This is mostly directed at Zoltan but, That feel when your 1080 is slower than each of your sea of 980ti's. Do you think you could explain this? Is this down to software optimization or perhaps the core of the 980ti is slightly faster and the memory is not fully utilized or needed in these WUs. I know a lot of the FPS gain in games for the new generation is down to a new compression technique which is not useful for this type of workload, so perhaps the 1080 is, in fact, slower than the 980ti? Everyone please chime in.

your 1080 is slower than each of your sea of 980ti's. Do you think you could explain this?

... so perhaps the 1080 is, in fact, slower than the 980ti?

Jumping in here because I previously ran a 980Ti and sold it to buy a 1080. My main observation is that while the 980Ti always ran in full boost speed on GPUGrid, my 1080 never jumps above its base clock. This has certainly made a larger difference between the cards than should exist. The card WILL boost, as it boosts often in games and there a couple PrimeGrid apps which cause it to always run at full boost, but all the other BOINC projects I've tried (GPUGrid, Einstein, SETI) never get the card running at top speed.

I don't know if this is because the projects are just not pushing the card hard enough for it to boost its speed or if there is something odd about my individual card (EVGA 1080 SC).

your 1080 is slower than each of your sea of 980ti's. Do you think you could explain this?

Yes, I think I can explain this. You didn't asked for the explanation itself, but I'm sure you'd like to know. :)
The major factor is WDDM (Windows Display Driver Model). This makes the OS (Windows Vista and up) more stable, as a display driver failure ("The display driver has stopped responding....") won't bring the whole OS down, but it puts a little delay into every interaction between the OS and the GPU, which makes the GPUGrid app slower. Different GPUGrid batches have different amount of CPU-GPU interaction (the CPU does the calculations which need to be done in double precision), but it's usually more than of games. I'm using Windows XP x64 because it does not have the WDDM architecture, which makes my GPUs crunch faster, but the Pascal cards does not work under Windows XP. I can do it because I don't use my PCs for such purposes which need a modern OS, or when I do I simply reboot my PC with Windows 10 (I also have laptops for such tasks).
The minor factor is that the Linux Pascal app does not respond to the SWAN_SYNC setting, which have different impact for different batches (the current one (SDOERR_BNBS) is less affected).
The third factor is that the GTX 1080 is a replacement for GTX 980, not for the GTX 980Ti, so it's an inadequate expectation that the GTX 1080 should be faster than a GTX 980Ti (regardless that it is faster in other applications like games). Truth is that judging by their specifications I've expected that the GTX 1080 will be faster than a GTX 980Ti. They are almost equal under WDDM OS. The real replacement for GTX 980Ti has not been released yet, but judging by the performance of the TITAN X (Pascal) cards the GTX 1080Ti (if it will ever be released at all) will be faster than a GTX 980Ti, even if the GTX 980Ti runs under a non-WDDM OS.

... so perhaps the 1080 is, in fact, slower than the 980ti?

The GigaFlops/Watt ratio is much more important, than the card's speed, and the GTX 1080 is way much better than the GTX 980Ti in this respect. But the facts are stubborn things, and there was hardly any GTX 1080 which was faster than my GTX980Ti's (under Linux perhaps). While the WDDM can't be turned off, the Linux app should be modified to respond to the SWAN_SYNC (or it should be automatically turned on for high-end cards) to fully utilize the high-end Pascals.

A 1070 is the replacement for a 980Ti in regards to performance. I have both and they crunch pretty similarly. The 1080/Ti are above the 980Ti.