[Artas1984]

Over the last 2 months HD6950 has been tested in various games, but mostly in new titles. Today i am testing it versus HD6850 simply to find out how much faster is it, and not "will it run that game". So i will be using 20 games, older and newer - but all can tell you the difference.

Today i will be reviewing these 2 cards:

Gigabyte Radeon HD6850 OC with 1 Gb of GDDR5:

http://www.gigabyte.com/products/pro...px?pid=3630#ov

XFX Radeon HD6950 with 2 Gb of GDDR:

http://xfxforce.com/en-us/products/g...0Standard.aspx

I am at my friends place, and brought the HD6950 for testing from my PC, and this is my friends test PC:




As you can see, HD6950 is a very large card, and actually if fitted inside the case, Cooler Master CM 690 II Advanced, it would block 4 sata ports, coming from the motherboard. HD6850 is just slightly behind the ports by a few mm, so it is perfect for the machine. And here is the full rig:

Intel Core 2 Duo E8400 clocked at 3600 MHz
Gigabyte EP-45-EXTREME
Gigabyte Radeon HD6850 OC
2X2 Gb Kingston Hyper-X clocked at 1066 MHz DDR2
Enermax MODU82+ 625 W EPS12V

We are using Catalyst 11.2 drivers.
Now, to save time and show the real potential of the cards, we decided to do benchmarks only on 1600x1200 resolution, and if possible - with max AA, with all settings and effects set to max in every game. If a game has an option to select DX9 or DX10, it means DX10 is used. Some games have their own benchmarks, and those who don't - we use fraps. To make the tests as precise as one can do with fraps, for both video cards we used the same custom scene, running 15 seconds. During this time fraps measures the minimum, average and maximum frames, those will be dsiplayed in every screenshot. All the screenshots of the games show the actual moments of the benchmarks.

Lets' start.

Assassin's Creed 2 is perfect game to show the huge gap between the 2 cards. HD6950 is 2 times faster here. We are using a custom 15 seconds fraps benchmark.



Avatar - The Game is not so demanding and runs smooth on both cards. HD6950 is about 15 % faster here. We are using a custom 15 seconds fraps benchmark.



Bioshock was chosen for testing instead of Bioshock 2, as the later game is set at 60 FPS cap, and both cards never scored below 60 FPS. Anyway HD6950 is 25 % faster here. We are using a custom 15 seconds fraps benchmark.



Battlestations Pacific is another system hog and HD6950 is 2 times faster here. We are using a custom 15 seconds fraps benchmark.



Cryostasis is one of the most demanding games ever, and in fact, while i had a GTX280, the game lagged more than Crysis. Shadows and ice particles are the main cause for that, as i tested this subject separately. This game is an NVIDIA Physx game, and as such, it kills the HD6850; however, HD6950 manages simply to ignore the fact, that this is a Physx game and it crushes the HD6850 by 4 times here!!! There is nothing wrong with HD6850, as my friend had a HD4870 before, it scored only 7 - 10 FPS on the same rig. I expect that HD6950 has something that makes Physx a no problem. We are using a custom 15 seconds fraps benchmark.



World In Conflict is very demanding strategy, but it is rather a cpu hog, and as such does not show that much difference between the cards, except for max frame rates. The game has it's own benchmark.



Dead Space results are rather disturbing for HD6850 as it gets wiped by HD6950 4 times here, although the game itself generally does not lag even on HD3870. We are using a custom 15 seconds fraps benchmark.



Dead Space 2 is great for both cards, however it is very important to notice, than when using older ATI drivers, HD6850 was lagging like hell with just 20 FPS on average. Now, with 11.2 it is fixed. We are using a custom 15 seconds fraps benchmark. HD6950 is 25 % faster here.



GTA 4 is one sluggish game, but with the new 11.2 drivers, HD6850 frames went twice as high as with previous drivers and now it is not far behind the HD6950. GTA 4 has it's own benchmark.



The old Far Cry still perfectly fits for testing video cards. HD6950 is 2 times faster here. We are using a custom 15 seconds fraps benchmark.



Far Cry 2 is the only game, where HD6850 was slightly faster than HD6950. I have no clue how is that... We are using a custom 15 seconds fraps benchmark.



FEAR was the most popular benchmarking game along with some others before Crysis appeared. Even to this day this game is perfect for testing video cards. As you see, HD6950 is from 2 to 3 times faster here. FEAR has it's own benchmark.



