is just Nvidia for better and a little more expensive
, AFAICT. The definition is loose if it exists at all. They mostly use it to differentiate similar models into budget
Regarding grades up and down: next-generation x50 may well match or exceed the old x60 in performance, AND be more power efficient
(which helps cooling, woo!
), AND have modernised I/O
, AND have extra/new software features
(like DX support, effects), etc. - not to mention a fresh warranty period.
It's all about what you want! As you said, you're okay with the 660 ('does a decent job'), but the card is turning into a blower. Well, let's compare to a 'next-gen' 750 Ti, shall we?The Asus GTX 660 2 GB non-Ti
, as reviewed by TPU
, consumed an average of 121 W
and had a peak of 129 W
in the TPU tests. Its relative performance in the performance summary of said tests @1200p was 9%-points higher than the stock
GTX 660 non-Ti. For reference it is also 10%-points ahead of the old 570
in the summary - hilarious, and a good example of how big the generational advantage can be in technology.The Palit StormX Dual 2GB
is the best-rated 750 Ti reviewed by TPU
. It consumed an average of 69 W
and had a peak of 75 W
in the TPU tests. Its relative performance in the summary of said tests @1080p was 14%-points ahead of the stock 750 Ti - and only 6%-points behind the vanilla-660.
Based on this general information, we can observe the following:
The 750 Ti eats up an average of 70 W against the 660's 121 W. 70 / 121 = 0.579, so the 750 Ti is, on average (and under heavy 3D loads), some 40% more power efficient than the 660
(i.e. consumes 40% less power doing the same work). The power consumption advantage translates into performance, too, as the card is only 6% behind the vanilla-660 in the summary @1080p, and hence some 15%-points behind the Asus model. The peaks were so close to the average that they don't need to be analysed separately.So, what is 6%-points less performance in actual figures?
(750 Ti vs Vanilla, Asus review uses 1200p; remember, Asus has a theoretical lead of 15%-points)
- -1.9 FPS in AC4 out of 28.5 FPS total (average, obviously)
- -2.2 FPS in Battlefield 4 out of 35.0 FPS total
- +4.2 FPS in Crysis 3 (advantage: 750 Ti!) out of 22.2 FPS total
- -23,1 FPS in Diablo III out of 124.4 total (so still 100+ on the 750 Ti)
- -0.4 FPS in SC: Blacklist out of 40.0 FPS total.
Sadly, the test suite does not include heavy (or any) RTS titles, so you'll have to look up those figures yourself. But as you can see, the new 750 Ti is already just about as capable as a vanilla 660, and not at all hopeless in games - at least based on average FPS figures, which is not the whole story
(The Tech Report on frame times and FPS as metrics). Basically, FPS alone does not tell you how fluid
the gameplay is, so that is why you have to know frame times, i.e. how much frame delay you get. Delayed network packets make it feel like you're warping, delayed frames make it feel like you're stuck in jell-o.
Did I mention the FPS averages are all at 1080p, 4xAA, and the HIGHEST QUALITY SETTING IN THE GAME? See method: http://www.techpowerup.com/reviews/Palit/GTX_750_Ti_StormX_Dual/6.html
. This means that if you let the AA go and drop down to second-highest or medium, it should be 60+ FPS, easy.
To get back to smoothness
, though; The Tech Report does a poor job of hosting comparable results, so comparing the frame times of the 750 Ti and the 660 is guesswork
. The closest resemblance we're going to get is BF3 (660, stock)
vs BF4 (750 Ti, stock)
, as the Frostbyte engine did not change drastically between the two.
The 660 vanilla spent no time beyond 50 ms (good, 20 FPS steady), was 39 ms in the 33.3+ ms zone (tsk, tsk, but not a horrible 30 FPS experience), and was stuck for 14,700 ms in the 16+ ms zone (yikes, no 60 FPS gaming for you). http://techreport.com/review/23527/review-nvidia-geforce-gtx-660-graphics-card/4
The 750 Ti vanilla spent no time beyond 50 ms (good), no time beyond 33.3 ms (VERY good), and only 5,635 ms in 16+ ms territory (impressively smooth compared to the 660, but obviously not optimal). http://techreport.com/review/26050/nvidia-geforce-gtx-750-ti-maxwell-graphics-processor/8
TTR tests are at 1080p with Ultra/4xMSAA and High/High-AA settings respectively. I don't know if BF4 no longer had Ultra, or if they just arbitrarily ran it at the second-highest setting. At any rate, there is good indication that the 750 Ti can perform very smoothly. Just dig up the 650 Ti figures and you'll see something ugly... and I still game with it successfully (though it is undeniably bad at twitch shooters and large-scale games, unless everything looks like lego blocks, but that's oldschool, so hey...)!The Take
With the 750 Ti alone, you can have a much cooler card (less power = less utility bill, less heat in the case and your workspace, less work for the cooler), newer tricks (the newer card's advantage in Crysis 3 is an indication of that no doubt), and similar - or perhaps even smoother? - real-world performance (especially in newer titles; in older or poorly optimised titles the new tricks do not always help, so it's all down to raw processing power). The first observation alone
should be enough to arouse the interest of anyone browsing SPCR.
I probably made some horrendous mistake compiling the data, but there you go. I like to do this exercise every now and then to show people just how far the mainstream cards have come - they're not the junk of old, and can put up a fight against the previous generation, even punching above their weight class (that 660 vs 570, oh man...
). Of course, the 960 will still beat the 750 Ti in performance hands-down, but not without a heat penalty and a heftier price tag - and how much performance is it that you need?So now we wait for the 950...
PS. While you're at it, peep out the 660 vs 960 frame times at TTR, starting on page http://techreport.com/review/27702/nvidia-geforce-gtx-960-graphics-card-reviewed/5
Edit1: corrected the power calculation, added some emphasis here and there.
Edit2: added supplementary note regarding 960 vs 750 Ti; should not ignore the current generation and original topic.