GTX 1050 is not much more, so that's a good, more modern comparison. hell, the upcoming 1630 is probably better. that's actually closer to what I expected from Dane at the top end
GTX 1050 is a good one, and was going to be my next step if I didn't get off of work lol. (Hey, everything I could do for the day was finished, don't judge lol).
It only has 640 shader cores, but has 50% higher clock speeds, and is a pascal architecture.
I really like you bringing up the 1630 though. 1630 is Turing, which is amperes predecessor. Now we are just one gen away
This one has even less cuda cores at 530, it uses it's smaller fab node size to increase clocks nearly 2x the Liverpool, to get that comparable performance. We know that's not the route drake's going to take though.
We are, to my knowledge, lacking an ampere gaming product going low enough to match the general fill and flop performance of the Liverpool and 1630. But since we are only one gen apart , we have psuedo direct comparisons from the maker! (Caveat being this is essentially marketing material from the maker)
Here is a comparison that stuck with me. Nvidias Jensen claimed Ampere has 2X the performance per watt as Turing.
Now the way Jensen arrived at this claim, is a little sneaky, it's not about an increase in fps per watt, which would actually be a 30 something percent increase, but by matching like for like between Ampere and Turing, with ampere matching the performance of Turing. So if I am grokking them right, you take a hypothetical ampere gpu, with the same component specifications as the Turing gpu, and run the same game, matching the same exact performance, and Ampere can do it using half the power draw.
so Jensens boast goes, that 530 cuda Core 75 watt 1630, a hypothetical 530 cuda Core Ampere could match it's performance at around 37 watts. I don't know which fab node/manufacturer he was basing this off of though? My guess would be the most favorable one for the comparison at the time, so TSMC N7 over Samsung 8nm. I know that if we doubled the cores, like 1,060, we could match the performance at an even lower power draw, but it's not linear and I don't know the curve.
Of course the desktop ampere gpus rebuked the power savings and used more transistors and more power draw, for big gains.
For drake we are definitely getting the transistor increase in spades..... But the focus will be on power draw/clocks that runs off a battery... So the traditional better performance at lower power draw. Something more directly in line with Jensens original boast of Turing performance at half the power draw.
But, without any hard info on clocks or fab, this is where I run out of road.