17 GPU comparison in Steam using Serious Sam 3

[RussianNeuroMancer]

btw, there is new AMD forum for such issues: http://devgurus.amd.com/groups/steam-linux.

Just in case, updated link: http://devgurus.amd.com/community/steam-linux.

[Kano]

I played the whole game and i think level 6 is the best to see if a gfx driver sucks or not as the framerate drops at the beginning already. Using an indoor level like 2 is completely stupid as there the framerate is not below 25-30 fps even with slower cards.

[blackout23]

What level did you play and for how long? I have played the beginning of "Unearthing the Sun" for 300 seconds multiple times the same way and average framerate was between 47-53 doing the same stuff everytime and using the same weapons. It sounds a bit odd that your runs always just differed by 1 FPS around the average.

[RussianNeuroMancer]

Generally speakin, AMD have serious issues with their drivers both in Windows and Linux. Once in a while they fix & patch things and then they release CAPs to improve performance over some hot game titles (such as Skyrim, Starcraft 2 etc). That has been told, it's quite possible that AMD Linux drivers in conjunction with Steam beta and its OpenGL-to-DirectX porting system, to behave strangely. Unfortunately, I have already shipped my 6970 back to AMD, thus I can't bench in Crossfire any more.

btw, there is new AMD forum for such issues: http://devgurus.amd.com/groups/steam-linux.

Kano, maybe raise CF topic on this AMD forum would be more effective?

[sonnet]

Of course and my results are fake. Oh yes they are! It's way more logical to score 3 times worse score with two cards, rather using just one. After all, it is no doubt that 1 is greater than 2. Right ? Right! So Crossfire technology is invented to drop your performance in Linux and so far AMD just don't give a sh1t -- no matter whether Steam is going to release soon enough. Right? Right

Game over

In your benchmarks, Kepler cards seems to perform too well compared to Fermi's generation cards.

For instance GTX 580 perform 10% better than GTX 660 in Windows.

http://www.techpowerup.com/reviews/Club_3D/HD_7870_jokerCard_Tahiti_LE/28.html

On your tests though, the GTX 660 is 13% faster than the GTX580.
And we are talking about the same drivers. And normally it should be logical to think that the gtx 580 had better optimizations since the cards are older.

[Kano]

I did SS3 benchmarks on Win as well, my CPU is i7-3770S, i don't think it is needed to OC it, just because you think my gfx card should be faster. It IS faster when i use DX rendering, then you would see a 5x (CF gives an impressive 5% boost) at the beginning of level 6. Indoor benchmarks are completely useless - there the speed is much higher. But why on earth do you write those benchmarks in an article about UBUNTU?! I even confirmed that when i use OpenGL renderer in Win i get basically the same speed as on Linux with the same CF speed drop. But did not even use the OpenGL renderer, what do you want to show use? The DX performance you never get with OpenGL and that game or what? Also why should your gfx cards scale by 87% and mine only 5%, forget your benchmarks. Set "gfx_iAPI = 1" and restart the game, then benchmark again. Also I did Heaven benchmarks on Win+Linux, where are yours - especially with CF, forget all other cards.

[Dr.Paneas]

noticing the last few post, I guess you're right about this being fake.
PS: thanks for your ss, I was wondering about a possible CF configuration, I guess there's not doubt now that it would be a very bad idea.

Of course and my results are fake. Oh yes they are! It's way more logical to score 3 times worse score with two cards, rather using just one. After all, it is no doubt that 1 is greater than 2. Right ? Right! So Crossfire technology is invented to drop your performance in Linux and so far AMD just don't give a sh1t -- no matter whether Steam is going to release soon enough. Right? Right

and yes this is not me playing with Dry Ice.

http://www.youtube.com/watch?v=xy_OGAjLlH8

and yes, this is not me playing with phase change.


and yes in a matter of fact, I am a alien-bot called Benchmatron, son of Ubuntron, Debian's heir, who spent his last 5 years as a hardware reviewer in many IT websites (Insomnia.gr, Outofspecs.gr, Guru3D.com, VR-Zone.com) just waiting for that moment. To fake all Linux community. Yeah!!! I confess, they have cookies in the dark side, I couldn't resist.

