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

Status
Not open for further replies.
https://videocardz.com/newz/amd-navi-23-graphics-processor-gets-gfx1032-id

The latest news is that on Twitter rogame today confirmed the GFX level for the Navi 23 graphics processor. The new Navi chip is now listed in between Navi 22 and Van Gogh as GFX1032.

Navi 21 also known as Big Navi has the GFX ID of 1030. One ID above sits Navi 22, which at this point is easy to guess it will succeed Navi 12. In case you missed the story, this chip is a custom GPU for Apple MacBook. The Navi 22 is likely to follow suit for next-gen MacBook.
 
So you think power consumption is irrelevant?
Depends on the context. HBM is mainly used there is a need for, well, high bandwidth.

There was of course that Intel-AMD processor and Navi 12 where HBM was in fact used to optimize power consumption, but those were premium products, where margins are high anyway.
 
20200193681
MECHANISM FOR SUPPORTING DISCARD FUNCTIONALITY IN A RAY TRACING CONTEXT
That patent hints that AMD is favoring a D-buffer like approach with a fully decoupled traversal hardware, as opposed to the inlined traversal approach Nvidia has opted for. As far as I understand, that approach would not be compatible with the inline ray casts as exposed by DXR 1.1, but in return is in line with the rumored minimal impact of ray tracing on RDNA2 platform.

Any estimates how plausible it is that this patent was already implemented in RDNA2?
 
Any estimates how plausible it is that this patent was already implemented in RDNA2?
Very high probability this was already implemented.
Unlike CPU teams which file for patents at least 3 years before implementation, the GPU teams file for patents and implement them even close to tapeout.
Also they seem to be more tolerant to minor silicon bugs which they work around in the compiler(GFX1010 is a prime example), in contrast to CPU teams which have very low tolerance to errata.
This patent was filed in December 2018.

As far as I understand, that approach would not be compatible with the inline ray casts as exposed by DXR 1.1
Isn't that in contrast to what Dave Oldcorn said in the DXR U YT video?
 
Very high probability this was already implemented.
Unlike CPU teams which file for patents at least 3 years before implementation, the GPU teams file for patents and implement them even close to product tapeout.
Also they seem to be more tolerant to minor silicon bugs which they work around in the compiler(GFX1010 is a prime example), in contrast to CPU teams which have very low tolerance to errata.
This patent was filed in December 2018.
Big companies usually produce a lot of patents, many of them are never implemented in anything.
 
Big companies usually produce a lot of patents, many of them are never implemented in anything.
Same can be said otherwise. We can list a lot which where implemented. Especially for CPUs. For a product in 2020 those patents would have been filed 1.5 years at least.
 
Do we have any idea of what Van Gogh is (other than it being an APU) and where we will see it? It also has a LITE option does that mean it's semi-custom for Sony/Microsoft/Chinese console or is it something for Apple?

Microsoft are rumoured to have a lower end console so i was thinking it would be for that. Sony also had the LITE codename for its console with NAVI 10 LITE and to my knowledge it hasn't revealed itself to be an actual product so that's why I'm thinking Its a console part.

It also not appearing in any of the roadmaps is mysterious.
 
Do we have any idea of what Van Gogh is (other than it being an APU) and where we will see it?
Mean, maybe green.
And soon-ish. Pretty sure soon-ish.
It also has a LITE option does that mean it's semi-custom for Sony/Microsoft/Chinese console or is it something for Apple?
LITE doesn't indicate console relationship as is, we've had other LITE stuff that's distinctly not consoles.
Also Apple is going in-house and they'll never forgive AMD anyway.
It also not appearing in any of the roadmaps is mysterious.
Bold of you to assume AMD has a concrete APU roadmap.
 
That patent hints that AMD is favoring a D-buffer like approach with a fully decoupled traversal hardware, as opposed to the inlined traversal approach Nvidia has opted for. As far as I understand, that approach would not be compatible with the inline ray casts as exposed by DXR 1.1, but in return is in line with the rumored minimal impact of ray tracing on RDNA2 platform.
Maybe that's what makes Big Navi „big“? But anyhow, if you're right, we would need an update to DXR for RDNA2. Proprietary stuff is not the colours AMD is flying.
 
Apple will never forgive AMD for.. a 9 year-old APU with a CPU architecture that Apple never adopted?
 
Status
Not open for further replies.
Back
Top