Valve sucks

[mohammedtaha]

if u think about it .. the only reason it works iis f u tell it it's an ati card since nvidia cards will still use 32fps when 24 is needed and maybe HL2 was programmed to recognize different cards and set somethings automatically .. what is probably happening is the extra 8 fps used are probably being dumped 24 - 16 fp ... and probably works if not needed but when it is needed u might see some artifacts at some spots .. and they maybe no visible but if another engine does come out and is gonna be using 24 fp then this hack will not work ...

maybe valve intentionally left it like hoping to send a patch that will fix this problem ... they're probably workin on it .. but impatient people out there are not waiting ...

This problem is nothing compared to my NFSU2 problem which EA can't fuckin fix and so i got rid of the game (game crashed to the desktop 4/5 times) .. and also being a boring ass game .. driving around the city to get from one spot to another just to get ur damm picture taking ... really tiring after the first few times ..

oh well .. enough complaining ... stop crying about it and enjoy the game .. u hacked it and it works ... stop blaming valve and get over it .. u found a fix before they did hooray

 

[el rolio]

Lawd gimme strength...

^ YARDIE!

 

[starhawk]

^ YARDIE!

say what?

 

[el rolio]

hahah nah i saw what you said there and i really cant think of anyone but someone with/who is caribbean heritage that would say that. guess i was wrong tho.

yea, my moms would say that 2-3 times per day on average while i was a teen hahah.

 

[starhawk]

but what does it mean?

 

[psikoticsilver]

Hey, did we even get any new benchmarks or screenshots?
(No way am I actually going to sit here and read 15 pages of flames to find a few benchmarks... you understand, right? )

 

[Kevin Lowe]

if u think about it .. the only reason it works iis f u tell it it's an ati card since nvidia cards will still use 32fps when 24 is needed and maybe HL2 was programmed to recognize different cards and set somethings automatically .. what is probably happening is the extra 8 fps used are probably being dumped 24 - 16 fp ... and probably works if not needed but when it is needed u might see some artifacts at some spots .. and they maybe no visible but if another engine does come out and is gonna be using 24 fp then this hack will not work ...

I can't even tell what you're babbling about.

You can't just dump bits. If you do, you lose part of the number. Since floating point numbers are stored in two parts (mantissa and exponent) you'd be completely changing the value of the number in unpredictable ways - suppose you drop the entire exponent; you just changed the value by 10^(max_value_of_exponent). It's not like it only uses the bits it needs to store the number - even if you're storing an integer that's been cast to floating point, all the bits are still used. Changing precisions just changes the number of bits used to store and pass each component.

 

[rcolbert]

Hey, did we even get any new benchmarks or screenshots?
(No way am I actually going to sit here and read 15 pages of flames to find a few benchmarks... you understand, right? )

There is a good article on Anandtech that compares mainstream cards using DX8/DX9. They don't cover the *controversy* here, but at least give a good idea of some of the performance and IQ differences:

http://www.anandtech.com/video/showdoc.aspx?i=2281

 

[tranCendenZ]

Hey, did we even get any new benchmarks or screenshots?
(No way am I actually going to sit here and read 15 pages of flames to find a few benchmarks... you understand, right? )

You can see here some benchmarks from 2003:
http://www.techreport.com/etc/2003q3/hl2bench/index.x?pg=2

As you can see a mixed precision path (FX path on the chart) is almost as fast as the DX8.1 path in some cases, and in all cases much faster than the full precision path for GeForce FX cards. Valve should have implemented this but failed to.

 

[starhawk]

that sort of settles things...

tho this thread is so d*** confusing from all the spam and flames that i can't say which side is the winner here...

 

[psikoticsilver]

You can see here some benchmarks from 2003:
http://www.techreport.com/etc/2003q3/hl2bench/index.x?pg=2

As you can see a mixed precision path (FX path on the chart) is almost as fast as the DX8.1 path in some cases, and in all cases much faster than the full precision path for GeForce FX cards. Valve should have implemented this but failed to.

Ok jeez, let me be more specific:

Do we have any new benches and/or pictures past page 18 that are UP TO DATE (ie not from 2003), and use the floating point precision "hack" (I don't know, what would you call it)?

I want to see benchmarks where they contrast DX8.1 out-of-the-box and DX9.0 out-of-the-box with DX9.0 with forced 16-bit and/or 3danalyze.

