AMD 6970 VS Dual 6850 at High Resolutions (numbers inside)

K

kevinnem

Weaksauce
Joined
Sep 2, 2010
Messages
94
Hello everyone,

The following data has been re created form the tom's hardware artical about the new 6000 card ati / amd has released.
Click below for my graph (someone please confirm this works)
http://www.screencast.com/users/kevinnem/folders/Jing/media/09f589a5-5a8f-4b4b-95b0-5e3294daff82

2010-12-15_1459.png


Link to artical: http://www.tomshardware.com/reviews/radeon-hd-6970-radeon-hd-6950-cayman,2818.html


The point: I am goign to be doing a 3 monitor eye finity set up, and as they are the same cost, I Am considering a dual 6850 set up , or a single 6970.

I assumed that with 2 gig of ram, then 6970 would win, but at the highest resolutions tested, it was clear that eh dual set up works better (in all but 1 situation). I may have missed soem aspect of this anaylsis - please point out any errors.



data:
#dmark Vantage
cost 2 3 4 5 6 7 metro lost planet AVP AVP (min) BF BC2 BF BC2(min) dirt 2 dirst 2(min Just cause
single 6970 360 2158.9 12.65 75 42.4 66.2 270.5 55 37 35.8 19 46.6 35 52.5 43.1 35.4
dual 6850 360 2059.8 12.85 65 31 59 175.7 58 49.7 43 14.7 62.5 42 69.9 57 47.1
benefit of dual% 0 -4.590300616 1.581027668 -13.33333333 -26.88679245 -10.87613293 -35.04621072 5.454545455 34.32432432 20.11173184 -22.63157895 34.12017167 20 33.14285714 32.25058005 33.05084746
 
Interesting, was looking for this kind of info myself after being sorely disappointed with the performance of DX11 games in Eyefinity/Crossfire
 
Hmm, I'm kind of regretting my 6870 purchase now. Looks like dual 6850s get better minimum and average fps. The only concern I would have is would the resell value of two 6850s be worse than a single 6970 a year or two down the road...
 
What resolution did you compile your graph from? I believe Tom's did a max resolution of 2560 x 1600.

Let's put the number of pixels into perspective.
1920 x 1200 = 2.3 MP
2560 x 1600 = 4.1 MP
3(1680 x 1050) = 5.3 MP
3(1920 x 1080) = 6.2 MP

Where 1GB of VRAM is usually fine for 25 x 16 if you do not use AA, Eyefinity at 1080P is still 2 MP more than that.

My point is: I think we have to see Eyefinity resolutions get tested with the 6800 series before reaching a conclusion on whether or not 1GB is enough. I predict that it would not be.

EDIT: When comparing 6850 CF and a 6970 at Eyefinity resolutions I don't think the RAM difference will matter. The 6970 will run out of raw performance while having RAM to spare, and the 6850 CF will have raw performance to spare while running out of RAM. They would both probably have similar image quality at the end. I would choose the 6970 because I would rather take a fastest single GPU available, and not deal with MGPU when I do not have to.

EDIT2: dook43's post affirms what I'm saying. AA increases the amount of memory used, and so does increasing the number of pixels, so even w/o AA at some point you're going to reach a point where 1GB is not enough. I'd estimate 1080P Eyefinity w/o AA is approaching that point.
 
Last edited:
Nobody has tested 6870 crossfire eyefinity that I have seen on the web. I can tell you from experience that AA is impossible to apply with a single 2GB 5970 or 5850 crossfire at >= 5040x1050, though.
 
thanks for the input guys, I am really interested to see how this develops.

I can see your concern with the high resolutions ,and the vram available.

I am looking for a 3 monitor 3X24 HD set up, adn these where the highest resolution numbers I could find.

I am sure I am nto the only one that is surprised that at least at this setting, (25x16) the results where in favour of the dual 6850 setup.

I also did teh cals RE: pixels, my consideration what that you would use morphological AA that would be much easier on teh system then antospoic.

anyway, please keep comments coming, and if anyone has some some study taht shows how these higher resolutions effect prformance it would be very usefull.
 
I use a single 5850 for 3 monitors. Works well but i wouldn't get 1gig cards. Do at least dual 6950s. They cost more but will do a better job. I


I can do mlaa in most games with the 5850 but i do get stuttering in some things like lotro. Getting a second 5850 (borrowed from a friend tos e performance ) didn't help with it so i'm assuming its a vram issue .
 
ManyCowsMoo said:
Just cancelled my 6970 and ordered a pair of 6850s!
Click to expand...

quick question.

What happens when a game comes out that the 6850s don't run well ?

With the 6970 when you need better performance you can allways buy a second one. Its a pain in the ars to sell 2 video cards.
 
Here's some more info.

Anand did a review of the 5870 2GB and has comparisons to the 5870 1GB. They test both 25 x 16 and 57 x 10, and at 57 x 10 there is no difference to speak of between 1GB and 2GB. I'm genuinely surprised. The three monitors at 57 x 10 are only about 0.7 MP away from the 57 x 12, so these results are more accurate.

But one thing that complicates this. 6870 CF is quite a bit more raw performance, almost twice as much as a 5870. If you have so much raw performance to spare that for example you want to turn on MLAA or CFAA or 8x AA or EQAA, the VRAM may hold you back if the extra memory burden is enough to pass 1GB. Add to that the fact that your setup is 57 x 12, and IDK.

I'll be finding out more about this, it's interesting.