FEAR 2 seems to have the same results for both cards as HD6950 is about 250 % faster here. We are using a custom 15 seconds fraps benchmark.



Jericho is another NVIDIA Physx game, and once again HD6850 is trashed to pieces, while HD6950 delivers excellent performance and outguns the HD6850 by some 300 %. Guess there is no need for NVIDIA card in your other PCI-E slot when you have HD6950. We are using a custom 15 seconds fraps benchmark.



Necrovision favors ATI cards in general, despite the NVIDIA logo, and from my experience, CrossFire scales over 90 % on high resolutions, while here HD6950 is 20 % faster. We are using a custom 15 seconds fraps benchmark, even though the game has it's own - but it is bugged as hell.



Resident Evil 5 is great for both cards and the difference is small. The game has it's own benchmark.



Company Of Heroes is said to be working fine in a Pentium 4 with Radeon 9600, but if you crank up the game and use specific scenes, it can lag on GTX280 from my previous experience. Here the HD6950 is 10 % faster. We are using a custom 15 seconds fraps benchmark.



Serious Sam 2 is a big surprise - max out it makes the HD6850 look bad as the HD6950 destorys the HD6850 by some 250 %. And yes, this game lags on powerfull cards with 8X AA and all settings max. We are using a custom 15 seconds fraps benchmark.



Bulletstorm is the newest game around, definitely one of the BIG games of this year and it needs powerfull video cards to work. HD6950 is from 33 % on average frames to 50 % on minimum frames faster here. The game is set at 62 FPS cap, so we could not see the max frame rates of HD6950, but the rest results show the scene very good. We are using a custom 15 seconds fraps benchmark.



Crysis - need i say more? Perhaps only that HD6950 is from 33 % to 50 % faster here, and you won't be able to play this game with HD6850 on max settings smooth. As all know, Crysis has it's own benchmark tool.




For final words the HD6950 is one impressive card, that just does not lag, though for the most games HD6850 is fine. Take a notice that new ATI drivers make a big impact and as time will go by, we might see a dramatic FPS improvement for HD6850 as was here in some cases.

The rest comments i will leave for you.

[hajalie24]

Great Thread, but I don't understand how the 6950 ignored the physx slowdowns unless that was turned off.

[||Console||]

Nice work , You should Contact some one on the news team and put this up on the Front page.
Good to see a review with c2d's because there are still alot of people (like me) running them .

[Artas1984]

Originally Posted by ||Console||

Nice work , You should Contact some one on the news team and put this up on the Front page.
Good to see a review with c2d's because there are still alot of people (like me) running them .

If you guys think that it is worth being in the news, it would be easyer for you to make that happen, as you are mods here.

And yes, i recommened my friend to keep the E8400 as i saw that in benchmarks it can beat Core i7 920 in many games. Besides he can OC it to 4 GHz stable.

Originally Posted by hajalie24

Great Thread, but I don't understand how the 6950 ignored the physx slowdowns unless that was turned off.

I do not understand that too! But one thing you shuold know: Physx software had to be installed, or those games would not even start! Perhaps HD6950 has some new tech, that makes no difference if Physx is processed or not.

[fntastiK]

wow its a really interestin thread! The HD6950 is better but the price is there too. I would to see the HD6850 vs the HD6870

[Janus67]

Interesting, I'm assuming that the 6950 is running completely stock w/o a flashed bios?

I would be interested to see how different the #s would have been had you been using an i7 (series) processor.

[Artas1984]

Anybody can explain the NVIDIA Physx thing that went smooth with HD6950 and bad with HD6850?

I will remind that 9.10 Physx software was used.

Originally Posted by Janus67

Interesting, I'm assuming that the 6950 is running completely stock w/o a flashed bios?

I would be interested to see how different the #s would have been had you been using an i7 (series) processor.

Default XFX HD6950 - not even overclocked. And about that Core i7 - can not do that, but i can do a Phenom 2 955 test instead.

[Janus67]

I'd be interested to see that test anyway, just to get a different look at things (see how CPU-bound some of the tests were that maybe the 6950 was able to take a bit of the load off that the 6850 couldn't).

Interesting stuff regarding the Physx framerate differences, I haven't really experienced it (then again I've only played a handful of the games you benchmarked)

[Artas1984]

Originally Posted by Janus67

I'd be interested to see that test anyway

It's on the way, this time in my own PC.
Will be comparing HD6850 to the GTX280 and CrossFire HD4830. Also will probably include GTX470 and HD6950 in the mix.
I already have the results, just need formating now.

