[fixed] Dragon Age 2 Low Performance on Nvidia Cards [fixed]

[notty22]

If they both do same which one you choose?

Clearly the one that is the least expensive, the one that consumes less power, the one that is faster, etc.

Now this differentiation by having features in certain games means that you aren't comparing hardware graphic cards head-on anymore - you are comparing game catalogs.

For the consumer this isn't ideal - the consumer wants the hardware companies to develop faster parts that consume less power and are cheaper and for the game developers to create fun games to play that run on the hardware available.

Seems to be a healthy system where everyone wins - but one where the consumer has great power.

Someone makes a crappy game - you don't buy it. Someone makes a crappy card - you don't buy it.

But I'm not even sure this is what happening with Dragon Age II -


my 6850, 1680x1050, takes a drop from 55fps at high settings to 25fps at very high settings (and this is without the hirestexture package!).

SSSHHHH !!! Your not supposed to be admitting that , it ruins the drama thats being played out here, and the pointing fingers.
Now seriously the game also plays perfect on Nvidia cards on high as well, its going to very high that something is happening.

 

[SirPauly]

If they both do same which one you choose?

Clearly the one that is the least expensive, the one that consumes less power, the one that is faster, etc.

Now this differentiation by having features in certain games means that you aren't comparing hardware graphic cards head-on anymore - you are comparing game catalogs.

For the consumer this isn't ideal - the consumer wants the hardware companies to develop faster parts that consume less power and are cheaper and for the game developers to create fun games to play that run on the hardware available.

Seems to be a healthy system where everyone wins - but one where the consumer has great power.

Someone makes a crappy game - you don't buy it. Someone makes a crappy card - you don't buy it.

But I'm not even sure this is what happening with Dragon Age II - my 6850, 1680x1050, takes a drop from 55fps at high settings to 25fps at very high settings (and this is without the hirestexture package!).

The consumer has always had great power and having more choice is much more ideal than not having it at all. If power, performance and price are so important to one, they can still buy their products just on this.

 

[exar333]

Totally awful......

Yikes!!

 

[Arkadrel]

ExarKun333

you can see the 6990 is CPU bottlenecked too, thats why the 5970 scores almost the same as the 6990 in 1920x1080 resolution.

To see the GPU's strength at those levels you need to look at the 2560x1600 resolution.

single 580 ~37 fps
single 6990 (dual gpu card) ~ 87 fps.

 

[zlatan]

I think some of the performance problems are related to a lesser know, but important DX10/11 feature named Conservative Depth Output. This is an optimization for the Depth Output. An early depth test run before the fragment shader so that the shader evaluation can be skipped if the fragment ends up being discarded.
Only the HD Radeon cards support this feature, so on the GeForce you need to use SV_Depth. The result is identical, but you gaining speed with the Conservative Depth Output.
http://szotar.sztaki.hu/dict_search...90222c74&vk=&L=ENG:HUN:EngHunDict&W=identical

 

[Arkadrel]

I think some of the performance problems are related to a lesser know, but important DX10/11 feature named Conservative Depth Output. This is an optimization for the Depth Output. An early depth test run before the fragment shader so that the shader evaluation can be skipped if the fragment ends up being discarded.
Only the HD Radeon cards support this feature, so on the GeForce you need to use SV_Depth. The result is identical, but you gaining speed with the Conservative Depth Output.

So its a Dx11 thingy nvidia didnt do right?
Kinda like how AMD tessellation isnt quite as strong as Nvidias.

And this is a game that shows it?

 

[RussianSensation]

Techspot.com just did a GPU&CPU Performance Test for Dragon Age II:

Testing Methodology:
"We used Fraps to measure frame rates during a minute of gameplay from Dragon Age II's first single-player level. The test begins as the player engages a number of hostiles on top of a rocky mountain outcrop. The scene has heaps of action taking place and should help to provide an excellent representation of the performance gamers can expect to see throughout the game."

Nvidia drivers used:
"For testing we tried the Forceware 266.66, 267.24, 267.31, 267.46 and even a set of leaked 270.32 drivers, none of which improved the performance of GeForce cards where it belongs. We understand both BioWare and Nvidia are working to solve this issue and we will update our benchmark scores if that happens sometime soon."

.....................................................................

Very High (2560x1600) - This test is very demanding. At 2560x1600 in DX11 with 4AA/AF you'll need an HD6990 (or similar).

High (2560x1600) - HD6970 is 89% faster than a GTX480=HD5830.

Very High (1920x1200) - Again pretty much hammers all graphics cards, except the 6990

High (1920x1200)

CPU Scaling - "For the most part the game only uses two cores while we saw very light usage on a third"

CPU Scaling 2 - Completely GPU limited with a Core i7.

"Needless to be said, at this point we were unable to get SLI working correctly, and in fact getting any Nvidia graphics card to perform as it should when using the DX11 rendering mode won't be possible until Dragon Age II is patched or a new Forceware driver fixes this issue. Having the mighty GeForce GTX 580 performing below the Radeon HD 5850 is evidence enough that there is something very wrong here.

