Crysis 2 DX11 Performance Discussion Thread

http://translate.googleusercontent....e.html&usg=ALkJrhioCKuA_x98DsgzDBh9SYtAvA7jXQ

Analysis of Crysis 2 dx11!!! Realy good!
Originaly in french but with google translate!

 

WOW!!Crysis 2 is working great on nvidia cards

 

Well going to windowed mode DOES disable Crossfire... :nerd:

 

I'ved done some test with Heaven engine 2.5...
1980*1080 af16x aa4x Normal ----->42.5fps Avg Close to a GTX580 Stock Score!!!
On ATI HD6950@6970@1000/1580

 

Someone can benchmark at 1680x1050/Ultra/2x/4x AA with Adrenaline pls?
I want compare with my system.
Catalyst 11.7 preview (not "b").

 

Great read. Thanks

For AMD users, it clearly shows that using a lower tesselation factor of 16 or even 8 will not reduce drastically the quality but will result in a great boost.

Shading and postprocessing should be put at extreme for an additionnal boost. Shading will remove quality reflections so you might want to skip and remove only SSDO through a console command.

Finally, postmsaa should be put to off in an cfg file. Beware changing options as you may reactivate it. Check status in console with postmsaa = ..


Edit:

To reduce SSDO, i would try these:

r_SSAO screen space ambient occlusion:
0 - disabled
1 - SSAO technique with normals
2 - SSAO technique with normals and temporal accumulation
3 - volumetric obscurance technique with jittering
4 - volumetric obscurance technique with jittering and temporal accumulation
r_SSAODownscale use downscaled computations for SSAO
r_SSAOQuality SSAO shader quality [CRYSIS21:0 - low spec, 1 - med spec, 2 - high spec, 3-3 highest spec]

Edit 2:

Just tried this config with MLAA and 8x tess factor.

g_skipIntro = 1
r_MultiGPU = 1

cl_fov = 65
pl_movement.power_sprint_targetFov = 60 (field of view while Sprinting)
r_DrawNearFoV = 70

sys_MaxFps = 120
sys_spec_GameEffects = 4
sys_spec_ObjectDetail = 4
sys_spec_Particles = 4
sys_spec_Physics = 4
sys_spec_PostProcessing = 3
sys_spec_quality = 3
sys_spec_Shading = 3
sys_spec_Shadows = 4
sys_spec_Sound = 3
sys_spec_Texture = 3
sys_spec_VolumetricEffects = 4
sys_spec_Water = 4

r_UseEdgeAA = 3
r_PostMSAA = 0
r_PostMSAAEdgeFilterNV = 0
i_mouse_smooth = 20
r_SSAO=2

Lots of jaggies but performance is great with 70-90 fps

 

Default/765MHz/1125MHz - 24,9 fps
Default/900MHz/1150MHz - 28,3 fps
8x Tess/900MHz/1150MHz - 34,1 fps
Catalyst 11.7 preview.



What do you think about my results?

I like to know 5850 CF performance, someone can help me?

 

Considering "Central Park" is the benchmark map using more tessalation, the results are about as they should be for you, although I was expecting a little bit more from that CF setup. Since you have the 5850s OC'ed and my HD5970 did 40 FPS @ stock clocks (same as HD5850 stock clocks), and did it @ 1080p, I was expecting from 45 to 50FPS @ 1050p.


Johnny

 

Heres my results, did the same result setup as pyshcok9

Default CCC setting , stock clocks for first run.
Default CCC setting , Oc'ed to 900/1200 second run.
8x Tess setting, still oc'ed to 900/1200 third run.
using 11.7 preview


They on par with other peoples 5970 results?

 

is this ok?

 

Here's another one for you all to compare to...

This is with Tesselation clamped to x8

http://imageshack.us/photo/my-images/33/unledcqpw.png/

The above bench was done on my crossfire system (2 * HD5870) at 900/1250 on water (not that water is needed for these clocks - it's an HTPC and the water is there for quietness)

My other system has a single HD 6970 in it.

-------------------------

For Crysis DX11 it has become very clear to me that my HD5870 crossfire is vastly superior to my other system, a single HD6970 in all respects when running Crysis 2 DX11 on Ultra settings.

This contrasts to a previous situation a year or so back where I found that my (then-single) HD5870 was as good as my previous cards (HD4870 in crossfire) for Crysis and Crysis Warhead (cryengine 2 games)

I put this down to the following reasons:

* The HD 5870 is tesselation-weak, and shader-rich (1600 shaders) - the bottleneck with this card was always tesselation, as could be easily seen from benchmarks at the time such as unigine and stone giant, and the fact that it's legacy performance in DX9 and DX10 was so good.

* The HD 6970 is tesselation-strong, but relative to itself, shader-poor (it has in fact slightly less shaders than the HD5870 at 1536, although they do have some design improvements). If you want proof of this point, run Heaven benchmark first on version 2.1 and then again on version 2.5 with a single HD6970. The tesselation load is

identical on both these benchmarks, the only difference being that the shader load got more complex in version 2.5 with a new SSDO implementation. You will be amazed at the negative difference in FPS due to shader overload in 2.5 on this card. Untick the SSAO box in the settings dialog and watch the frames leap back up!

