GTX670 VS GTX570 benchmarks - 20 games tested on 1920X1200

[Artas1984]

Forgot to inform you xtremesystems guys about my personal recent benchmark!

Look what i've got now!


These are the Gigabyte GeForce Windforce GTX670 and GTX570 models!
Respectively the GV-N670OC-4GD and GV-N570SO-13I - in other words the most powerful in their class. And the deal is simple: one card logically replaces another, so i am here to find out is the replacement a valid deal.

My objectives for this review are:

1) To find out how much faster the GTX670 is in terms of FPS difference.
2) To compare the real performance difference between theoretic performance difference.
3) To compare price per performance ratio
4) To find out the CPU bottlenecks if possible.

The benchmark rig is

Core i7 3820 @ 4550 MHz
Asus X79 Deluxe
8 Gb Patriot Viper DDR3 1600 MHz CL8 in quad channel config.
WD Caviar Black sata3 1 Tb game drive
Corsair Force GT sata3 120 Gb Windows drive.

Windows 7 Pro 64 bit.
NVIDIA Forceware 306.97 drivers.





20 various games will be tested on their maximum settings, AA on all games will be 4X and the resolution will be 1920X1200. Each game will have a different testing methodology, but all the methods will be repetitive for you.

All the tested games will have different engines! Noted will be their versions. Some games will have their own in-game build benchmarks, but for the majority fraps testing will be used.

1) TESTING GAME PERFORMANCE

ALAN WAKE {1.02.16.4261}
Engine - MAX FX 3.0
Testing - Fraps 15 seconds, from the start of episode "Signal".
DX10. Brightness 15/20. All settings max. 4X AA.



Aliens vs. Predator {1.0}
Engine - Asura
Testing - Fraps 15 seconds, from the start of Predator campaign.
DX11. Brightness 6/10. All settings max. 4X AA.



James Cameron's Avatar {1.02}
Engine - Dunia
Testing - Fraps 15 seconds, from the start of campaign when you a free to interact.
DX10. All settings max. 4X AA.



Cryostasis: Sleep Of Reason {1.02}
Engine - AtmosFear 4
Testing - Fraps 15 seconds, start of the campaign.
DX10. Physx On. All settings max. 4X AA.



Crysis {1.2}
Engine - Cry 2
Testing - in-game build benchmark.
DX10. All settings max. 4X AA.



Crysis 2 {1.9}
Engine - Cry 3
Testing - Fraps 15 seconds, from where you free to move.
DX11. Brightness 100 %. All settings max. High Resolution Textures OFF. 4X AA.



Deus Ex Human Revolution {1.2.633}
Engine - Crystal
Testing - Fraps 15 seconds, from the start of campaign.
DX11. All settings max. 4X AA.



F.E.A.R. 3 {1.0}
Engine - Despair
Testing - Fraps 15 seconds, from the start of campaign, where you a free to move.
DX11. All settings max. 4X AA.



Formula 1 2011 {1.0}
Engine - EGO 2.0
Testing - Fraps 15 seconds, select time trial, red bull, Albert park, wet weather, and as soon as you see the car - start benchmarking, do not drive the car...
DX11. All settings max. 4X AA.



Tom Clancy's H.A.W.X 2 {1.01}
Engine - ?
Testing - Fraps 90 seconds on top of in-game build benchmark from the start.
DX11. All settings max. 4X AA.



Hard Reset Extended Edition {1.5}
Engine - Road Hog
Testing - in-game build benchmark.
DX9. FOV 90. Gamma 1.5. All settings max. 4X AA.



Homefront {1.0.384501}
Engine - Unreal 3
Testing - Fraps 15 seconds, start of campaign.
DX11. FOV 90. Gamma 2.5. All settings max. 4X AA.



Lost Planet 2 {1.1}
Engine - MT Framework
Testing - Fraps 90 seconds, on top of in-game build benchmark.
DX11. All settings max. 4X AA.