[Janus67]

Sounds great, can't wait to see them!

With all that hardware, have you considered trying your hand at joining the benchmarking team?

[Artas1984]

Originally Posted by Janus67

Sounds great, can't wait to see them!

With all that hardware, have you considered trying your hand at joining the benchmarking team?

Won't have time to mess around with all of the benchmarks, and i am not very rich on hardware anyway. But here is my new mainstream benchmark of HD6850 VS GTX280 and other cards, like said, on Phenom 2 this time:

http://www.overclockers.com/forums/s...d.php?t=674776

[raIDERgeek]

Damn know you have me thinking to stick with 2CD and just a get a video card.

[deathman20]

Don't know how I missed this before but good job. That is some impressive results across the board. Curious how it stacked up to your old 280 though, i know I wish I did some more tests with my 285 before I sold her.

Have you tried any of the benchmarks again with the 11.4 drivers? As well with the 6950 unlocked for more performance?

[Artas1984]

Originally Posted by deathman20

Don't know how I missed this before but good job. That is some impressive results across the board. Curious how it stacked up to your old 280 though, i know I wish I did some more tests with my 285 before I sold her.

Have you tried any of the benchmarks again with the 11.4 drivers? As well with the 6950 unlocked for more performance?

No, this benchmark is over, since HD6950 is no more. I also have sold the GTX280, although you can see it's results vs HD6850 in the other thread.
The only future thing that might happen is when my friend will upgrade his monitor, and do some more benchmarks. A friend of mine has two HD6970 cards, so heck knows against what we could benchmark those...

[bigben6]

Good stuff, but for a (typically) less than 50% increase, I could Xfire 2 6850's I wonder how they would compare, what IS amazing is the 2-3 instances where the 69xx was 200+% faster... crazy!!!!

6850 vs 6950: http://www.anandtech.com/bench/Product/291?vs=331
Seems ot support your findings

6850 CF vs 6950: http://www.anandtech.com/bench/Product/302?vs=331

Xfire seems to win


EDIT: Anandtech does NOT have a 2Gb version of the 6950 in their data base... so take those with a grain of salt...

[Artas1984]

The big difference between the cards might have been due bandwidth limitations, or frame buffer, where the 2 Gb of RAM on HD6950 might have proved to be useful in place. And yes - in the near future in that same rig a CF HD6850 setup will be tested.

[EarthDog]

How do you figure the 2GB of ram helped at a mere 1600x1200 resolution?

[Artas1984]

Originally Posted by EarthDog

How do you figure the 2GB of ram helped at a mere 1600x1200 resolution?

8X anti-aliasing was used too.

If you look at the other thread, where the cards were tested on 1680X1050 and 4X anti-aliasing, you will see little difference between HD6850 and HD6950. This said, the conclusion comes that it must be the memory bandwidth that comes to play in some of the games here. What else? Being the smart *** that you are dog, explain yourself why.

[EarthDog]

Im not being a smart ass. Im just asking to explain your conclusion a little better.

Most other reviews show that 2GB of ram doesnt matter on MOST games until 1920x1080 and with gobs of AA... Im just wondering why your statement is contradictory. Let me find a link for you.

EDIT: Why dont you load up MSI Afterburner or GPUz and see exactly how much memory its actually using across all your tests.

EDIT2: Here is H2H 6950 1GB v 2GB.

1GB: http://www.techpowerup.com/reviews/A...50_1_GB/6.html
2GB: http://www.techpowerup.com/reviews/H...HD_6950/8.html

Cliff's notes: You can see that outside of what appears to be some margin of error results (1-2FPS difference) and maybe some specific games where texture loads are heavy (not old games surely), it really doesnt matter until 2560x1600 res. I am reading that correct, right? Again, not trying to be a smart ass at all. I just thought from seeing many reviews it doesnt matter much until 1920x1080+ with AA. The lower the res the less textures are loaded. Obviously with a lower res, more AA needs to be used and of course the more memory is used, but, after looking at those reviews, Im going to have to stand by my original statement...

[bigben6]

Originally Posted by Artas1984

What else? Being the smart *** that you are dog, explain yourself why.

I hope this was intended as playful, dog is the biggest meanest stinkiest smart ass you will ever meet, hes pretty smart though, if he says something he can prolly back it up with data and if he asks a question is prolly a question worth asking...