In fact, most single GPU owners will want to use the high quality preset regardless of whether it’s an AMD or Nvidia card."

 

[GaiaHunter]

Techspot.com just did a GPU&CPU Performance Test for Dragon Age II:

Testing Methodology:
"We used Fraps to measure frame rates during a minute of gameplay from Dragon Age II's first single-player level. The test begins as the player engages a number of hostiles on top of a rocky mountain outcrop. The scene has heaps of action taking place and should help to provide an excellent representation of the performance gamers can expect to see throughout the game."

Nvidia drivers used:
"For testing we tried the Forceware 266.66, 267.24, 267.31, 267.46 and even a set of leaked 270.32 drivers, none of which improved the performance of GeForce cards where it belongs. We understand both BioWare and Nvidia are working to solve this issue and we will update our benchmark scores if that happens sometime soon."

.....................................................................

Pretty much mirrors my results with a 6850+Phenom II X4 @3.6.

 

[Seero]

I think some of the performance problems are related to a lesser know, but important DX10/11 feature named Conservative Depth Output. This is an optimization for the Depth Output. An early depth test run before the fragment shader so that the shader evaluation can be skipped if the fragment ends up being discarded.
Only the HD Radeon cards support this feature, so on the GeForce you need to use SV_Depth. The result is identical, but you gaining speed with the Conservative Depth Output.
http://szotar.sztaki.hu/dict_search...90222c74&vk=&L=ENG:HUN:EngHunDict&W=identical

Hey guys, here is a guy with some insight, and knowledge to Dx11. Most importantly, this is related to the issue people are having.

My questions, how do you know DA2 is using this function, and can it be the sole reason of the huge performance impact? If SV_Depth is the cause, doesn't it also impact graphic setting high too? If it does, then the effect is exponential from high to very high, but reports indicate minimal difference between high and very high, and it is mostly on tessellation.

 

[Arkadrel]

I kinda like that fact that the "high settings" -> "very high settings", the 6990 ONLY loses 2fps, all the other cards... get the air taken out of them.

If SV_Depth is the cause, doesn't it also impact graphic setting high too?

what should worry more, is that *IF* this is a dx11 feature not commanly used, when it does get more used... same thing with other games?

 

[zlatan]

So its a Dx11 thingy nvidia didnt do right?
Kinda like how AMD tessellation isnt quite as strong as Nvidias.

And this is a game that shows it?

Well, technically you can support Conservative Depth Output with a driver based emulation, as GeForce support it this way. It's not a problem because the DX11 enable this solution, but the hardware based approach is faster.
This game just showing some benefit about it, but I didn't see the source code. I just said that Conservative Depth Output is one of the main reason of the performance difference, but not the only one.

I think most people misunderstood tessellation. You don't able to do micropoligon render with today's GPU. Rasterizing small triangles on a quad pixel pipeline is inefficient. Maybe DX12 improve this.

 

[Skurge]

Welcome to the forum zlatan :thumbsup:

Good to have someone with your knowledge around here.

 

[zlatan]

Hey guys, here is a guy with some insight, and knowledge to Dx11. Most importantly, this is related to the issue people are having.

My questions, how do you know DA2 is using this function, and can it be the sole reason of the huge performance impact? If SV_Depth is the cause, doesn't it also impact graphic setting high too? If it does, then the effect is exponential from high to very high, but reports indicate minimal difference between high and very high, and it is mostly on tessellation.

In the GDC there are some document about the Lycium Engine. The renderer is forward, but there is a depth pass for visibility and early pixel rejection. But as I said this is just one reason for the performance impact, not the only one. To use Conservative Depth Output you need a DX10/11 renderer. The settings are independent from this solution.

 

[bryanW1995]

The demo can only manage 34 fps on a GTX580 at 2560x1600. In comparison the HD6990 with AUSUM gets 87 fps. 3x as much speed means this game is clearly GPU limited. Also, in their testing they "only" used a 3.3ghz Core i7. CPU speed is a non-issue in this game compared to GPU horsepower.

hmmm, it looks very similar to dao but maybe they changed a bunch under the hood. I'll do some testing when I get home.

 

[bryanW1995]

I've never seen a ranged attack miss. Such bullshit that the arrows curve in mid flight to track you. D:

ruh roh... I haven't even played yet and I'm already starting to regret purchasing it.

 

[bryanW1995]

You shouldn't blame AMD if nVidia cards are performing poorly. and you shouldn't blame nVidia if AMD cards are performing poorly. Except in special cases where features are locked away from the other vendor. nVidia have done this, I don't remember AMD doing this.

nVidia or AMD should NOT be performing so poorly in any AAA game no matter who helped the devs more.

Just like AMD in CIV5, they shouldn't be performing so poorly when all their cards highend are basically performing the same as their low end cards. That did improve with Cat 11.4 though.

