The purpose of this benchmark was to show the difference between the cards, when heavy memory bandwidth is in action, having 8X anti aliasing and 1600x1200 resolution in all of the games. This clearly proved that HD6850 is a mainstream card and that HD6950 is for enthusiasts.
This time, in todays benchmark, i focus on the mainstream settings and the whole purpose is to find out how does the old GTX280 hold up against the new HD6850. Many HD6850 VS GTX460 benchmarks have been made around, but all of them threw in your regular newest and most popular games, and i was not happy about that at all, as i don't play them. Also few of those benchmarks had older cards to compare like GeForce 200 or Radeon HD4800 series. I also tossed in HD6950 and GTX470, to give a preliminary view of how these cards work on mainstram settings, not fully exposed to their potential.
AMD Phenom 2 X4 955 BE RB-C2 1.4 V, overclocked at 3600 MHz
Gigabyte MA790XT-UD4P F7
Corsair 2X2 Gb DDR3 XMS3 DHS, downclocked at 1333 MHz, CL8
Inter-Tech Coba Nitrox IT-7600SG 600 W EPS12V
Catalyst 11.3 & ForceWare 266.58 used.
For games, those do not have their own benchmark tool, i used custom fraps 15 s setting of the same episode.
All of the tested games are set at 1680x1050 resolution, with all settings set to max and all effects enabled, AA set to 4X.
We will start with first person shooters.
Nothing unusual - GTX280 outperforms the HD6850 by some 10 %. CrossFire HD4830 scaling is excellent, doubling the performance and slightly beating HD6850. HD6950 is about 15 % faster than GTX470. Bioshock was used instead of Bioshock 2, since the later is reduced to 60 FPS. Fraps used for benchmark.
SERIOUS SAM 2 DX9
No surprise to see NVIDIA wining, with GTX280 beating the HD6850 by some 20 %. CrossFire HD4830 scaling is about 33 %. Notice that the high end cards are bottlenecked by resolution. Fraps used for benchmark.
In Necrovision HD6850 beats the GTX280 by some 20 % and HD6950 also slightly beats GTX470. CrossFire HD4830 scaling is overwhelming, delivering double the performance and beating HD6850 slightly. Fraps used for benchmark (the in-game benchmark exists, but it is buggy).
CRYSIS WARHEAD DX10
Victory for ATI! HD6850 moderately beats the GTX280 by some 30 %, and HD6950 beats the GTX470 by some 20 %. Unfortunately for HD4830, it get's trashed badly and CrossFire crashes for some reason... Fraps used for benchmark (since original game was not installed).
FAR CRY 2 DX10
Another Crytek game, where HD6850 beats the GTX280 by some 15 - 20%, but GTX470 manages to slightly outperform the HD6950. CrossFire HD4830 scaling is very good - delivering double the frames on average. Used the in-game benchmark: 3 sequences of small ranch with disabled AI.
Bulletstorm is fresh and has issues.. One of them is broken support for any resolution, except for the one dividable by 8. I, however only encountered this bug with NVIDIA cards. So in this benchmark you can not compare NVIDIA to ATI, but you sure can compare how do the cards of the same brand stack with each other. To add on problems, CrossFire HD4830 scaling is negative. This game is the perfect example why older, matured games should be benchmarked instead of new freshies.... Fraps used for benchmark.
FEAR 2 DX9
FEAR always liked ATI cards over NVIDIA, and it comes to no surprise here, in FEAR 2, that HD6850 beats the GTX280 by some 15 - 20 %, and HD6950 beats the GTX470 same way. CrossFire HD4830 scaling is once again, rewarding - delivering double the performance and slightly beating the HD6850. Smooth frames in general. Fraps used for benchmark.
In this adventure first/third person shooter, HD6850 beats the GTX280 by some 15 % and HD6950 takes the lead over GTX470 slightly. CrossFire HD4830 scaling is just 30 %, although the game is playable with a single HD4830. Fraps used for benchmark.
DEAD SPACE 2 DX9
Dead Space 2 - another mix of first/third person shooters and works like a charm - delivering such nice frames for a new game is something worth an applause. GTX280 ties with HD6850 and GTX470 ties with the HD6950. CrossFire HD4830 scaling is slightly over 50 %, which is, of course, a good deal of improvement. Fraps used for benchmark.
RESIDENT EVIL 5 DX10
An iconic game that ironically became a shooter out of survival horror. Funny to see all the cards basically performing the same. Resident Evil 5 is efficiently optimized, but it does bottleneck the high-end cards on these settings. Even a single HD4830 works great. Used the fixed value in-game benchmark.
This game is know to be one of the most laggy and poorly coded games ever - it's patches were made till 2009 and even though i have the latest version, i have to say that the results are always different on different resolutions.
In this case, NVIDIA is just dominating the ATI, but this is only true for this specific resolution. For example if i change the resolution to really low, i actually can get even more lag than on high resolution. But generally speaking, GTX280 is better than HD6850 is this game, no mater what the resolution is, unless it is broken. Fraps used for benchmark.
COMPANY OF HEROES DX9
What is the most important here is that GTX280 greatly beats the HD6850 in minimal frame rates by some 40 % and GTX470 just crushes the HD6950 in minimal frame rates by some 70 %. Amazingly, CrossFire HD4830 scaling is so good, that in minimal frame rates, it surpasses both the HD6850 and HD6950. While on average frames the HD6950 is still the fastest card, it is sad to see it delivering really low minimal frame rates. Used the in-game benchmark.
WORLD IN CONFLICT DX10
Cards are trading blows here, although in minimal frame rates GTX280 beats the HD6850 by some 15 % overall slightly too. GTX470 beats the HD6950 by some 17 % in minimal frames, but looses in average and maximum scoring. Excellent CrossFire HD4830 scaling at around 66 % enables it to tie the frames with HD6850. Used the in-game benchmark.
BATTLESTATIONS PACIFIC DX9
A modern naval/aircraft simulator shows pure NVIDIA dominance - GTX280 just destorys the HD6850 by the access of 75 - 90 % and so does GTX470 destory the HD6950. Some games are just made for NVIDIA cards and there is nothing you can do about it. To add salt on flesh, CrossFire HD4830 scaling is negative. Fraps used for benchmark.
Another modern aircraft simulator HAWX has it's own benchmark, where average and maximum frame rates are calculated. HD6850 ties with GTX280 and HD6950 ties with GTX470. Maximum frame rates are not important as they are limited to 322 FPS an all cards hit that wall. Used the in-game benchmark.
The settings used here perfectly work to show the difference between the mainstream cards, but do not show the full potential of high end cards, so this benchmark should not be used as an example for HD6950 and GTX470 - only in some games was the full potential of those cards shown.
Now for the ultimate conclusion:
GTX280 and HD6850 are roughly equal in performance, and CrossFire HD4830 setup, when scaling is 100 %, is equal to them. The results vary too much to establish one's card constant dominance. There will be always cases where one card will majority outperform another or dual card scaling will not work - that should not impact the final verdict.