4090 vs 7900xtx re-test, who has aged like fine wine?

[Marine-RX179]

There should be no scenario where an RX6700 can be quicker than an RTX3070,or an RX480 compared to an RTX3060.

There is ONE scenario- the performance tanking into oblivion when the vram runs out.

There were situations where the modded 3070 16GB was getting like doubled the frame rate of the 3070 8GB at 1440p and 4K

 

[KompuKare]

Honestly outside being a test vehicle for chiplets,not really sure what RDNA3 bought to the table? It has better machine learning capabilities which appear to be barely used. Also the weird dual issue thing too. So not surprised if drivers are more mature.

I'm sure chiplets will be really important but with their marketshare at an all time low, who in their right mind at RTG would embark on so risky a strategy?
One halo chiplet GPU done "crazily" big (450mm² GCD with cache heavy MCDs would have made far more sense. Then maybe later a volume part to slot below Navi32. But what do we outsiders know!?
Looking at the Blackwell rumours, MCDs with a decent amount of L3 makes a lot sense at 3nm and below where cache doesn't scale and price per mm2 keeps going up.

So many confounding factors, what a waste of time.

I guess what PCGH were not going to do is waste lots of their time resting with the old suite and platform from the launch.
It still has some value, it's that the @Grim5's headline isn't that accurate.

They'll sell it at half capacity and charge a monthly subscription to double your VRAM. You heard it here first. DLSS license sold separately. RTX premium upgrade by the hour.

Don't give... Never mind others have already made that point!

 

[mrk]

It's most likely nothing to do with drivers and more to do with more powerful cpu's around which can get more from the 4090 as it does tend to be bottlenecked more than any other card. A better test would have been the 4080 v 7900xtx as they are similar in raster.

For raster in some games that aren't multithread optimised yes but not for RT in optimised titles where still the gap is big. The optimisation point is massively true and the proof is in Starfield which boosted nvidia performance by up to 60% all due to engine optimisations which Bethesda refused to acknowledge initially telling gamers to upgrade their PCs...

In raster in cyberpunk there is only one set of settings (dlss on but fg off) where my 4090 drops to 81% utilisation and cpu increases to 52% which makes no sense logically as the fps is 122... with max settings and psycho ssr. This could even be a run variance bug as all other modes inc raster, RT and PT the GPU use is over 90% and CPU usage between 30% and 48%

At 3440x1440 the cpu bottleneck is minimal on modern cpus that are capable.