Yes, you can say AMD did a great job by making sure the game worked great on their hardware, but nVidia should also be lambasted for not having the game work right on their hardware. Same can be said to AMD about CIV5 although to a lesser extent.

civ 5 is one of the few games out there to make extremely heavy use of tesselation. NV cards shine in that scenario. I wish I could say that I bought the gtx 460 b/c of civ, but I was just lucky that I got it when I did.

 

[Seero]

In the GDC there are some document about the Lycium Engine. The renderer is forward, but there is a depth pass for visibility and early pixel rejection. But as I said this is just one reason for the performance impact, not the only one. To use Conservative Depth Output you need a DX10/11 renderer. The settings are independent from this solution.

I am not aware there exists a hardware solution for Conservative Depth. I however know that it is a Dx10.1 function that Nvidia doesn't support. Like you said, Conservative Depth Output is a Dx11 FP, it isn't possible that Nvidia did not implement this. While it is possible that AMD's algorithm is better than Nvidia for this particular FP, but I have yet seen any reports on that.

Like you said, Conservative Depth Output is a better approach then SV_depth, but AFAIK both camps implement it fine.

Maybe that part of code was developed before Dx11 arrives, and Conservative Depth, with is a Dx10.1 call which is only available to AMD was used, and developer designed to use SV_depth approach for Nvidia. In that case ... it will take a while to fix. Edit: this is a wild assumption.

 

[RussianSensation]

If Conservative Depth Output was so critical, what explains NV cards getting such a significant boost in performance from Very High to High? Even without the high resolution texture pack, NV cards almost double their framerates by simply going to High. Almost the same can be said for AMD's cards.

1920x1200 4AA/16AF

HD6950 VH = 35
HD6950 H = 65 (+86&#37

GTX570 VH = 24
GTX570 H = 45 (+87.5%)

Both cards suffer almost identical performance degradation in percentage terms when subjected to the more intensive visual settings. If Conservative Depth Output was such a severe issue for NV, wouldn't their cards suffer an even larger performance drop as you continue to increase the visual settings? This doesn't seem to be the case. On the other hand, if CDO was a bottleneck, I doubt the 570 would be able to net an 87% performance increase when switching from VH to H. It seems to me, there are a lot of other things at play here.

 

[sticks435]

ruh roh... I haven't even played yet and I'm already starting to regret purchasing it.

Note that this is only for enemy ranged attacks like arrows. You can still miss plenty, but their arrows pretty much never miss. Though it was like this in DA:O too.

 

[Skurge]

Note that this is only for enemy ranged attacks like arrows. You can still miss plenty, but their arrows pretty much never miss. Though it was like this in DA:O too.

Its like that in almost every RPG, the spell or arrow hits the character, but it can still "miss", depending on a bunch of factors.

 

[sticks435]

Its like that in almost every RPG, the spell or arrow hits the character, but it can still "miss", depending on a bunch of factors.

Yea, but that's usually if your standing still. I've never really seen a "Kennedy style magic bullet" arrow that can do 90 or 180 turns to hit you while your running.

 

[Idontcare]

Note that this is only for enemy ranged attacks like arrows. You can still miss plenty, but their arrows pretty much never miss. Though it was like this in DA:O too.

Its to make up for the timing distortion that goes into the imbalance of speeds of you and a flying arrow.

Were the arrow's speed actually scaled up to be appropriate to the speed at which you could move with your legs then you'd nary have a second to even see the arrow coming at you let alone "time" to move out of its way.

In the interest of keeping it enjoyable they slow the arrows down, now you seem to have super-human speed and can dodge arrows all day long. To balance it back out the arrows now have locking abilities and mid-flight course correction capability.

How much reality were you looking for in your fantasy-based game? I'm happy I don't have to limit my perspective to the eyes of my character, I kinda like seeing the battlefield as if my character's eyeballs hover some 20ft above and 10ft behind his head. Makes seeing those unrealistic arrows coming at me all the easier.

 

[MentalIlness]

Wow, My 6870 "if id play this game" identically matches a GTX580 at the res I would play it. LOL

Whats even more weird, is a 5830 outruns a GTX480, and a 5770 outruns a GTX470 :/

 

[sticks435]

Lol, I understand where your coming from, it's just strange to be running across the battle field and have an arrow do a 90 degree turn to "hit" you. It's like that sci-fi movie with Tom Selleck and Gene Simmons where the bullets can travel around corners and stuff. Usually only time I really notice it is when my mage is trying to run from an archer.

 

[chris2012]

The 267.59 driver dramatically improves performance apparently. Like almost double it making very high dx11 playable smoothly.

Now the problem is it's only for 550 ti and 450. So you need to mod/get a modded inf file for the rest of the cards.

Sidenote: can anyone with xp tell me if it's dangerous in any way to use inf modded drivers? As in what's the worse that can happen? Game crashes? Or can they actually damage the cards they weren't meant for.