I would personally wait on buying a new card.
Save some cash while waiting and splurge on the card later.
Also, GTC 2016 is only 10 weeks away. We'll know more by then.
I don't usually recommend waiting, since, you know, there's always that next thing just behind the corner!
But this is a slightly different situation considering the manufacturing node we've been stuck with for a while.
If you can't wait then, well, buy the fastest card you can afford right now and don't look back. Either new or used.
Or go SLI and get 1.5-2x more performance (around 970/980 level) than you now have with minimal investment if you don't mind dealing with double the cards, heat and noise and potential SLI problems..
Here in Finland (the land of thousand lakes and ridiculously expensive used tech) the 670's go for 100-150€ depending on the model. That's 110-160 USD for you americans, cheapest 670 I found on Ebay.com was an EVGA FTW model for 65 USD. I'd buy that in a heart beat.
But commence rambling:
The smaller manufacturing node (14/16nm FinFET) is going to be amazing.
- Same performance + lower power consumption
- Higher performance + same power consumption
We've been stuck on the damn 28nm manufacturing node for four years now, think about what the node shrink is gonna give us!
(Power consumption numbers and relative performance figures taken from TechPowerUp)
Remember the GTX 580 > Titan jump.
Performance on 1920x1200 resolution
GTX 580 = 58% off of Titan (100%)
Performance on 2560x1600
GTX 580 = 54% off of Titan (100%)
Average/peak power consumption under gaming load:
GTX 580 : 214W/229W
GTX Titan : 208W/238W
That's insane.
And yes, I'm not comparing GTX 580 to a GTX 680 since GTX 680 is a main stream chip (GK104) where as GTX 580 was a high end chip (GF110).
GF110 to GK110 is the logical comparison, big core vs big core.
Although this is a comparison of full core vs partial core.
Perfect comparison would be the GTX 480 vs Titan and GTX 580 vs Titan Black.
Unfortunately TechPowerUp didn't include the 480 to the Titan reviews relative performance graphs AND TechPowerUp hasn't tested the Titan Black so...580 vs Titan will do..
Or we could compare GF114 (560Ti) to GK104 (680)
In that case
Performance on 1920x1200 resolution
GTX 560Ti = 59% off of 680 (100%)
Performance on 2560x1600
GTX 560Ti = 53% off of 680 (100%)
Average/peak power consumption under gaming load:
GTX 560Ti : 148W/159W
GTX 680 : 166W/186W
Since then without any node shrinks we've gotten these results with
- tweaking the architecture
- mastering the manufacturing process (improving yields and so on)
Which allowed for
- better utilization of the shaders
- throwing more (+35%) area (read shaders) at the problem (which is 28nm)
While keeping the power consumption in check, at least with the main stream chip.. GK110 to GM200 is a different story..
GTX 680 > GTX 980 (GK104 > GM104)
Performance on 1920x1080 resolution
GTX 680 = 64% off of 980 (100%)
Performance on 2560x1600
GTX 980 = 63% off of 980 (100%)
Average/peak power consumption under gaming load:
GTX 680 : 166W/175W
GTX 980 : 156W/184W
Where as GTX Titan > Titan X (GK110 > GM200) is a different story like I said.
This is just about the limit of what you can do on 28nm, and it shows. ~550-560mm^2 on 28nm was nuts, 600mm^2 is outright ridiculous.
The power consumption didn't "quite" stay in check. Look at the average power consumption.
Performance on 1920x1080 resolution
GTX Titan = 64% off of Titan X(100%)
Performance on 2560x1440
GTX Titan = 62% off of Titan X (100%)
Average/peak power consumption under gaming load:
GTX Titan : 186W/252W
GTX Titan X : 223W/243W
If you look at Titan X vs 980 Ti, you'll see.
Performance on 1920x1080 resolution
GTX Titan X = 103% off of 980 Ti (100%)
Performance on 2560x1440
GTX Titan X = 104% off of 980 Ti (100%)
Average/peak power consumption under gaming load:
GTX 980 Ti : 211W/238W
GTX Titan X : 223W/243W
tl;dr?
With Pascal + 16nm FinFET:
I'm going to expect similar jumps in performance we've seen in the past with 580>Titan, 560 Ti>680, 680>980, Titan>Titan X