05-07-2016, 09:26 PM
(05-07-2016, 09:05 AM)Flomac Wrote: I respect your knowledge but in this case you were miles away from being right.
I was joking but I also wasn't "miles away from being right." Looks like you've got it all twisted.
Our previous conversation on these forums was about Tesla P100 vs Titan X (or a GTX card closely resembling P100.) Given the fact that there is zero overclocking headroom on the P100, I still maintain the P100 is at most 14-16% faster with hashcat. Our conversation was not about GTX 1080 vs GTX 980, or GTX 1080 vs Titan X. And indeed, the GTX 1080 does not even remotely resemble P100. It looks a hell of a lot better.
On IRC I've stated numerous times that I expected the GTX 1070 to perform about like a GTX 980, maybe midway between a GTX 980 and 980Ti, and the GTX 1080 to perform about like a GTX 980Ti, maybe midway between 980Ti and Titan X. And I was correct:
GTX Titan X - 3072 * 1515 = 4654080 MIOPS
GTX 980Ti - 2816 * 1515 = 4266240 MIOPS
GTX 1080 - 2560 * 1733 = 4436480 MIOPS
That literally puts it smack-dab in the middle between the two, which meets my expectations.
The x-factor here, something I didn't predict, is that this card can apparently run at 2114 MHz, maybe even higher. That's absolutely insane, nobody predicted this. 1800 MHz maybe, but 2100+ MHz? No fucking way.
Now, we still have yet to verify that the non-founders edition can be overclocked that high, and we have yet to verify that the clock can actually be sustained that high when running hashcat. But assuming it can,
GTX 1080 - 2560 * 2114 = 5411840 MIOPS
Fuck me, that's fast. But even if we can't sustain 2114 MHz while running hashcat, we only need 1818 MHz to be as fast as the Titan X, which should certainly be no problem.
And this is the one thing I did get wrong: I said we likely wouldn't see a card faster than the Titan X until the "GTX 1080 Ti" (or whatever they will call it) is released 6 months later. Now mind you the GTX 1080 probably won't be much faster than Titan X, only just, but it looks like it will indeed be as fast and possibly up to 10-16% faster, depending on where we can pin the clocks.
So no, I absolutely wasn't miles away from being right. I was actually pretty damn close to right. But nobody predicted 2100+ MHz clocks.