We have another slide for you, but this time bit more technical.
NVIDIA Turing gets a bigger L2 cache
For Turing NVIDIA doubled the bandwidth for the L1 cache. Architectural design changes improved the latency and capacity of L1 cache, which is now 2.7 times as big as Pascal’s.
The L2 cache has been upgraded to 6MB, which is 3MB more than Pascal.
This slide was presented at Editors’ Day yesterday in Cologne. You are not supposed to see it till September 14, so don’t tell anyone.