 

[Scali]

The Anandtech article from last year, http://www.anandtech.com/video/showdoc.aspx?i=1863&p=2 gives good info on what the 'mixed mode' entails:

Working closely with NVIDIA (according to Gabe), Valve ended up developing a special codepath for NVIDIA's NV3x architecture that made some tradeoffs in order to improve performance on NVIDIA's FX cards. The tradeoffs, as explained by Gabe, were mainly in using 16-bit precision instead of 32-bit precision for certain floats and defaulting to Pixel Shader 1.4 (DX8.1) shaders instead of newer Pixel Shader 2.0 (DX9) shaders in certain cases. Valve refers to this new NV3x code path as a "mixed mode" of operation, as it is a mixture of full precision (32-bit) and partial precision (16-bit) floats as well as pixel shader 2.0 and 1.4 shader code. There's clearly a visual tradeoff made here, which we will get to shortly, but the tradeoff was necessary in order to improve performance.

Unlike what some people seem to think, this path was not just about running in FP16... It also used DX8.1-level shaders for some things... Apparently in the final version, they decided to run everything in DX8.1.

Anandtech said this about the quality last year:

The sacrifices that you encounter by running either the mixed mode path or the DX8 path are obviously visual. The 5900 Ultra, running in mixed mode, will exhibit some banding effects as a result of a loss in precision (FP16 vs. FP32), but still looks good - just not as good as the full DX9 code path. There is a noticeable difference between this mixed mode and the dx82 mode, as well as the straight DX8 path. For example, you'll notice that shader effects on the water aren't as impressive as they are in the native DX9 path.

Are the visual tradeoffs perceptive? Yes. The native DX9 path clearly looks better than anything else, especially the DX8.0/8.1 modes.

Now, one theory that I have is that Valve managed to improve the image quality of the DX8.0/8.1 modes, and perhaps decided that there is no longer a use for the mixed mode, since DX8.1 will look and perform about the same on an FX card (they mention banding in the mixed mode path, so apparently it didn't look all that great then either).

Another theory that I have is that Valve decided that DX8.1 is the only way to get the FX series to run at comparable framerates to their Radeon counterparts, and Valve decided to drop the mixed mode, because they were afraid that people would say the game wasn't as playable on FX cards as on others (the mixed mode is still quite a bit slower than the DX8.1 mode, in those benchmarks).

 

[Optimus]

I would consider doing a benchmark run if someone can tell me which files are modified so that I can back them up first. I'll investigate further on my own until I get this info.

 

[CleanSlate]

The Anandtech article from last year, http://www.anandtech.com/video/showdoc.aspx?i=1863&p=2 gives good info on what the 'mixed mode' entails:



Unlike what some people seem to think, this path was not just about running in FP16... It also used DX8.1-level shaders for some things... Apparently in the final version, they decided to run everything in DX8.1.

Anandtech said this about the quality last year:



Now, one theory that I have is that Valve managed to improve the image quality of the DX8.0/8.1 modes, and perhaps decided that there is no longer a use for the mixed mode, since DX8.1 will look and perform about the same on an FX card (they mention banding in the mixed mode path, so apparently it didn't look all that great then either).

Another theory that I have is that Valve decided that DX8.1 is the only way to get the FX series to run at comparable framerates to their Radeon counterparts, and Valve decided to drop the mixed mode, because they were afraid that people would say the game wasn't as playable on FX cards as on others (the mixed mode is still quite a bit slower than the DX8.1 mode, in those benchmarks).

That's the key to fixing the entire situation!

Just change your 5600/5700/5800/5200 (5200 might not work well with it) to a 5900 ultra in your settings and it will run in the 5900 ultra's mixed mode setting with fp16 and all the corrections made for it to look great done by valve. That's a good game.

You have your solution.

~Adam

 

[Optimus]

Come to think of it... Do we know that they removed the mixed mode? I wonder if it is simply another "-dxlevel" setting. Perhaps "-dxlevel mm" or "-dxlevel mixed" or "-dxlevel 9m"?
I'll try a couple things later, after work... maybe someone who has a legal copy of the beta (HardOCP or another hardware review site) can check and find out what it was called back then. I'll try a few guesses tonight and get back to you guys.

 

[Scali]

Come to think of it... Do we know that they removed the mixed mode? I wonder if it is simply another "-dxlevel" setting. Perhaps "-dxlevel mm" or "-dxlevel mixed" or "-dxlevel 9m"?

That's a good point really, I had assumed they did, from all the reactions posted here.
But as CleanSlate points out, it was only supposed to run on the 5900Ultra anyway (the Anandtech article I posted above mentions that).
In that case I have another theory... Valve removed the mixed mode because the card it was supposed to support is too rare to be worth the effort (1.95% for '5900 Series' according to Valve survey... and how many of those are Ultras? Less than 1%?).
That is, assuming they did remove it. Would be interesting to see some tests of people with a 5900Ultra running at default settings, and forcing DX8.1. And also trying to use a different card ID, and running that at default settings, and DX8.1. That way we can get a good idea of what kind of path the card is actually running.