Sorry I can't discuss with you guys especially when there is no doubt that my benchmarks are fake. Have a nice day and don't ever visit UbuntuXtreme. This is a bad, satanic place where ... hmmm it seems that there are 3 page saying true (not fake) analysis with screenshots and animated gifs but the last page is fake... 1 page to fake them all. 1 page to cheat them and into fake numbers bind them.

Game over

[Dr.Paneas]

You should not believe ANY number, he shows incredible CF performance increase, even when this would be win, but there is not even a 5% increase when you use DX renderer, the OpenGL decrease with CF is on win as well. I can also create a random number table, this one is 100% FAKE.

dun DUN DUN


germans vs greeks go:

No no no. It's not germans vs greeks. From my point of view, it's germans + greeks. We are in open source community, working together in order to find out what's happening here. So if you want to start a flame, I am not going to join you. I don't care if you are developer of Kannotix, but you should learn some proper manners before accuse someone in public. Especially when you you are not 100% sure about what's going on.

From my experience Serious Sam 3 gets good crossfire scaling indoors, but outdoors (which is like most of the game) is still crappy. This is a confirmed critical bug by CroTeam. Bare in mind that my tests are based in Level 2 (The Spider's Nest which is mostly an indoor's level).


You score ~30 FPS with a single GPU, this rings me a bell that 99% you experience CPU bottleneck !! Please share your CPU model and what's your SS3 settings and resolution ?

Judging from your screenshots, CrossFireX is a bad feature for your system playing SS3. Two thinks come in my mind:

1) This is most likely due to your CPU bottlenecking your performance. Meaning that your CPU is not going to allow as much performance as your GPUs allow. That's why I had to overclocked my Core i5 2500K up to 4.5GHz. Crossfire-wise, CPU bottleneck can decrease performance in some cases when it doesn't have the CPU muscle to feed it. Unigine heaven has almost no CPU stress, so a slow CPU doesn't show up in there and that's why you saw such a performance boost in DirectX11.

2) Possibly poor performance from a current driver. Try running the game in Windows7, but don't forget to install the latest CAP profiles. Without those, crossfire doesn't perform as well.

Generally speakin, AMD have serious issues with their drivers both in Windows and Linux. Once in a while they fix & patch things and then they release CAPs to improve performance over some hot game titles (such as Skyrim, Starcraft 2 etc). That has been told, it's quite possible that AMD Linux drivers in conjunction with Steam beta and its OpenGL-to-DirectX porting system, to behave strangely. Unfortunately, I have already shipped my 6970 back to AMD, thus I can't bench in Crossfire any more. However, any time soon, I will try to find a couple HD 7000 series in order to play with Serious Sam 3 and Team Fortress,

Just for you information, phoronix has published some Crossfire benchmarks back in 2008. In QuakeWars, there is a huge scalling and crossfire works great. But in Quake 4, there is no Crossfire boost. Now supposing that this article was written back in 2008, we have 4 years of improving drivers. Do you really believe that AMD guys are so dumb that they managed to ruin Crossfire performance that bad as you imply in your in-game screenshots ? I mean, you don't even get equal performance, but worse. If I were you I would probably do some more testing with proper hardware and then fire accuses. Anyway, no hard feelings.

my 2 cents

PS: I agree with you for Unigine Heaven, no Crossfire profile for OpenGL both in WIndows and Linux.

[sonnet]

noticing the last few post, I guess you're right about this being fake.
PS: thanks for your ss, I was wondering about a possible CF configuration, I guess there's not doubt now that it would be a very bad idea.

[Kano]

You should not believe ANY number, he shows incredible CF performance increase, even when this would be win, but there is not even a 5% increase when you use DX renderer, the OpenGL decrease with CF is on win as well. I can also create a random number table, this one is 100% FAKE.