaming prowess today?

 

Hopefully 6900XTX. Or 9700XT/Pro.

 

I was more surprised by the size of the heatsink. That is one beefy cooling system.

Quite possible, the console chips likely taped out at least a few months before the desktop chips. Use of Zen2 certainly suggests as much.

Seems unlikely. XSX is using the 4 core CCX with 4 MB L3 just like Renoir and I'd be very surprised if PS5 isnt using the same physical design.

5700XT at 1080p would be my guess, since that's where the CPU plays much of role in gaming performance.

Using a nice AGP to PCIE converter I'm sure

 

Considering the rumors that RDNA 3 is going to be chiplet based, it would make sense to have a big Infinity Cache to mitigate the downsides of the chip to chip Infinity Fabric.