Valve sucks

First (before I explain how this is relevant to the "video card" thread), guess what graphics card this is rendered on? It's not my pic (I'm just hosting at the moment). Do pay attention to the FPS counter in the lower right.

Another hint....some tweak was done....

Using the Valve CS:Source stress test, this tweak boosted performance on a certain card from 36 FPS to 64 FPS with virtually no image quality loss.

Guess the card yet?

OR the tweak?

Fx 5700?
tweak = textures medium?

answers?

Your 5700, and you switched to OpenGL.

Radeon 9200, because Valve belongs to ATi

heh, can this tweak be performed on other nvidia cards? Not that I'd need it, but it'd be nice.

Originally Posted by bboynitrous

Your 5700, and you switched to OpenGL.

Don't toy with my emotions!!!



........



That's not possible!..... is it?

Oh, it's pretty sad really.

Basically, some guys on Guru3d figured out what Valve did to cripple nVidia cards.

First off, you need 3dAnalyze. I'm assuming everyone knows that you can force HL2 to run in DX9 mode on FX cards, right? Only, you get artifacts in the water and other areas?

Well, that's pretty easy to fix. Just have the 3dAnalyze util report your card as an ATI Radeon instead of a GeForce FX.

*taddah* All the artifacts go away, and you get true DX9 reflections!

Okay, but there IS a performance hit doing that. How to get around that?

Well, the funny thing is that Valve coded Half-Life 2 to use FP24 shaders all the time every time. And it's really not needed. Nope. In fact, FP16 seems to do the trick most the time - as seen in that above pic. FP16 and FP24 are indistinguishable in Half-Life 2 for the most part.

Again, using 3dAnalyze you can test this. It is capable of forcing a card to use only FP16 shaders no matter what is requested. You'll see virtually no image quality difference doing that - just a HUGE performance boost. Why? Well, because while FP16 is all that Half-Life 2 *needs* almost all the time, if they let the GeForce FX cards do THAT, they might have been competitive! So, instead, they forced full precision in every shader op (unneeded), which caused the GF-FX cards to render the DX9 mode in FP32 all the time. With the obvious associated performance hit.

Try it yourself. The link to the article is here. Download 3dAnalyze, and follow these instructions:

Originally Posted by Presi

Open it and follow the numbers:
1. select HL2.exe file in half-life 2 folder
2. select any file inside the folder half-life 2\bin
3. select Steam.exe
than check these options:
- Under the section Pixel and Vertex Shader: FORCE LOW PRECISION PIXEL SHADER
- Under the section Remove stuttering: PERFORMANCE MODE
- on the bottom left: FORCE HOOK.DLL

If you haven't change the file dxsupport.cfg with the method described in the beginnig of this thread, you can obtain the same result typing in the section DIRECTX DEVICE ID'S the ATI Vendor and Device ID, there are just two device though.
....
In the end 3D ANALYZE gives me an error, CREATEPROCESS FAILED, I launch HL2 anyway, the water looked awesome, awesome detail and I noticed a boost in performance too. I think around 20/30% which allowed me to play the WATER HAZARD level with this setting: 1024x768 everything max, water relection to ALL, 2xAA, 4xAnisotropic with a range of fps of 40 and >150.

Amazing, huh?

AND NOW, AN EDIT:
With more data!

Originally Posted by Frosteh

While writing that marathon I decided to try out some of this stuff.

I ran downloaded the program, installed tryed it out and it wouldn't load HL2 itself, however testing showed that in fact it was effecting HL2 when it was loaded and HL2 was ran from steam.

GameSettings:
1024x768
Maximum everything
DX9.0 enforced with fixed bugs (using method of ATI 9800 product ID)


Driver Settings:
Driver Version 61.77
2xQAA (acts like 4xaa with speeds of 2xaa, aa that i have grown to love)
4xAF
High Quality
V-sync: OFF
Trilinear Optimisations: OFF
Ansiotropic Filtering: OFF

http://www.hackerz.tc/hzinternal/tot...-and-32bit.jpg

The picture is around 700k, and is a side by side comparison of the two 1024x768 screenshots, added together with adobe photoshop, saved to jpg with maximum quality (100) and no other alterations made.

The "cl_showfps 1" command was used to display the current average FPS and the screenshots were taken with the in game screenshot capture.

32bit is on the left, 16 bit is on the right, frame rates are roughly 29FPS and 41FPS respectivly, and the performance was a lot better in game with 16bit forced obviously, while this area ran particuarly badly compared to most other areas I considered 30FPS playable, but with 41 FPS I could easily up the resolution one step to 1280 960.

