17 GPU comparison in Steam using Serious Sam 3

[sonnet]

It seems to me that performance are in line with windows counterpart.
The 6850 makes 37 fps at 1920x1080 with ultra quality, which is similar to what other players recorded on youtube with windows (between 35-45).

I'm just really surprised how good kepler cards are doing, particularly compared to fermi generation.
The gtx 680 score twice the fps of the gtx480 at high resolution, while usually in windows the gtx480 score the 70% of the performances of the gtx 680.

[Kano]

As some nice person sent me SS3 as gift i could now check cf with my hd 5670 cards, fglrx 12.11 beta 11, after pressing alt-enter twice. cf has definitely an impact for that game, but a NEGATIVE one. At beginning of level 6:

CF disabled



CF enabled

[Pallidus]

So who tested the things better, you or me?

dun DUN DUN


germans vs greeks go:

[Kano]

Yesterday i repeated my Heaven 3.0 tests with Win and Linux and the result was the same like 7 month ago on Win. On Linux the rendering artefacts have been fixed but no speed increase. On Win you can enable CF logo, but you see that also wenn you run Heaven with OpenGL renderer, but only the DX renderer which is 50% faster than the OpenGL one without Tesselation scales perfectly (over 90% increase). OpenGL has got no speed diff on Win or Linux. So who tested the things better, you or me?

[Kano]

I am pretty sure that there is no Heaven CF profile for OpenGL. I did lots of benchmarking in April and only DX had a CF profile (on W7). Why should i get new card, i have got already 2x HD 5670 for exactly that purpose and installed em yesterday again in my box for testing. I could not even see an improvement with Doom 3 (which had some speed differences in an early phoronix test years ago) - i only noticed that when you rename the binary then you get around 15% more speed - that works for dhewm3 as well, when the binary (not a symlink) is called doom.x86 and does not matter if i enable or disable CF.

So to sum up, CF is win only feature, if a game/benchmark has got 2 renderers usually only the DX codepath has got a CF profile. The only interesting thing i found was that Rage was accellerated by CF as well but Nvidia does not provide a SLI profile.

[Dr.Paneas]

@Dr.Paneas

I do not say CF is not working on Win but you pretend to benchmark on Linux. Also you can not replay demos with SS3, do you play always the same or do you benchmark a standing picture?

I tried to play always the "same" pattern, that's why it took me 2.5 weeks benching. Way too much time consuming. I wish there would be a timedemo for SS3...

That has been told I had to play the same thing twice and then put an average on that. Most of the time I got:
1st run: 55 FPS
2nd run: 53 FPS
3rd run: 55 FPS
4th run: 54 FPS
5th run: 55 FPS

sum and divide ... it's 54,4 ~ 54 FPS.

Same level, same duration and as much as I could to simulate the same walkthrough. My results are not zero point accurate, but they are not fake as you imply. Since there is no a timedemo, there is no other way to benchmark the game. So if anyone would like to know about Serious Sam 3 performance (an not Steam games in General) my article is okay. May not accurate, but definetaly not fake.

IF there was a timedemo, I would have already use it. BUT since there is not, I tried my best to give you a glimpse of my experience with SS3 in my setup. That's all about. Have a first taste, nothing more. Everything are in expermental, beta phase, from driver and SS3 to the Steam itself. The numbers are expected to be innacurate, but not fake.

That has been told I will benchmark now with Team Fortress2 beta, where there is such a timedemo feature in order to replay the same demo and avoid such misunderstandings.

As for CFX, find a couple of 6970 or 6950 and run Unigine Heaven benchmark for Linux (not Windows).

my eyes are squared, need some sleep. good night

[Kano]

@Dr.Paneas

I do not say CF is not working on Win but you pretend to benchmark on Linux. Also you can not replay demos with SS3, do you play always the same or do you benchmark a standing picture?

[Dr.Paneas]

Also SS3 has got no real benchmark feature.

bmkResults() - Print the previous benchmarking results to the console, if any.
bmkStartBenchmarking(float tmStartIn, float tmDuration) - Start benchmarking, starting in tmStartIn seconds and lasting for tmDuration seconds. When done, print the results to the console.
bmkStopBenchmarking() - Immediately stop benchmarking and print the results. This does nothing if benchmarking isn't currently being done.

Source: Serious Sam 3 Wiki (Console Commands)

Did really anybody get a speed increase by using CFX with Linux?

I use a pair of HD 6970 using CaymanXT architecture. These cards work so good together that AMD decided to make them a sandwich and release AMD Radeon HD6990.

So depending on:

• GPU architecture
• Drivers
• OpenGL, DirectX9, DirectX10, DirectX11
• Game engine itself
• Resolution
• Anti Aliasing
• Tessellation

the results could vary from one to another.

For example:
In Call of Duty: Modern Warfare 2, crossfireX doesn't work
- Radeon 6970: 140 FPS
- 2xRadeon 6970: 139 FPS
link: http://www.guru3d.com/articles_pages...review,14.html

or there is very low scaling
- Radeon 6950: 137 FPS
- 2xRadeon 6950: 150 FPS
link: http://www.guru3d.com/articles_pages..._review,5.html


However, if you pick another game, like Far Cry 2 there is double performance (huge scaling)
- Radeon 6970: 83 FPS
- 2x Radeon 6970: 154 FPS
link: http://www.guru3d.com/articles_pages...review,15.html

one more
- Radeon 6950: 76 FPS
- 2x Radeon 6950: 147
link: http://www.guru3d.com/articles_pages..._review,6.html

As you can see in some cases there is huge scaling, and in some other does not. In Serious Sam 3 I got that scaling. Sorry but... I did
I suggest you start benchmarking with real games instead of obsolete OpenArena, Doom3, Unreal Tournament and similar stuff. If you want to see scaling you have to use Unigine Heaven and start messing with settings. Check my last review for Radeon HD 6970 and Radeon 6950 in Crossfire using DirectX9, 10 & 11. The review is in greek language so you have to use Google Translate to figure out what I am saying in there.

This must be a FAKE.

Oh c'mon... give me a break. Even a dead clock is right twice a day.
This is me, flashing my Radeon 6950 into a 6970 using TechPowerUp's mod last year.

http://www.youtube.com/watch?v=0aQ5gQ7EUSo

[Kano]

This must be a FAKE. Did really anybody get a speed increase by using CFX with Linux? I want to see benchmarks. Also SS3 has got no real benchmark feature.

[Pallidus]

wow nice work


but that is what I feared, almost double the perfomance....


I can safely say that is due to several things


1. years of perfomance optimization and driver development in windows

2. poor linux drivers

3. unity