These days VideoCardz is the best place for video card leaks. This time the site got its hands on another slide from yesterday's NVIDIA Editor's Day, this one explains the new cache and shared memory architecture of Turing.
In a nutshell, Turing has double the L1 cache bandwidth, a lower L1 hit latency, up to 2.7x higher L1 capacity, and double the L2 capacity than Pascal. The latter got upgraded from 3MB to 6MB.
NVIDIA Turing has bigger L1 and L2 caches than Pascal
Posted on Thursday, August 23 2018 @ 22:15 CEST by Thomas De Maesschalck