Metro 2033 {1.02}
Engine - 4A
Testing - Fraps 15 seconds, start of campaign, from where you are free to move.
DX11. All settings max. 4X AA.



Necrovision {1.2}
Engine - Painkiller
Testing - Fraps 15 seconds, start of prologue campaign.
DX10. Brightness 1. Contrast 1. Gamma 1.5. FOV 110. All settings max. 4X AA.



Red Faction Armageddon {1.01}
Engine - Geo-Mod 2.5
Testing - Fraps 15 seconds, start of campaign.
DX11. All settings max. 4X AA.



S.T.A.L.K.E.R.: Clear Sky {1.5.1}
Engine - X-ray 1.5
Testing - Fraps 15 seconds, start of engine cinematic sequence.
DX 10.1. All settings max. 4X AA.



Serious Sam 3 {1.0}
Engine - Serious 3.5
Testing - Fraps 15 seconds, start of the third autosave.
DX11. All settings max, except video ram set to medium. 4X AA.



Syndicate {1.0}
Engine - Starbreeze
Testing - Fraps 15 seconds, start of the fifth autosave.
DX11. All settings max. 4X AA.



World In Conflict {1.9}
Engine - MassTech
Testing - in-game build benchmark.
DX10. Brightness 6/9. All settings max. 4X AA.





2) Difference between real and theoretic performance

As we know theoretical performance is typically measured in texture fillrate, pixel fillrate and memory bandwidth, so let's take a look what these cards offer.


GTX670 has 20 % more memory bandwidth and 7 % more pixel fillrate, but texture fillrate of GTX 670 is twice higher! This reminds me the transition from 6800 Ultra to 7800GTX where memory bandwidth and pixel fillrate increased by a significantly smaller margin than texture fillrate - what does that mean for my 20 games?

CONCLUSIONS

Minimal frames: GTX670 is ~ 1,3 times faster than GTX570
Average frames: GTX670 is ~ 1,3 times faster than GTX570
Maximum frames: GTX670 is ~ 1,3 times faster than GTX570

The small difference between minimum FPS means than GTX570 did not suffer due it's smaller RAM or memory bandwidth.

The real world performance is significantly smaller than theoretic.

3) Price/performance ratio

GTX570 last price on newegg was 329 $. GTX670 is now at 449 $.

CONCLUSIONS

The difference between prices is 1,36, which is roughly equal to real world performance difference. The performance difference on 2560X1440 would be increased in GTX670 favor even more due to 40 GB/s more memory bandwidth, so the price is valid.

4) CPU bottleneck?

In a CPU bottleneck situation you are most afraid of the minimal frame rate drop (i will explain and proove that statement in my other CPU dedicated benchmark) and don't care that much about maximum frame rate.
How much FPS would drop if i lowered the Core i7 3820 to 4 GHz?
Without bothering you with new charts, i will just state the difference in minimal frame rate going from 4000 MHz to 4550 MHz in %.

Alan Wake - 0 %
Aliens vs. Predator - 0 %
Avatar - 2 %
Cryostasis - 0 %
Crysis 2 - 0 %
Deus Ex Human Revolution - 0 %
Formula 1 - 3,5 %
F.E.A.R. - 0 %
Hard Reset - 7 %
H.A.W.X - 21 %
Homefront - 0 %
Lost Planet - 2 %
Necrovision - 0 %
Red Faction Armageddon - 1,5 %
Serious Sam 3 - 1,5 %
S.T.A.L.K.E.R.: Clear Sky - 0 %
Syndicate - 0 %
World In Conflict - 8,5 %

CONCLUSION: there is almost no bottneck for GTX670 when CPU speed is 550 MHz lower. Only Hard Reset, World In Conflict and H.A.W.X would benefit from CPU OC.

[Artas1984]

Made a video presentation out of this benchmark if anybody is still interested in upgrading their cards:

https://www.youtube.com/watch?v=_-s6djWQz00