Back on topic to Crysis 2 on Ultra, I belive that thee reason that the Nvidia cards seem to take less of a hit with Tesselation is simply that AMD don't yet have any single GPU card that adequately balances tesselation vs. shading. The Nvidia Fermi architecture did this from day 1, meaning you can't bottleneck those cards with either tesselation or

shading, as they share these resources internally. We do however, have CCC and the ability to clamp tesselation on, say, x8 to partially mitigate this issue on AMD cards however.

AMD will finally hit the sweet spot with the new 7XXX which will basically be the 6XXX architecture but with lots more shaders - hopefully nearly 3200 (they can now fit on the chip since the 40-28nm die shrink will double the transistor count). This architecture should finally give single-GPU cards that nicely balance Geometry and Shading

Performance - but that remains of course to be seen.

So Ironically, (despite my investment in the HD6970 the day it came out last year) the best balance available to me at the moment for Crysis 2 is 2 * HD5870 in crossfire, since these effectively reduce the tesselation bottleneck in these cards by load balancing frames over two tesselators in two GPU! Therefore assuming a GPU-bound scenario

where tesselation is the bottleneck, we can go from 25fps to nearly 50 in this instantaneously occuring scenario and keep the game smooth. And at 3200 shaders in total, this combo can also handle SSDO and other new expensive lighting techniques.

The same theory applies to other cards and combos using crossfire-x such as the combos 2*HD58XX, 2*HD68XX, as well as the 'crossfire-on-a-card' HD5970 and this can clearly be seen from some of the benchmarks posted in this thread.

As for a new CAP profile helping matters, this has been debunked, since in GPU-bound situations crossfire scaling can already be nearly 100 percent on Cryengine 3. It's clear however, that AMD have everything to gain by driver-optimisation in Crysis 2 DX11.

Back in 2008/2009 almost every other Cat release had some crysis performance improvements. When the first Crysis game launched, there was 10-15 fps difference between running DX9 and DX10 on the same card. Now try it today, and you will see that the gap was closed. Hardware permitting, they may be able to do similar again.

Also, as has been mentioned here, Crytek are unlikely to have done any optimisation for AMD cards (probably only Nvidia). The fact remains though that the future looks bright, since AMD did a great job of reverse-engineering Cryengine 2 and will hopefully do so again with Cryengine 3.

 

Your post makes sense, but there is no way an HD5870 should outperform an HD6970...

Crytek simply didn't bother to optimize the game for AMD cards. I guess once AMD gets new drivers with performance boosts then we will see the HD6970's outperforming the HD5870's, or so at least I hope.

 

Can you make an Heaven engine bench 2.5 with One 5870
with same settings as this!!
I'ved done some test with Heaven engine 2.5...
1980*1080 af16x aa4x Normal ----->42.5fps Avg Close to a GTX580 Stock Score!!!
On ATI HD6950@6970@1000/1580 with an E8600@4.7ghz :bang:

 

Sorry if I wasn't clear - the nature of my comparison was actually:

2*HD5870 vs 1*HD6970

You are quite right a single HD5870 does not outperform a single HD6970 in any test I've seen.

 

You were pretty clear to me! However, a single HD5870 does beat a single HD6970 in Crysis 2 for whatever reason, rumors say the HD6xxx cards are tesselating through shaders and not even using their dedicated tesselation unit. :bang:

That would explain why the HD5870 is outperforming an HD6970 in Crysis 2... Sad but true. :wanker:

Hopefully AMD can fix this sooner or later, I just hope it is not the same story for Battlefield 3 and hopefully AMD performs as good as Nvidia.

 

Mine HD5970 @ stock did 39.8 FPS; yours did 39.4 FPS, so, yes, indeed, results are typical.
Also, I experienced about the same performance boost as you did when OC'ing the card: got Crysis 2 over 60 FPS average on Central Park bench.

This "little" card of ours, although being 18 months old, is pretty much capable to runs Crysis 2 @ 60 FPS. Needs OC'ing but it does eat it nevertheless. :banana:


Johnny

 

I just repeated your exact test with Heaven 2.5 and got:

Single HD5870 FPS average 31.1
CrossfireX HD5870 FPS average 61.8

Your score of 42.5 with the 6950 unlocked+clocked is good - similar to my HD6970 at 980/1500

I have seen that on the GTX 580 although the average framerate is about the same as this, that the overall framerate is steadier with less deviation from the average (smaller peaks and troughs).

 

I must admit I haven't tested a single HD5870 against a single HD6970 on Crysis 2 DX11 ultra.

I will go and do it now...

 

Thanks! Keep me updated! The HD5870 will beat the HD6970 sadly.

Hopefully AMD can fix this performance issues to do justice to the AMD HD69xx cards against the GTX570 and GTX580...

A GTX570 SLI is almost double performance of an HD6970 CF in this game.