Oh, and my suggestion would be -dxlevel 82.
Anandtech mentions DX82 somewhere in the article, could be a typo, could also be the name of the path.

Edit: after reading more through the article, it seems that -dxlevel 82 was the name of the path for 5200/5600 at the time, part DX9, part DX8.1. Not sure how that differs from the 5900U path though.

At any rate, the 5900U got beaten by a R9600Pro (and sometimes Ti4600) when using mixed mode, so I think it's quite likely that Valve decided to go for the DX8.1 path, no 5900U owner will accept it when it is beaten by a card of what... 1/4th of the price?

 

[starhawk]

something like that.

 

[Optimus]

For one game out of many, I would accept it, especially considering the obvious, though conceivably unintentional, bias this particular game has.

Consider this starhawk:

Had the thread been closed when you suggested it, the discussion would never have taken such a productive turn as it has.

 

[Optimus]

Good News! It worked! "-dxlevel 82"

I noticed but one problem in the CS:S Video Stress Test and this was a slight banding in a pattern somewhat like 1/(x^2) + 1/(y^2) = 1. If you can live with that, then it still looks much better than dx81 and I got a framerate increase of 20 fps bringing me to a total of 48.88 fps.

That's an increase of nearly 71% (i.e., [20/28]*100%)

I would still prefer a partial precision dx9 only path, but this certainly satisfies me. It would also be nice to see them finish this path. Of course, they may have and then just were not happy with it's performance.

 

[starhawk]

Good News! It worked! "-dxlevel 82"

I noticed but one problem in the CS:S Video Stress Test and this was a slight banding in a pattern somewhat like 1/(x^2) + 1/(y^2) = 1. If you can live with that, then it still looks much better than dx81 and I got a framerate increase of 20 fps bringing me to a total of 48.88 fps.

That's an increase of nearly 71% (i.e., [20/28]*100%)

I would still prefer a partial precision dx9 only path, but this certainly satisfies me. It would also be nice to see them finish this path. Of course, they may have and then just were not happy with it's performance.

ATTENTION!!! WE HAVE A FIX!!!

now let's throw a party!!

 

[Optimus]

Bad News...

The 8.2 shader path is not finished. It does not support "Reflect All" in water. It does however default to "Reflect World."

It would not surprise me if this is why it was canned. Reflect all probably uses nothing but fp32 due to the obviously complex nature of the shader. Thus, it could not be implemented without a drop to "unplayable" framerates.

If one changes the setting to "Reflect All" HL2 will hang for a minute and then change the hardware dxlevel to 7.0. A tad drastic but that's probably the lowest shader path available.

 

[tranCendenZ]

now, time to start bitching about valve requiring steam for retail cd buyers, and fixing the stuttering bug

 

[starhawk]

Bad News...

The 8.2 shader path is not finished. It does not support "Reflect All" in water. It does however default to "Reflect World."

It would not surprise me if this is why it was canned. Reflect all probably uses nothing but fp32 due to the obviously complex nature of the shader. Thus, it could not be implemented without a drop to "unplayable" framerates.

If one changes the setting to "Reflect All" HL2 will hang for a minute and then change the hardware dxlevel to 7.0. A tad drastic but that's probably the lowest shader path available.

so perhaps it wasn't all a vast conspiracy... just something they couldn't fix.

i will post my email address if flamers so request, just be warned: i will flame you back enough to make your modem smolder.

 

[Optimus]

OK, nothing but "Simple Reflections" works with it. That sux monumentally as even 8.1 at reflect all gives me 19fps at very worst. Granted 9.0 gives me 12fps at very worst so not much improvement there. No comparison was possible for 8.2 due to this water crippling.

 

[R1ckCa1n]

i will post my email address if flamers so request, just be warned: i will flame you back enough to make your modem smolder.

LOL

 

[starhawk]

LOL

well... lemme put it this way:
Poppy didn't want to pay a single shiny penny for my college education cuz he'd just had a $30,000 wedding... it only took ONE email to get those payments flowing. ONE.

don't p*ss me off. it's bad for your health.

 

[Zxcs]

well... lemme put it this way:
Poppy didn't want to pay a single shiny penny for my college education cuz he'd just had a $30,000 wedding... it only took ONE email to get those payments flowing. ONE.

don't p*ss me off. it's bad for your health.

lmao

 

[starhawk]

lmao

if you want to find out for sure... pm me your email address... and we'll see just who's laughing in 2-3 emails.

i'll give you a hint as to the laughter... it ain't gonna be you.

