RDNA 3 / Navi 31 could have much more power to unlock?

[ryzenmaster]

What do you think about this?

What do you think? whats the potential here for NAVI 31?

 

[Vayra86]

The hardware is definitely there and yes it also feels asthough there is more to refine...

I think it is likely too because these are the first gen gpus on chiplets and AMD is all in on that.

 

[ModEl4]

The resources in theory are there, but the software/drivers side isn't optimized enough (although we are half a year after launch) and developers also need to optimize their engine to RDNA3 which probably isn't happening en masse with the consoles being RDNA2 (even if PS4 Pro launches in Q4 2024 it will probably be RDNA4 based and how close will be to RDNA3 is anybody's guess) Also it seems that doubling the shading throughput it brings only 10% (or less...) general performance increase and the performance realised isn't due to chiplet design only, as monolithic Navi33 will probably also show.
My prediction was that 6nm full navi33 will have at least 4% higher clocks vs 7nm RX6600XT (around 2697MHz) and with 18Gbps memory accompanied it will be around +15.6% faster than RX6600XT (clock difference and 10% difference for doubling the shading throughput), that was my worst case scenario but we already seeing reports for only 11% difference and 2.6GHz frequency, also the naming of the full Navi33 being RX 7600 instead of RX 7600XT doesn't bring much confidence regarding performance so I'm not too optimistic regarding what performance improvements we can expect in the future regarding RDNA3... (Navi31 will see Improvements in performance as CPUs get faster and games more demanding but also RTX 4090, these will not be RDNA3 based improvements)

 

[ryzenmaster]

The resources in theory are there, but the software/drivers side isn't optimized enough (although we are half a year after launch) and developers also need to optimize their engine to RDNA3 which probably isn't happening en masse with the consoles being RDNA2 (even if PS4 Pro launches in Q4 2024 it will probably be RDNA4 based and how close will be to RDNA3 is anybody's guess) Also it seems that doubling the shading throughput it brings only 10% (or less...) general performance increase and the performance realised isn't due to chiplet design only, as monolithic Navi33 will probably also show.
My prediction was that 6nm full navi33 will have at least 4% higher clocks vs 7nm RX6600XT (around 2697MHz) and with 18Gbps memory accompanied it will be around +15.6% faster than RX6600XT (clock difference and 10% difference for doubling the shading throughput), that was my worst case scenario but we already seeing reports for only 11% difference and 2.6GHz frequency, also the naming of the full Navi33 being RX 7600 instead of RX 7600XT doesn't bring much confidence regarding performance so I'm not too optimistic regarding what performance improvements we can expect in the future regarding RDNA3... (Navi31 will see Improvements in performance as CPUs get faster and games more demanding but also RTX 4090, these will not be RDNA3 based improvements)

From chatting with some people in the industry, the thoughts I share with some, is that AMD would as time goes, have more mature drivers, game engines will get more mature with AMD specific APIs/optimizations (AMD put efforts working with studios, as it seems), and together with FSR 3 and Fluid Motion, we might also see the more mature drivers to unlock the first big bang of performance.
As AMD did before, at times unlocking 40% out of no where... lets see! FSR 3 is what? 1-2 months away?