Por que va a pintar peor ? por que el chip se llame TU106 en vez de TU104?
Sacado de reddit:
Ppl are upset about 2070 being a fully enabled TU106 instead of cutdown TU104. But I found the reaction irrational. Whichever name they use, 2070 (3GPC) will be 75% cores of 2080 (4GPC cutdown), the same proportion of 1070 vs 1080. Remember last gen GP106 is 50% of GP104, this gen TU106 is 75% of TU104. Nvidia just decided to make a 50% bigger 106 chip, but the change of chip naming has irritated a lot of ppl.
Eli5 what's the real reason behind the reaction?
FYI, Turing has 64 CUDAs (FP32s) per SM, 12 SMs per GPC. So
TU102 (RTX 6000/8000) has 4608 CUDAs, 72 SMs, 6 GPCs.
TU104 (RTX 5000) has 3072 CUDAs, 48 SMs, 4 GPCs.
TU106 (RTX 2070) has 2304 CUDAs, 36 SMs, 3 GPCs.
So 2080Ti:2080:2070:2060(supposed) = 68:46:36:24.
By comparison:
GP102 (Titan Xp) has 3840 CUDAs, 30 SMs, 6 GPCs.
GP104 (GTX1080) has 2560 CUDAs, 20 SMs, 4 GPCs.
GP106 (GTX1060) has 1280 CUDAs, 10 SMs, 2 GPCs.
So 1080Ti:1080:1070:1060 = 28:20:15:10.
La proporción de chip y de SMs es exactamente la misma que la anterior gen.