Originally Posted by ZealotKi11er
Well the leaks show that its not as fast as 1070. Probably RX 590 speeds.
It should definitely be closer to a gtx 1070 than an RX 590.
The leaks already shows this. The GTX 1070 is 7% faster in this than the RTX/GTX 2060 benchmark which is closer to the gtx 1070 than the RX 590.
If the RTX 2060 has 1536 shaders and 192 bit bus, it has the same component ratio of the GTX 1060 to GTX 1070. This being 50% more resource.
However because the performance is CPU bound and the decrease in efficiency in an architecture as it scales up, this is significantly lower in practice.
The GTX 1070 is 34.8% faster than the GTX 1060.
So if the differential in performance is the same, to get the performance of the 2060, we simply have to divide the 2070's by the 1.348 which translates into 160/1.348 = 118.69 which is very similar to the GTX 1070s score of 120.
Another calculation to determine the 2060's performance is to determine the IPC gain of RTX in terms of performance per tflop.
To ensure we are not CPU bound to make this calculation, we have to look at the 4k Results. This is fair for the RTX 2060 because it is unlikely to be CPU bound like the RTX 2070/280 at 1080p meaning it will be able to achieve it's optimum performance per flop unlike the GTX 2070/2080 at CPU bound(which is illustrated by the 1.35 performance advantage of the GTX 1070 over the 1060).
So if we consider the GTX 1080 at 9.01 tflop card considering the GTX 1080 1760mhz running clock and the RTX 2080 as a 11.18(the actual clocks on the cards at 1900mhz) and look at the difference in performance in the same chart below(notes most of these clocks are coming from Ananddtech),
We see there is a 44.1% gain in performance. From the above, you can determine the gtx 2080 has a 1.242 tflop advantage
1.44.1= tflop advantage* performance per tflop 144.1/ normalized tflop advantage = perfomance per tflop.
1.441 /1.242 = 1.160
Similarly for the RTX 2070 vs gtx 1070, 1.4333 difference in performance.
The 1070 has clocks about 1779mhz(6.83tflops) and the rtx 2070 has clocks of 1803mhz(8.31tflops) = 1.217
1.43/1.217 = 1.178 advantage in t flops. This being higher should be no surprise as an architecture scales worse per tflop. So lets take this performance per tflop and times it by the tflop ration difference between the gtx 1060 and 2060.
Lets make this last calculation simpler and assume the clocks will be the same and the difference will be entirely related by the number of cores. Since he RTX 2060 has 1536 cores and the gtx 1060 has 1280 cores, the RTX 2060 has 1.2times the number of cores and lets assume for now tflops.
Take an aggregate average from the above and say the 1.17 advantage for performance per tflop.
To get the actual performance, you take 1.17 *1.2 = 1.404.
This means that a RTX 2060 should be 40.4% faster than a GTX 1060 atleast in less CPU bound situations. This puts it around GTX 1070 level in performance which I suspect it will be at. Definitely not RX 590 levels of performance because for that to happen, the RTX 2060 would have to a regression in performance per tflop because the 20% increase in cores should put it slightly above an RX 590 alone, add in the increase performance/tflop and you got something that should be substantially faster. What I suspect will happen is the 2060 will be about 35% faster than a gtx 1060 at 1080p but end up being 40% faster at 1440P.