GTX 460 768MB suffering huge FPS dip when going from DX9 to DX10/DX11

[jut703]

Just curious, how much of a performance hit do you guys get when going up from DX9 to DX10/11?

In Crysis, DX9 gives 32.8 FPS while DX10 gives 28.85.
In Warhead, DX9 gives 48.31 FPS while DX10 gives 45.24.
In F1 2010, DX9 gives 45 FPS while DX11 gives 30 FPS.
In DiRT 2, DX9 gives 59 FPS while DX11 gives 47.8.

Is this normal? If it isn't, what solution can I apply? Will reinstalling drivers help?

Here are my pertinent specs:

Intel Core 2 Duo E6550 @ 2.8 GHz
4GB DDR2-800 RAM
Inno3D GTX 460 768MB

 

[celpas]

that is completely normal.DX 10/11 is supposed to be faster than dx9 if done right but devs don't do it as it was supposed so it becomes slower.your fps loss seems right to me.

 

[gamerk316]

Most games still support XP, so you have a "base" DX9 path, with occasional branches for DX10/11. As a result, the DX9 path is the most optimized.

 

[jut703]

So my huge dips are fine?

From what you're saying, upgrading CPU will still cause dips like these though at increased framerates?

 

[gamerk316]

Probably.

 

[accolite]

It's normal!

 

[Kewlx25]

Some games enable extra special effects when using 10/11 because 9 can't do them.

I would compare screen shots and see if there is any gained quality. If not, stick with 9 if it's more performant.

 

[notty22]

Here is a great example of what you have concluded.
Dirt 2 CPU benchmarks with DirectX 9 and DirectX 11: Phenom doing well, quad-cores rule
http://www.pcgameshardware.com/aid,700780/Dirt-2-CPU-benchmarks-with-DirectX-9-and-DirectX-11-Phenom-doing-well-quad-cores-rule/Practice/

 

[bystander]

that is completely normal.DX 10/11 is supposed to be faster than dx9 if done right but devs don't do it as it was supposed so it becomes slower.your fps loss seems right to me.

This is a misunderstood concept.

DX10 is faster than DX9 when doing the same code.

DX11 is faster than DX10 when doing the same code.

Games written for DX10 usually include new visual improvements that DX9 cannot perform which will cost you some performance.

Games written for DX11 usually include new visual improvements that DX10 cannot do, so it also slow things down. Things like tessellation and ocular occlusion are added.

The general guideline most dev's follow is to not write a game in DX10 or DX11 if you can get the same visuals out of DX9. Even if it would be faster to write it in DX11. This is because more people can play it if it's written in DX9. Same goes for DX10 vs DX11.

 

[renz496]

^+1

i think that many people still don't get this thing right. i still remember right before dx11 comes out many people think that dx11 games will still run faster than dx10/dx9 even with the new visual improvement added in the game. (which is obviously eating your performance if you try to think about it logically) 😉

 

[iam2thecrowe]

i wouldnt say what you are getting is "huge dips" id say theyre exactly what i would expect. As renz said, your getting better visual quality, of course performance is not going to remain the same.

 

[jut703]

I guess I'll be sticking with DX9. I'd prefer consistently smooth framerates rather than eye candy that I won't be looking at when I'm shooting at aliens or hitting the apex.

Are there any games that look substantially better in DX10/DX11 than DX9? Crysis and F1/DiRT don't look particularly different.

 

[celpas]

This is a misunderstood concept.

DX10 is faster than DX9 when doing the same code.

DX11 is faster than DX10 when doing the same code.

Games written for DX10 usually include new visual improvements that DX9 cannot perform which will cost you some performance.

Games written for DX11 usually include new visual improvements that DX10 cannot do, so it also slow things down. Things like tessellation and ocular occlusion are added.

The general guideline most dev's follow is to not write a game in DX10 or DX11 if you can get the same visuals out of DX9. Even if it would be faster to write it in DX11. This is because more people can play it if it's written in DX9. Same goes for DX10 vs DX11.

I still think that dx 10 is not optimised well.take for example crysis warhead,when I run it in dx 9 at everything max I get the performance of dx10 gamer settings(high).Also it feels much more responsive than dx10.From comparing dx10 vs dx9 in the game all I could find was that dx10 had a slightly more draw distance and a brighter sun but it took nearly 5fps away.The only games in which dx10 was actually done right were metro 2033 and just cause 2.In metro 2033 there was only a 2fps difference between dx9 and dx10 and the difference was enormous.the dx9 image was way too bright and less realistic while dx10 looked darker and more realistic.

 

[celpas]

Just curious, how much of a performance hit do you guys get when going up from DX9 to DX10/11?

In Crysis, DX9 gives 32.8 FPS while DX10 gives 28.85.
In Warhead, DX9 gives 48.31 FPS while DX10 gives 45.24.
In F1 2010, DX9 gives 45 FPS while DX11 gives 30 FPS.
In DiRT 2, DX9 gives 59 FPS while DX11 gives 47.8.

Is this normal? If it isn't, what solution can I apply? Will reinstalling drivers help?

