It bugs me so much that they're inconsistent in how they calculate throughput. On one hand you have TITAN X and 980TI whose compute throughput is calculated AT BASE CLOCK On the other hand you have 1080/1070 whose compute throughput is calculated at boost clock. What the **** NVIDIA. What the ****. What I'm interested in seeing is improvement in Pascal (over Maxwell) at throughput parity. Will GP104 suffer because of reduced memory bandwidth + fewer ROPs/TMUs ? Will 6.5tflops on Pascal do better than 6.5tflops on Maxwell in AotS for example ?