EDIT: Rage3D also has a comparison of 1GB vs. 2GB Grid is one of the few games to give an advantage to the 2GB 5870 at 57 x 10.

EDIT: This will be right up your alley. Tweaktown comparing 5870 2GB vs. 1GB and 5970 at 5760 x 1200.

The conclusion I have reached is that at 5760 x 1080/1200 and with 4x AA, older games and more recent games will not benefit from 2GB.
 
Last edited:
XacTactX said:
Here's some more info.

Anand did a review of the 5870 2GB and has comparisons to the 5870 1GB. They test both 25 x 16 and 57 x 10, and at 57 x 10 there is no difference to speak of between 1GB and 2GB. I'm genuinely surprised. The three monitors at 57 x 10 are only about 0.7 MP away from the 57 x 12, so these results are more accurate.

But one thing that complicates this. 6870 CF is quite a bit more raw performance, almost twice as much as a 5870. If you have so much raw performance to spare that for example you want to turn on MLAA or CFAA or 8x AA or EQAA, the VRAM may hold you back if the extra memory burden is enough to pass 1GB. Add to that the fact that your setup is 57 x 12, and IDK.

I'll be finding out more about this, it's interesting.
Click to expand...

Well, Anand's review also included crossfire numbers (next page from your link), and those are almost 80-90% CF scaling - something the HD5870 didn't see in general. Probably the HD5870, alone, was simply not powerful enough to make 2GB matter. But once in a group, it was capable of making a real difference. It also shows poor CF scaling on the HD5800 series probably was more influenced by a disbalance somewhere in the HW config (that the HD6800 solved, the HD4800/HD5700 were perfect) than drivers. Though that is not going to take any fault away from the drivers, IMO.
 
VERY interesting studies you guys have found there.

So is the conclusion we have come to that 2 gig of ram doesn't help performance over 1 gig?

Does this begs the question : when is 2 GB of Vram usefull? (and can you prove it)?

This is much like an artical I read a while back about higher ram counts, and it concluded that after 4 GIG ram(system), there was a some improvement to 6 gig, and from 6 to 12 gig, made little difference (on typical home use, not industrial).


I wish there was a tool, like resource monitor, that told you how much ram was being used on your video card, so that we could manipulate the AA, and resolutions to see the effects.
 
kevinnem said:
VERY interesting studies you guys have found there.

So is the conclusion we have come to that 2 gig of ram doesn't help performance over 1 gig?

Does this begs the question : when is 2 GB of Vram usefull? (and can you prove it)?

This is much like an artical I read a while back about higher ram counts, and it concluded that after 4 GIG ram(system), there was a some improvement to 6 gig, and from 6 to 12 gig, made little difference (on typical home use, not industrial).


I wish there was a tool, like resource monitor, that told you how much ram was being used on your video card, so that we could manipulate the AA, and resolutions to see the effects.
Click to expand...

All they have found is that cypress doesn't benfit from 2GB frame buffer in the games tested. Hardocp has some games in which the 2GB verison performs better.

At the end of the day it comes down to having to test cayman and see if there are situations where it pulls ahead. We know a 6950 1gig is coming
 
When you go Eyefinity/Surround, then the 2GB helps out. That's been discussed, there were reviews on it on overlock.net, etc.
 
Bez- My gut feeling was that the 2 gig was better then 1 gig, when you have high resolutions , and some AA involved. But I am having trouble finding the data to support this.

XacTactX's post - and the link there in, seem to paint a pretty clear picture that in the mass majority of cases, the raw processing horsepower is more important then ram size.

If you know of studies that show differently please link them.


when we get 1 gig version of the 6900's, (or 2 gig versions of 6900's) I really hope people do a review focusing on that.
 
OK, here's the review I was referring to: here

I think it is pretty clear that if you use AA and eyefinity/surround, you need more than 1GB
 
Last edited:
I'd say having nearly 50% of the games not perform acceptably in Surround on 1GB cards is pretty telling. Nobody has yet done the analysis to that level.

I'd say all of you needing to make a purchase should go out and get a pair of $279 XFX 6950's today.
 
I can speak from experience. My one 6870 runs 3x 1920by1080 monitors with low amounts of AA very nicely. I recorded many numbers but I did try one game.

Dragonage origins - Max settings - 8x AA - 5760x1080 - was getting 30 - 60 fps. 30 is the large open areas 60+ in closed areas. ( also to add, I didnt see it drop below 30 fps once, which is fine for me)
 
I think it's time he starts OC'ing that 920 for his reviews or swaps it out for a 950 so that someone else can put that 920 to good use!
 
6950 scaling is better than 6970 across the board. I knew I made the right decision :)

I can also echo 75% scaling in JC2 at 5760x1200.
 
bezant said:
When you go Eyefinity/Surround, then the 2GB helps out. That's been discussed, there were reviews on it on overlock.net, etc.
Click to expand...

kevinnem said:
Bez- My gut feeling was that the 2 gig was better then 1 gig, when you have high resolutions , and some AA involved. But I am having trouble finding the data to support this.
Click to expand...

bezant said:
OK, here's the review I was referring to: here

I think it is pretty clear that if you use AA and eyefinity/surround, you need more than 1GB
Click to expand...

Heres some backup for overclock.net
http://www.widescreengamingforum.com/wiki/NVIDIA_GTX460_in_Surround:_1GB_vs._2GB_-_Featured_Review

2GB is basically required at 3x1920x1200.
 
You must log in or register to reply here.
Back
Top