Exploring bandwidth on the 660 Ti -
http://www.tomshardware.com/reviews/geforce-gtx-660-ti-memory-bandwidth-anti-aliasing,3283.html
http://www.tomshardware.com/reviews/geforce-gtx-660-ti-memory-bandwidth-anti-aliasing,3283.html
What leaves me wondering, is the fact that they offer no explanation for GTX670's almost as bad decline in performance despite having more memory bandwidth than 7870 for example.
I think it was a copy from hardopc, i edit the post, excuse for the time but i read your post just now
Because an early leak like that is at risk of being withdrawn quickly, which it since has, so it makes sense to mirror the images.
Weird those nvidia results really were inconsistent (gtx560 vs. gtx660 vs. 7850) - the reality isn't really that bad. Not quite up to 7870 level but very nicely between 7850 and 7870, with low power consumption (at idle and load, the latter despite the rather high clock) too.
Now where's the GTX650
FWIW there's also a review of the GTX650 (the only one I found): http://www.tomshardware.com/reviews/geforce-gtx-660-geforce-gtx-650-benchmark,3297.html
Looks like the gtx650 is a decent competitor to hd7750 (some weird settings though and no power draw measurements). Needless to say a 109$ gtx650 is a way better deal than a 99$ gt640 (offereing nearly 50% more performance), though ideally the price shouldn't be higher than that of a hd7750.
Looks like the gtx650 is a decent competitor to hd7750 (some weird settings though and no power draw measurements). Needless to say a 109$ gtx650 is a way better deal than a 99$ gt640 (offereing nearly 50% more performance), though ideally the price shouldn't be higher than that of a hd7750.
Hardocp also puts the 660 on nearly even footing with the 7870, and Tomshardware shows the 660 closer to the middle, but still closer to the 7870 than the 7850.
So which reviews are you talking about?
Weird those nvidia results really were inconsistent (gtx560 vs. gtx660 vs. 7850) - the reality isn't really that bad. Not quite up to 7870 level but very nicely between 7850 and 7870, with low power consumption (at idle and load, the latter despite the rather high clock) too.
Now where's the GTX650
Hardware canuck find different results for TDP, 7870 > 7850 and the 660 use more of the 7870 .. then again i can imagine it differs from the situation used.
http://www.hardware.fr/articles/876-1/nvidia-geforce-gtx-660-asus-directcu-ii-top-sli-test.html
Like allways, it completely depend of the games, zone tested and settings used
(extreme quality is in general by use 8xMSAA when possible )
Last edited by a moderator:
Yeah that's what I meant. Just about everywhere else puts the card in-between the 7850 and 7870.
Yes and 8xMSAA in particular seems to hurt all Kepler cards quite a bit more than GCN cards, compared to 4xMSAA.
What is interesting to me is that while GK106, proportionally, is cut down from GK104 in almost the same exact way that Pitcairn is, and their performance for the most part aligns that way, the die size reductions did not scale together. Pitcairn definitely inherited a slightly better die size reduction AND, probably, power use reduction as well.
Really though, compared to 40nm, Nvidia has managed to more effectively reign their die sizes AND power consumption, making bigger improvements in both raw TDP and perf/mm^2 AND in percentage of change over AMD. GK106 is 35-40% smaller than GF114 (depending on the source you use for GF114's die size), yet it's 30-35% faster and consumes 20-40% less power (depending on the program/game). Compared to GF114, the chip that it replaced from a development point of view, it's smaller, uses less power, and is essentially twice as fast. It's quite a huge turn around, and hopefully a sign of things to come with Maxwell. It's incredible progress from where Nvidia was at on 40nm.
Really though, compared to 40nm, Nvidia has managed to more effectively reign their die sizes AND power consumption, making bigger improvements in both raw TDP and perf/mm^2 AND in percentage of change over AMD. GK106 is 35-40% smaller than GF114 (depending on the source you use for GF114's die size), yet it's 30-35% faster and consumes 20-40% less power (depending on the program/game). Compared to GF114, the chip that it replaced from a development point of view, it's smaller, uses less power, and is essentially twice as fast. It's quite a huge turn around, and hopefully a sign of things to come with Maxwell. It's incredible progress from where Nvidia was at on 40nm.
But I agree, nV made quite some progress this iteration. But it doesn't look that bad on AMD's side either. Pitcairn and CapeVerde are also quite potent for their size (slightly in front of GK107 and GK106 in performance at the cost of a roughly proportional higher power consumption). Only Tahiti appears to carry quite some additional burden (significantly higher DP performance, ECC and that stuff).
I've yet to see any conclusive power consumption tests this generation. Dynamic clocking and significant per-game variability in GPU resource use mean that just using a single application to determine power usage is no longer adequate.
Most sites have said that the HD 7870 draws more power than the GTX 660 under load (using a single game as the criteria) while at least one has said the opposite.
Its clear it will be interessant to see reviews do a % ratio based on all the games they are testing performance wise and not just give the "higher reported"; or by test one games as typical gaming tdp..
The fluctuation between games is unpredictable before you test it. Use Crysis2 and run it for 3min and say, this is the average tdp you will get over all games is completely wrong today.
( just have to look the Hardware.fr reviews who use 2 games with completely different results )
Looking at the Toms hardware review, its clear a 650TI will be released soon for goes against the 7770.. Surely with 4SMX enabled on the GK106
Last edited by a moderator:
