We are well aware that the 7850 is ImSpartacus's favorite GPU. That card has been a decent performer in the sub-150 W TDP range for the past 2 years. But it looks like NVIDIA might shake things up with today's announcement of the GTX 750 and GTX 750 Ti, and not in ways that you might expect. NVIDIA has, at least in roadmaps, a plan to release a new GPU architecture about every two years. So far they’ve been Tesla, Fermi, and Kepler. Their GPUs haven’t been without delays though, and the slide appears to show the "big" compute-heavy chips (e.g. GF100, GK110) instead of the smaller, leaner chips designed primarily for gaming (only the big chips have high DP rate), but it was definitely not unrealistic to expect some sort of Maxwell architecture GPU release in 2014. Recall that in June 2013, with the launch of the GTX 760, NVIDIA presented a roadmap showing no additional retail desktop cards through at least most of the rest of 2013. For a company known for rebranding its parts it seemed a bit weird to me that they didn't just rename existing 600 series parts to put in the lower half of the 700 lineup. It's not like they were waiting for revised Kepler chips since the 770 and 760 used the same GK104 as the 670 and 680. So I made a guess that the sub-760 area was deliberately left empty because it would be filled up by Maxwell parts within a year (although I also thought they would go straight to the 800 series numbers for that). It turns out that NVIDIA is releasing Maxwell in early 2014 (despite some people adamantly claiming that there would only be refreshed Keplers), and the first retail Maxwell chip is codenamed GM107, and it is used in the GTX 750 Ti and the GTX 750, both announced today. A few points should be noted. The "7" in the chip codename indicates a small low-end GPU. The 750 Ti's performance ends up below the GK106-based 660 (more on that below), unlike all other new architectures in at least the last 8 years where the first GPU of the architecture beat the top-end chip of the previous architecture. In addition, the process is 28 nm, not 20 nm, a reflection of the slowing of process nodes lately. However, that’s not the whole story. Some of the architectural changes from Kepler to Maxwell involve reducing the "core" count of each SMM (Maxwell SMX) from 192 to 128 and dividing the SMM into 4 parts. The memory bandwidth might seem a bit low for a chip of this level, but they added much more L2 cache (GK107 has 256 KB, GM107 has 2 MB) to compensate, lowering power consumption compared to if they had used less cache and more bandwidth. The performance of the 750 Ti falls in between the 260X and the 265 on the AMD side and between the 650 Ti and the 660 (slightly below the 650 Ti BOOST) on the NVIDIA side. The 265 is a higher clocked 7850 and overall the 7850 is slightly ahead of the 750 Ti. So the GM107 slots between the GK107 and GK106 in terms of performance, much like how Bonaire is between Cape Verde and Pitcairn. I suspect we will see more of these "in-between" chips in the future. But what may be even more impressive is the performance per watt. The listed TDP of the 750 Ti is 60 W and the 750 Ti is 55 W. Thus, neither the 750 nor the 750 Ti come with external power connectors. While TDP ≠ power consumption, there’s still a decent difference in real power usage and perf/W, which is especially helpful for laptops (when mobile GM107 parts come out). Now these GPUs replace the 650 Ti and probably the BOOST too. The 650 and 660 are still "current" parts. Unlike the former, the latter is surrounded on both sides by 700 series parts, so its number stands out like a sore thumb, at least for now (maybe it’ll be replaced soon). There's also supposedly a smaller GM108 chip coming out soon which might fill in some places below the 750. As for pricing, the 750 Ti comes in at $149 and the 750 comes in at $119, compared to $119 for the 260X and $149 for the 265 after their recent price drop. That’s not good performance/$ compared to AMD’s parts after the price drop, and the 265 stays noticeably ahead of the 750 Ti. However NVIDIA has the advantage of lower power consumption, which may or may not matter to you. Also, the GTX TITAN Black, an update to the GTX TITAN, has been announced. It’s a fully enabled GK110 (the TITAN only had 14/15 SMXs active) with higher clocks, the same TDP and price, and a more black shroud. If the GTX 750 Ti is too cheap for you, maybe consider this card instead. Source: AnandTech.