Thanks for the reply. I actually did switch the cards, I forgot to mention that. I was just hoping I wouldn't have to so that I'd have an additional PCI-E slot free.
This is actually my home build, heh. My dept built a decent server with 32 cores and a Titan (thankfully) on my suggestion, so I have that available as well. Long story short.. my univ dept didn't want to budge for the $ for the K6000, so I just bought one off eBay a while back (way below retail) and I'll resell it when I'm done with it. 🙂
In re: ECC, I do not, however I do tend to exceed the > 6GB on occasion with EM simulations. Had a Titan before and overclocked pretty nicely indeed. I can overclock the K6000 as well, but I haven't done too much work on that. I'm going to try the new Kepler BIOS Tweaker eventually.. that one seems to have more control over power restrictions. If I attempted to overclock the K6000 at the level I was able to take the Titan (anywhere from 1075-1150 w/ DP units on and roughly a 400MHz memory underclock), and run a very compute-bound CUDA algorithm, the card throttles, and I'm pretty sure it s because of a power limit breach... seemed to happen pretty much when i hit 230W.
In re: K40m, are there higher application clocks available? What's the maximum clock rate supported? I had seen the 745MHz figure quoted elsewhere, but not sure if NVIDIA would've given more (or higher!) clock options than they did on the K20... but maybe not since the Tesla is usually their 'stable' product.
You mention 'automatic superclocking' -- is that your way of saying you flashed a BIOS with a higher clockspeed? If you know of a way of overclocking recent NVIDIA cards in Linux without a BIOS flash I am all ears.
