5850 vs 6950: my results

  • Thread starter Deleted member 174368
  • Start date
D

Deleted member 174368

Guest
Recently got my XFX 6950 and decided to post a mini-review against a modestly overclocked 5850 (775/1050).

Other system specs:
-AMD Phenom II X3 720 @ 3.4 GHz
-4GB DDR3 Ram

Picked four games which I felt best represented modern games. F1 2010, Need for Speed Hot Pursuit, Metro 2033 and Amnesia.

All tests were done at 1920x1080 and max settings unless stated otherwise. Gameplay runs were done multiple times for consistency between results.



F1 2010

To start off, the 6950 disappointed me slightly in F1 2010, but here you can see a discrepancy between a canned benchmark and real gameplay:

f1.png


While not a huge difference, in gameplay it did feel smoother than the 5850. If I were to go by the canned benchmark however, I’d say there was no difference.



Need for Speed Hot Pursuit

This game has some very pretty visuals, but unfortunately has no AA setting anywhere in the options. Forcing AA from the driver panel works, however.

nfs.png


Barely any improvement with 0xAA. Switching it up to 4xAA and it was much better. In fact, the 6950 did not lose any performance at all, compared to the 5850 which rendered 20% slower.



Metro 2033

I used the canned benchmark because this game uses a checkpoint system which makes benchmarking real gameplay quite tedious. Even so, it did paint the 6950 in an excellent fashion:

metro.png


With Advanced Depth of Field enabled the game isn’t as playable as I’d like, but it still doubled the 5850’s framerate. With ADoF off, the results are the same, and an average of 40 is more than satisfactory.



Amnesia: The Dark Descent

Setting SSAO to 128 samples, while extremely good looking, is incredibly taxing on hardware. I expected some improvement, but the results below are shocking to say the least:

amnesia.png


Echoing the Metro results, the 6950 literally doubled the 5850's FPS in this game and made this setting playable. It seems like in addition to Tessellation, SSAO was greatly improved with the 6900 series. I'm hoping we see more of Amnesia in hardware benchmarks.


I included the unlocked 6950 results (at 6950 clocks) to show that there is virtually zero performance improvement. I’m wondering if most of this performance increase people are reporting is due to higher overclocking limits, rather than unlocked shaders.

Going by the Metro 2033 and Amnesia results, it’s quite obvious that AMD did design the 6900 with the more taxing features in mind. On the other hand, it seems like buying AMD or Nvidia “based on the games you play” really does matter this time around. Regardless, next year (Crysis 2 and Witcher 2) will be quite interesting for the 6950 and the 6970.
 
Awesome work AceCR42!! I really enjoyed reading your post.

I wonder would it be worth trying a 6950 unlocked at the default 6970 clocks and then a 6950 at the same default 6970 clocks. Maybe that would show a difference in performance due to shaders better? Maybe at 6950 default clocks it can't get full value out of the unlocked shaders?

I am only guessing here :)

But, I really did like the way you compared the two cards and your statement about the 6950 been designed for more taxing features, well that's part of the reason I bought the card. To me it seems like the more you push the card the better it performs.
 
It would be more interesting if you show an unlocked 5850 as the ref cards can be unlocked for voltage and clockspeeds.. some can reach 1000Mhz core speed but i fear your system is holding both cards back.
 
something is up , you must be cpu lmited or something. In some games my frame rates have doubled going from a 5850 to a 6950 unlocked. I play at 5760x1080
 
eastmen said:
something is up , you must be cpu lmited or something. In some games my frame rates have doubled going from a 5850 to a 6950 unlocked. I play at 5760x1080
Click to expand...
Yes he is. He is only using a Phenom X3 720 at 3.4ghz. He will see improvements in the minimum frame rates despite the cpu bottleneck. But the cpu will in fact be holding the gpu back the maximum frame rates at the paltry resolution of 1920x1080.

I'm guessing he would need atleast another 400 to 600 mhz overclock to help alleviate the cpu bottleneck at that low resolution, use a bigger monitor, use more eye candy (higher levels of AA), or go eyefinity to push the bottleneck back to the GPU. The 400 to 600mhz I mentioned is just a guess especially since not many i5's or i7's bottleneck the newest cards.
 
Thanks for the benchies! As someone who owns a 5850 and is wondering whether to pick up an unlockable 6950 before they disappear, it's nice to have an apples-to-apples comparison of the two.
 
yup the cpu is definitely the bottleneck at that resolution. the clear example of that is your benchmark results in Need for speed: hot pursuit when you raised the AA to 4x and lost 0 performance, it means the cards not working at all. its also why you see 0 improvements between locked and unlocked. change your AA settings to 4x or 8x on both cards then try running the benchmarks again. it will show a much better example in performance difference.
 
Please overclock your HD5850 to these clocks 925MHz Core / 1175MHz RAM and post some results.
 
Yeah guys, I am aware I'm CPU-limited in several games. It's one of the reasons I post these benchmarks in case someone has a similar CPU and is thinking about an upgrade.

And I don't really like pushing my video cards to their OC limit; typically only hit the max that CCC allows. Reason is I live in a country where getting a RMA (or replacement card) is an expensive and lengthy procedure, so I'd like to minimize the risks of such.

eastmen said:
In some games my frame rates have doubled going from a 5850 to a 6950 unlocked. I play at 5760x1080
Click to expand...

In case you didn't notice, it did double my frame rates in both Amnesia and Metro.

sirmonkey1985 said:
its also why you see 0 improvements between locked and unlocked.
Click to expand...

This doesn't make sense though, because the GPU usage hits 100% in both Amnesia and Metro, yet the average framerate is largely the same.

edit: My particular CPU seems to hit a wall at 3.4GHz, getting 3.5GHz stable requires far too much voltage and it only gets worse from there.
 
Yea but when i say most games i'm talking all the ones i'm playing are close to doubling . L4D , Fear 2 , Dead Space , Crysis wars , TF2 and a few other small ones.


As for Metro , it could simply be the physx . With the 5850 your gpu limited and then with the new gpu you shift the bottleneck back to the cpu. DOF may also require more cpu cycles thus when you disable it you gain speed by shifting the bottleneck again ? I mean your getting close to a 100% improvement in both.

If you really want to test , downlock your cpu by say 500mhz and see what happens with the 6950. You don't have to switch back to the 5850. If your gpu limited then you should get the same framerate no matter your cpu clocks until it shifts back to cpu limited.
 
I find it hard to believe the unlocked 6950 is giving no speed difference vs. the regular 6950. Something is definitely wrong there.
 
tc17 said:
I find it hard to believe the unlocked 6950 is giving no speed difference vs. the regular 6950. Something is definitely wrong there.
Click to expand...

My guess he was still running it at 6950 clock speeds. There really isn't another explanation (other than a typo)
 
You must log in or register to reply here.
Back
Top