I watched the 7900XTX presentation and there was a lot that couldn't be said, because the AIB partners have not announced any cards.
Unfortunately the presentation left a lot to be desired.
Some people might be disappointed in what was shown especially with regard to performance and truth be told, AMD should get someone in to polish their scripts.
For instance I got the joke with the "8K" and I am sure the material looked pretty good on paper, but there wasn't the comedic timing present.
So here is how I would have introduced it:
"Two years ago our competitor spoke about "8K Gaming" in their presentation and I suppose they were right - if your game engine is PowerPoint. Now however .."
And that would have set up Scott Herkelman for the rest of the presentation.
Another thing, AMD really needs to vet the script for the next presentation to make sure that what is quoted from it is not the equivalent of shooting oneself in the foot with a pistol set to full auto.
So here is the bottom line with regard to my thinking on the 7900XTX.
From my own personal knowledge going from an optimally configured 5950X (7nm) with a CBR23 score of 31,105 at about 180 Watts CPU power to an optimally configured 7950X (5nm) with a CBR23 score of 40,210 at about 177 Watts CPU power (and no, not at 95 degrees, but rather at 78 degrees Celsius) so at the same wattage and temp (I have the same AIO cooler on both, in fact aside from Mobo CPU and RAM my 5950X and 7950X systems are identical) the 7950X has at least a 30% performance gain over the 5950X.
In other words, if AMD had done nothing else except shrink the 6900XT down from 7nm to 5nm you can take any benchmark results you see out there and multiply them by 1.3.
But wait, that's not all; all things being equal the 7900XTX has 96 CUs, the 6900XT has 80 (how they work is a bit confused at the moment, so I am not factoring anything related to that into the equation yet). So that is 20% more CUs.
Thus we take the 1.3 and multiply that by 1.2 = 1.56
So if AMD had done nothing except sit on their backsides and scaled down the 6900XT from 7nm to 5nm and added some CUs then you would expect to see the benchmark results for the 7900XTX to be 1.56 times the result of the 6900XT.
Now it gets a bit trickier, but we can also apply some simple arithmetic.
In the AMD presentation the 7900XTX was compared to the 6950X and the lowest number they showed was a 1.5 time increase in gaming performance.
OK so the difference between the 6900XT and 6950X is about 6% in performance.
So now we take the number above for the 6900XT and multiply that by 1.06 = 1.6536
The 7900XTX was clocked at 2.3 GHz for the presentation and the 6950XT number that I see for the same performance level out of the box (the "Game Clock") is 2100. So simple arithmetic would be 2100/2300 = 0.913. This will give us an estimate for the architectural uplift in performance.
So now we can make a guesstimate of how much performance you could expect out of the 7900XTX when it is not running at stock but rather overclocked (or rather at the level of the Nvidia Boost 3.0) which would be:
(1.6536 x 1.5) x 0.913 = 2.265 times the performance of the 6900XT.
I wasn't too far off on my prediction for the 7950X which I made in June (I tried to be conservative then just as I am now), when the name of the CPU hadn't been finalized and I will be excited to see how close I came with this to the actual performance of the 7900XTX.
You can see my prediction for the 7950X and how close I came in reality here:
Prediction for the 7950X