Here are my pertinent specs:

Intel Core 2 Duo E6550 @ 2.8 GHz
4GB DDR2-800 RAM
Inno3D GTX 460 768MB

btw what settings do you play crysis warhead at?Just interested because you are getting 45 fps in the game

 

[jut703]

Those are controlled benchmark runs.

Both Crysis and Warhead are at 1920x1080, High Settings, 0xAA.

In Crysis I use the heavy Assault Harbor bench.
In Warhead I use the Frost bench which is the heaviest of the bunch but not as heavy as Assault Harbor.

 

[celpas]

your card is being heavily bottlenecked by your RAM and your processor bcause I just ran the crysis assault harbor benchmark and got 36fps average.My rig is
Intel core i7 920 2.66ghz
MSI Geforce GTX 260 1792MB Single card
6gb DDR3 1600MHZ RAM

1/25/2011 8:34:26 PM - Vista 64
Beginning Run #1 on Map-harbor, Demo-Assault_Harbor
DX9 1920x1080, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=1, Time Of Day= 5
Global Game Quality: High
==============================================================
TimeDemo Play Started , (Total Frames: 4100, Recorded Time: 132.23s)
!TimeDemo Run 0 Finished.
Play Time: 112.02s, Average FPS: 36.60
Min FPS: 24.77 at frame 2902, Max FPS: 55.44 at frame 1851
Average Tri/Sec: 12700486, Tri/Frame: 346998
Recorded/Played Tris ratio: 0.04
TimeDemo Play Ended, (1 Runs Performed)
==============================================================

Completed All Tests

<><><><><><><><><><><><><>>--SUMMARY--<<><><><><><><><><><><><><>

1/25/2011 8:34:26 PM - Vista 64

Run #1- DX9 1920x1080 AA=No AA, 32 bit test, Quality: High ~~ Last Average FPS: 36.60

Try going to www.geforce.com and update your drivers.

 

[accolite]

You cant compare your setup to his!?

And of course your system will run faster because of your components.


You have a GTX260 with 1792MB??? He only has 768MBMB!

Different models will average different results, common sense there.

 

[bystander]

I still think that dx 10 is not optimised well.take for example crysis warhead,when I run it in dx 9 at everything max I get the performance of dx10 gamer settings(high).Also it feels much more responsive than dx10.From comparing dx10 vs dx9 in the game all I could find was that dx10 had a slightly more draw distance and a brighter sun but it took nearly 5fps away.The only games in which dx10 was actually done right were metro 2033 and just cause 2.In metro 2033 there was only a 2fps difference between dx9 and dx10 and the difference was enormous.the dx9 image was way too bright and less realistic while dx10 looked darker and more realistic.

Crysis in DX10 mode used new shading methods not availible in DX9. It definitely added new features. However, those newer, more demanding features may not be worth the performance loss, and might not even be noticeable, but they did new techniques regardless. That is why it performs slower in DX10.

Metro 2033 was interesting to me. I was playing with some of the advanced features on the benchmark, like DoF, advanced PhysX and such, but could hardly tell a difference. Even going from high to very high, the quality difference was small, yet the performance impact was huge. Looking at the description field, the difference in a lot of areas was whether the game would approximate things or if it would calculate it out.

 

[celpas]

You cant compare your setup to his!?

And of course your system will run faster because of your components.


You have a GTX260 with 1792MB??? He only has 768MBMB!

Different models will average different results, common sense there.

He owns a gtx 460 which should run faster than a gtx 260 no matter what the components.He is running at 1920X1080 which is more of a gpu bound resolution.

 

[bowzef]

He owns a gtx 460 which should run faster than a gtx 260 no matter what the components.He is running at 1920X1080 which is more of a gpu bound resolution.

yep u are correct a GTX 460 would out perform ur GTX 260 and yes i would say few things is bottlenecking his PC maybe if he upgraded to Quad Core and got more ram and OC the RAM he would get lil more out of it

 

[accolite]

Oh my mistake, I was in a hurry and misread there what you have and what op has, all this 260 460 560 is starting to blend together for me.

 

[celpas]

There is also a possibility of vram due to which my gtx 260 can be faster.Crysis,metro2033,gta 4 all gobble up a lot of vram and he only has 768mb of vram which is not sufficient for crysis,metro.Also accoording ot the grapgh the 460 768mb is 6%faster than the gtx 260

 

[jyjjy]

He owns a gtx 460 which should run faster than a gtx 260 no matter what the components.

Not when the components are an E6550 vs i7.

 

[bowzef]

yeah that list of cards needs updateing missing GTX 570 and yeh i know GTX 560 out but too new to be posted in 😉

 

[celpas]

Not when the components are an E6550 vs i7.

he will see bottleneck in cpu bound games like bfbc2 and gta 4.crysis ,metro 2033 don't utilise cpu much at all.In fact metro 2033 showed no difference when moving from core 2 duo to an i7
http://www.legionhardware.com/articles_pages/metro_2033_performance_guide,9.html

besides he is playing at 1920X1080 resolution which is more gpu bound.