 

[^eMpTy^]

if you want to find out for sure... pm me your email address... and we'll see just who's laughing in 2-3 emails.

i'll give you a hint as to the laughter... it ain't gonna be you.

lol...mighty strong words

 

[starhawk]

lol...mighty strong words

i owe you a few words too i think... since Lord Almighty Kyle sez i can't knock people off my fs/t thread just for incredibly moronic posts anymore.

 

[Moloch]

well... lemme put it this way:
Poppy didn't want to pay a single shiny penny for my college education cuz he'd just had a $30,000 wedding... it only took ONE email to get those payments flowing. ONE.

don't p*ss me off. it's bad for your health.

Look out everyone, its an internet bad boy!

 

[Skystalker]

i owe you a few words too i think... since Lord Almighty Kyle sez i can't knock people off my fs/t thread just for incredibly moronic posts anymore.

My friend... a very good, rather revived discussion just went down hill. What do ya say we direct this back to the thread at hand?

I understand frustration... lets rise above it, yes?

-Skystalker

 

[starhawk]

consider it done. now what were we talking about?

 

[gsboriqua]

The difference is that SM3.0 will be the standard until at least 2006, probably another 2-3 card generations. SM2.0B is a one card, one manufacturer generation shader model. If you were to pick three shader paths, SM3.0, SM2.0, and SM1.x would be the obvious choices. SM2.0A and SM2.0B are both "black sheep," as they are both primarily specific to single card series. Therefore next year a dev who was just coding based on percentages as Rick suggested would never code for SM2.0A/SM2.0B, because the percentage for those shader models is small and will never grow. SM1.x covers a lot of cards, SM2.0 covers a lot of card series including ones that use SM2.0A and SM2.0B, and SM3.0 will cover a lot of cards through the future. SM2.0A/SM2.0B don't, and therefore will fall into the small percentage category. So if you want to use the percentages/money argument, realize you are dooming the X800 to SM2.0 vanilla shader performance, because supporting SM2.0B just for the X800 is not much different than supporting partial precision just for the GeForce FX, if anything its more complex and more costly. A few months after ATI SM3.0 cards hit the streets next year and penetrate the market, X800 owners will be very much in the same boat as GeForce FX owners are now in terms of getting devs to maximize performance for their card, as they both need exclusive coding to get the most performance.

I'm not sure why you spend so much time spreading around so much garbage about a card you have never even used is beyond me.

Anyway you are making some pretty silly assumptions. To think the next gen ATi card won't support SM2.0b is silly. Which means that coding SM2.0b won't just be for x800 owners exclusively, and therefore a wast of time. Gimme a break!

And whats with the guy threatening to spam you with emails until you cry??? lol wtf! You dont want to mess with me! I'll email your ass!!! Look out!

 

[tornadotsunamilife]

i thought all next-gen cards were going to be support sm3.0?

 

[scout007]

Can you tell the difference? Both of the pictures are 1280x1024 and with high everything except for the AA and AF.

zero AA and zero AF

4xAA and 8xAF

 

[peacetilence]

I haven't read this but I haven't played anything related to Hl2 since I've owned EQ2. One thing I will say after trying out HL DM tonight is that it's a jittery unplayable mess to me. FPS is good but players skip 10 feet at a time. It's really crappy.

 

[Advil]

Sorry for diving off the shader topic, but I think everyone has beat to death the lack of visual difference between SM2 and 3.

Boy those screenshots are a poster child example of why 4xAA is good in HL2. Not only does the bottom screenshot look a lot better, but those chain link fences that they are sooooo fond of are a love/hate thing. They suck with no AA, but with 4x or better, the shimmering as you move past them completely goes away and suddenly the effect is believable.

The engine could stand to push a few more polys but overall it's a 9.9 out of 10. Looks, physics and the feel of the gameplay are all spot on.

 

[gsboriqua]

i thought all next-gen cards were going to be support sm3.0?

They will, but that doesn't mean that all of a sudden ATi's new cards will no longer be able to run SM 2.0b

 

[fallguy]

Can you tell the difference? Both of the pictures are 1280x1024 and with high everything except for the AA and AF.

zero AA and zero AF
http://ucsu.colorado.edu/~zhongs/d1_canals_070002.jpg



4xAA and 8xAF
http://ucsu.colorado.edu/~zhongs/d1_canals_070001.jpg

Yeah, easily. There is a pretty large difference in quality. The bottom one looks a lot better. Still more jaggies than I would have thought with 4x though.

The water looks a lot better too, you sure you had the same setting for the water for both?