You seem to foget it has almost 1x more SMX, more ROPs, higher texture fillrate and higher vram bandwidth.. Those clock predictions mean squat.
It's already concluded on that calculation mate. You've could understand that already, from the theoretical 50% you cannot only lose 8.5% from the almost 300mhz core down. Stop the dreaming of twice 670/680 performance it's impossible with a refresh, Everybody should understand the difference between refresh and next gen.
The clocks rate times shader is pretty accurate guestimate of performance. 732mhz on the '780' would put it in the ballpark of a GTX 680 boosted to about 1280mhz. Yeah it's got more, but its got more running significantly slower.
Ain't gonna happen, that clock is for tesla with 225TDP, Titan will have bigger wattage - thus higher clocks. 2x 670gtx for sure, not lousy 40-50%.
If the bandwith difference between Titan and 680 is 50%, then the performance difference will be 50%.
So clocks mean nothing? That's new. Bandwidth it's not the most important as it's a value of link-to-transfer data, a number to give the performance of the bits/bytes diode. Speed it's also an equally critical factor.
But we don't know what the clock speed is yet. I personally don't see i being much more 732mhz myself as i don't think Nvidia will want to push the card anywhere near 300w TDP, and who is to say if it can go actually go much higher than 732.
"Bandwidth" it's a multiply of speed * bus width. Bus width matters of how much utilization of GPU happens. Also it's the maximum difference, so it's up to 50% as soon as proper utilization is going on. Up to 50% difference it's not the same with at least or standard 50% difference. There is always a minimal loss because of electric circuit. The average difference with 680 will be among ~41.5% - ~ 49.5% depending on title.
Ok you say ~ 40 - 50%, im still confident it will be much higher then that btw its not a refresh but a completely different 28nm Kepler design (high-end). Fermi GF104 refresh was a GF114 and not GF110.
I'm with TJ, it'll be x2 670s and perhaps be the fastest card this year. For me though the gtx 780 will be more appealing, or even if i get another 680 i'm sure they'll hold up against this new kid.
It will definitely be fast, i doubt it will be as fast as some think though, it would be poor business from Nvidia to make it too fast plus claims are all based on best case scenario which never turns out to be accurate. The fact its a 3k Tesla GPU cut down to a cheap gaming one suggests a stop gap for the real GTX780 which will be cheaper with similar performance.
i don't think the 4gb on the 680 was completely pointless especially to those who play at very high res beyond the 1920x1080 (like 3800x200 or so). however the card wasn't as fast as the 2GB version in various games, nothing a little o/c couldn't fix, but avg gamers who waited for the 4GB found out it wasn't worth the wait, more vram may be useful but not at the expense of performance, every single fps counts. as for me, my gtx690 is more than enough, of course gonna have to wait and see the real numbers of the Titan, but the numbers i reckon are going to be too close to the 690 that it won't be worth an upgrade. a Titan dual-gpu though would annihilate anything in its path performance wise, let alone it might feature 12GB of vram (6x2) or something, now that's a completely unnecessary and pointless amount of vram.