| JEMC said: Those Tesla/Quadro cards run that fast because Maxwell is rubish when it comes to computing, and they need every MHz to compensate that. This is how I see the situation right now (Note that this is just my opinion, and I don't know as much as I would like): Both professional and gamers want and ask for two basic things, MORE and FASTER, and Nvidia and AMD not only know it, but share that same vision and work in the same direction. What happened with the last gen is that they choose a different focus. With Kepler and Maxwell Nvidia payed more attention the FASTER side of things, while AMD focused on the MORE aspect (probably because they needed the GPGPU capabilities of their GPU to compensate the poor performance of the CPUs in their APUs). And the truth is that it worked well for both camps, offering noticeable improvements over previous cards while remaining competitive with their main competitor. Now, with Pascal and Polaris, I expect them to pay attention to the other side of things. Because of that, I believe that Nvidia will focus on MORE (and the Pascal diagram seems to point in that direction), and AMD will pay more attention to the FASTER aspect. And just like the last time, I expect to see noticeable improvement. In any case, we'll find out in a couple of months. |
Maxwell is perfectly fine in compute performance and in fact had better single precision performance ...
It's double precision performance that's more of a concern with power consumption. The Tesla K40 was only 144MHz slower than the Titan Black but it was running at least 15 watts lower than the Titan Black too ...
If a consumer version of the GP100 existed it would at BEST have 100MHz breathing room even if Nvidia disabled most of the FP64 units ...
Essay Pro