The following section may be a bit disjointed as I wrote this late at night trying to process the staggered launch of the chips, the reviews, the performance, and my own observations at the time. Please bear with me if you can.
- AMDs (then ATI) last decent lead over Nvidia, was during 2006, with an architecture born from a time before AMDs acquisition of ATI. The (R400) X8** series and the later R5*0 X19** series, saw many successes against Nvidia (Geforce 6000 and 7000 series, respectively) and ultimately won the fixed-pipeline/fixed-shader battle. To end ATIs reign of 2006, Nvidia released the (G80) 8800 GTX. The X1950XTX still managed to trade punches with the hot and noisy G80, but ultimately lost out in performance, particularly when optimizations for the newer GPGPU architecture came about.
- Come May 2007, ATI releases the abysmal (R600) HD 2900 XT; this was hot, noisy and performed worse, in most cases, than the prior R5*0 architecture. Nvidia fixes the G80s issues and releases the (G92) 8800GT that same year in October with ATI quickly releasing their (RV670) HD 3870 to fix the horror which was the R600. The HD3870 was not powerful enough to topple Nvidia's G80s or the later G92s, so ATI, perhaps with a hint of desperation, releases dual-GPU cards to try and take performance crown. To their credit, the HD 3870 at least corrected most the issues with the HD 2900XT. To add further insult to ATIs failings, simply refreshed the G92 for the Geforce 9000 series, possibly enjoying decent profit.
- Mid 2008 comes around and Nvidia releases their new (GT200) GTX 280 just before ATI releases a decent answer to the G92/G80, the (RV770) HD4870. Unfortunately, while the HD4870 finally took the lead from the G92s, it could not match the GT200s so, again, ATI relied on dual-GPU cards to hassle Nvidia latest offerings. This can't be cheap for them to do.
- 2009 sees some refreshing from both sides with the RV790 and GT200b appearing. By the end of 2009, ATI releases their new TerraScale2-based (Cypress XT) HD 5870.
- We had to wait till the beginning of 2010 to see Nvidia's next architecture: the Fermi-based (GF100) GTX 480. While the Fermi took the outright performance title, it came at a cost; the GPU was hot and noisy and to make matter worse, not that much faster than ATIs latest offerings (or less so, if you consider the dual-GPU cards). This was the first time in a long while ATI/AMD had released something that was arguably better than what Nvidia could offer. To try and recover from their embarrassment, Nvidia releases the (GF110) GTX 580 at the tail end of 2010, possibly with the added pressure from AMDs latest Barts XT chips. Luckily for Nvidias sake, December saw AMDs (Cayman XT) HD6970s flop (to a degree); the VLIW4 architecture and performance would simply not scale as expected.
- 2012 is the year something major occurs; it's the first time, even with a staggered launch, that the companies don't go head to head with the best the architecture can offer. January sees AMD release the (Tahiti XT) HD7970, but in response, Nvidia only releases their mid-tier Kepler GK104 as the GTX 680. As I have stated before, the GK110 was revealed the same month of the GK104 release and was released November the same year.
So back to the original argument; while there was always a staggered launch, it wasn't until the last few generations did something like the Kepler v Tahiti occur. A mid-tier GPU going against a top-tier GPU with the same or better performance (until Tahiti XT2 atleast). This meant Nvidia, rather than fully destroying AMDs offerings with the release of the GK110, enjoyed large profits on a marked up mid-tier GPU whilst keeping an illusion of competition. History has now repeated itself with the release of the Maxwells. Is this Nvidia being kind to AMD? Or are they just looking to fool the consumer and enjoy larger profits with a marked-up chip? Price and performance has always conveniently slotted between the two brands, even when such a difference in architecture performance occurs.
