So check my post history...I've been saying between 11-13TFOPS since back in May last year. This won't be the last opportunity for me to say "told ya" before the system launches (cough* RAM, BC).
Anyway, yes the next gen of gaming should be pretty exciting. Given how bottlenecked this generation has been in terms of HDD speed, weak CPUs, and memory bandwidth, we should finally see the game software evolve in some meaning ways. Notice i said "next gen gaming" because it'll apply to PC, consoles, streaming etc. The consoles will up the baseline and the developers will follow with the bump in software which will benefit virtually all platforms. I'm emphasizing the software because the hardware is just metal and plastic...it's the software bump that will really bring the "next gen" experience.
But how much of a graphical leap with 12TFLOPs of RDNA 2 really be? Well consider this: The current flagship AMD GPU on the market is still the Radeon VII which is a Vega (GCN 5) architecture spec'd at roughly 13.8 TFLOPs. The current top RDNA GPU from AMD is the 5700 XT spec'd at roughly 9.75 TFLOPs (10TFLOPs if overclocked i.e. Anniversay Edition). So let's just round up and say 10TFLOPs RDNA vs 14TFLOPS GCN 5 to have easier numbers to work with. Now by most accounts, the performance delta between the Readeon VII and 5700 XT (OC) is anywhere from 3% - 7% (
Radeon VII vs 5700 XT (1),
Radeon VII vs 5700 XT (2)). Let's say 5% on average. First, that's real world data that shows you just how much more efficient RDNA is over GCN (40% more compute power = 5% increase
).
Now a 12TFLOPs RDNA GPU is about 20% higher theoretical performance compared to the 5700 XT. So first of all, that says that the XBSX (and likely PS5) will come with GPU horsepower ahead of anything that AMD will have commercially available at least through the 1 Half of 2020. Compare that to the 7850 GPU in the PS4 at launch in 2013 which was decidedly mid-range by the time PS4 released and AMD had at least 6 GPUs more powerful than it available by that time (7870, 7950, 7970, R9270, R9280, R9 290).
Now when compared to Nvidia, we know that the 5700 XT is within roughly 12-15% of a Nvidia RTX 2080 on average (
link) and within ~18-20% of a 2080 Super on average (
link). Thus a 12TFLOP RDNA based GPU (20% higher perf than the 5700XT) should exceed a RTX 2080 and be very close to matching a RTX 2080 Super.
HOWEVER, all of that is assuming RDNA 1.0! The big X-Factor currently is how much more efficient is RDNA 2 really? Rumors suggest 30-50% which does seem a bit high. But even if we very conservatively say that RDNA 2.0 is 10% more efficient than RDNA 1.0, then a 12 TFLOP RDNA 2.0 GPU should be roughly as performant as a 13.2 TFLOP RDNA 1.0 card. That's >30% increase when compared to the 5700 XT. So yeah, that card would be above a RTX 2080S in real world performance. In fact, that is definitely approaching RTX 2080TI levels which is ~35% faster than a 5700XT in most games (
link). Remember i was very conservative with the efficiency gains of RDNA 2.0. If the gains exceed 10% and is even closer to 20%, then folks we would be looking at RTX 2080TI level perf in a console for real!
Now everyone is throwing around the TFLOPs figures trying to make sense of it. The truth is that TFLOPS are just a theoretically calculation that by itself cannot be used to determine actual game performance. But notice what I did here was use actual game performance data from commercially available cards to estimate performance. This takes much (but not all) of the "theory" out of the numbers.
Expect great things next gen. Exciting times ahead!