Machine specs for anyone who missed them:

XP3000 @ 11.5x200
1Gb PC3200 Ram @ 200
FX 5900 Ultra Modded to FX 5950 Ultra, further overclocked to 550/975

Let me know if the screenshot method is not accurate enough, if you guys want it done again with other methods it will have to wait untill tomorrow im afraid.

Now, this tweak is GREAT for proving what kind of performance hit Valve kindly provided GeForce FX users....but it's obviously not usable to play the game with. Why?

1) Well, first, 3dAnalyze is simply not stable enough to use as a workaround to play the whole game
2) In the case of some specific shaders (some windows, a few surfaces), there ARE visible artifacts - color banding - as a result of forcing partial precision. Is this a problem? Not really - the whole point of this observation is that Valve should have allowed partial precision MOST of the time, not ALL of the time. GeForceFX cards have a more-than-is-needed full precision mode (which they are stuck with running full-time currently) that would be perfectly suitable for the few times full precision is actually NEEDED in the game.

So, in short, Valve handicapped the GeForceFX cards by 'picking and choosing' which part of the DX spec to follow - they chose not to implement the partial precision hints allowed for in the spec, and which are obviously usable in almost every case in the game, and which would have made the GeForce FX cards competitive!

Ouch, tsk tsk Valve. Anyone going to do some in depth benches / results on this? Brent?

Originally Posted by 6800GTOwned

Don't toy with my emotions!!!



........



That's not possible!..... is it?

lol, I don't think so. I was just saying random crap hoping to be right.

any effect on 6800 series?

lol i called this so long ago

funny even at FP32 all the time, 6800 still kicks this game's ass. But it would be even faster if Valve used good shader programming and put in partial precision calls where appropriate.

hmmm, even as a 6800 gt owner I need some side by side screenshots before I jump on this bandwagon. Maybe I'll have time to test it out, but it is sad if true considering that the 6800 gt still ties with with the x800 pro.

Hey Kyle and Brent are you going to investigate this? I would like a second opinion on this.

Top of the line video cards from both companies are supposed to be neck-to-neck with each other this round of the fight anyways. Good discovery though.

Originally Posted by tranCendenZ

lol i called this so long ago

funny even at FP32 all the time, 6800 still kicks this game's ass. But it would be even faster if Valve used good shader programming and put in partial precision calls where appropriate.

Yeah, that's the real kicker.

All 3dAnalyze can do is force FP16 *all the time*. Although all the screenshots taken to date of it show *no* image quality difference....you gotta wonder.

All that means is that it's a damn shame Valve didn't code this right. Let's assume we DO somewhere find some artifacts by forcing it to FP16 (none found yet, just saying). Let's say as much as 5% of the rendering needs at least 24-bit floating point instructions to actually render properly.

If they'd used _pp hints for the REST, then 95% of the time, the FX cards would be running FP16, and 5% of the time running FP32 (instead of 100%, as they are now). There would be literally *NO* image quality difference in this hypothetical situation - anywhere, at all - and the performance would still be 95% of the boost we are seeing by forcing it to always use FP16.

Valve could have done this, and then the FX cards would be running in DX9 mode perfectly competitively instead of using DX8 mode. I'll grant the FX 5900s would probably still lose to the Radeon 9800 Pros and XTs....but we would be talking about playable framerates still, and identical image quality - rather than the 9800s slaughtering the FX cards we 'see' now.

(Course, it's *entirely* possible Half-Life 2 really doesn't EVER need more than FP16, and forcing it to use that 100% of the time will come up with no artifacts at all. In which case....shame on Valve, seriously!)

the 9700-9800 line and relatives do 24 percision while the FX line does 16 or 32 depending.

so yes.. its official by forcing the program to do 24 and not letting teh card decide its like valve locked in ati and exluded valve. can you say black mail for not bidding the highest?

haah what bs.

fuck valve..and fuck steam.

too bad for the fx owners..

but the 6800 or 6600 owners should be just fine.

Unfortunatly because the compatition is so tight between both vendors we are gonna start seeing more such instances as this i think personally It really is Not on when companys start to disadvantage one card in pref to the other
I just REALLY hope to god that Game Devs Dont adopt this trend more often?

have you used the http://www.ati.com/developer/compressonator.html or similar image quality compare by showing the differences between the images to verify the quality if almost if not as good?