Na Chiphell`u piszą że z 5080 vs 4090 może być podobna sytuacja jak dawniej pomiędzy GTX 1080 ti a RTX 2080 .Według mnie może być coś w tym na rzeczy.
"5080 may reproduce the situation of 2080=1080ti, with a higher frequency, and then the actual game is comparable or slightly higher than 4090.
But 4090 is more than just running points. In the AI era, computing power = CUDA scale growth, so Huang can keep the architecture unchanged for many years. Because the game field is not important, the established AI ecosystem is very important, so the optimized code does not need to be changed, and the own cudnn and tensor rt can also maintain compatibility. If you find a pytorch benchmark, 4090 has no difficulty in surpassing 5080, the gap = CUDA scale gap"
Żródło : www.chiphell.com