HUMUS ATI/Doom3 tweak explored

[rbV5]

Originally posted by: Rollo

Originally posted by: rbV5

To me it would be worth it to buy a 6800NU just to play this game. If you love Doom3, and hate nVidia, you could buy a 6800NU for $283, play the game, sell it for $250 and be out $30-$40

To me that would be silly since it plays and looks well on less powerful cards. Unless things change in later levels,....my 6800 offers no real advantage over my 9700 even with a little higher resolution and quality settings. The darkness of the game coupled with the atmosphere and action, I'm not seeing a whole lot of difference in the game between either of my setups, there's just not a dramatic difference between medium and high quality settings, or 800x600 and 1024x768, at least not enough to justify buyig a card to play at slightly higher settings IMO.

You could really say that about ANY video card comparison though when you're talking about upgrading. It's not $300 either- a guy with a 9700Pro has a card worth $140- half the cost of a NU. If you consider the 9700Pro on average 10% slower than a 9800XT, and a 6800NU on average 20% faster than a 9800XT, you're talking about a 30% increase in performance for $140 out of pocket. (and Far Cry or other games thrown in) When you add in the Doom 3 factor being able to run it at a decent setting like 12X10, 2X8X, to me that seems like an excellent bargain. (and that is a LOT higher setting than any 9700Pro can run it at)

I agree with most of what you say, but I just wouldn't think it worth it "just" for Doom3 if you already have a 9700/5800 or better card. IMO, the 9700 playability is better than I expected, and the difference in the quality settings is not as dramatic as I expected (this game looks damn good at lower settings than I expected) To be fair though, I've been playing mostly with my 9700 rig and I've only had the game a few days now. I haven't even tried the Humus tweak, and the game is plenty playable and looks good.

 

[BenSkywalker]

There is no explanation as to why it would not be mathematically equivalent unless the gpu was botching an equation; the tweak doesn't add a third-party equation nor does it remove any equations.

It does remove equations, got to B3D's main page and check out the snippet they have with Carmack. They replace the properly equated look up table with a simpler equation.

 

[BFG10K]

and I have to say that gameplay and enjoyment is every bit as good with my AIW 9700pro rig as my 6800 rig

I have to disagree with this comment; my 6800U absolutely blows away the 9700 Pro in D3. It's like playing a completely different game.

 

[nRollo]

Originally posted by: BFG10K

and I have to say that gameplay and enjoyment is every bit as good with my AIW 9700pro rig as my 6800 rig

I have to disagree with this comment; my 6800U absolutely blows away the 9700 Pro in D3. It's like playing a completely different game.

What what what?!?!

BFG! You bought a 6800U instead of the X800XT PE you were planning on?!

I know what's at the top of my shopping list now.....

 

[BFG10K]

My main reason was ATi's driver issues, although having the fastest card for Doom III is a nice bonus.

 

[nRollo]

Originally posted by: BFG10K
My main reason was ATi's driver issues, although having the fastest card for Doom III is a nice bonus.

Oh well. I suppose we'll be back to you telling me stuff like "If you would have bought the Ultra model instead of the GT, you'd know what I'm talking about. You have to run those butt ugly 12X10 Doom3 settings!" but congrats on the card. You made the right choice, IMO.

 

[rbV5]

have to disagree with this comment; my 6800U absolutely blows away the 9700 Pro in D3. It's like playing a completely different game.

I would imagine an Ultra (or GT) would make a difference, but I have a 6800NU which dies when you up the resolution or AA/AF. I think the faster/more ram makes a big difference in this game once you crank everything up. If I overclocked the 6800NU, it would probably make a difference, or a CPU more powerful than a mobile barton @ 2.3 GHz.

Maybe I'm having driver issues or something not set right in the drivers with my 6800nu, but 12x10 high quality is pushing it for smooth gameplay.

 

[gururu]

Originally posted by: BFG10K
I have to disagree with this comment; my 6800U absolutely blows away the 9700 Pro in D3. It's like playing a completely different game.

oh come on, isn't that a little much? we already know the quality settings only sharpen the game up a little. the sounds don't change. the monsters aren't scarier or smarter. the only thing left is higher resolution. and never before has resolution made a poor game into a good game (for me).

edit: not that D3 is a poor game (I love it), but other games with no gameplay.

 

[gururu]

Originally posted by: BenSkywalker

There is no explanation as to why it would not be mathematically equivalent unless the gpu was botching an equation; the tweak doesn't add a third-party equation nor does it remove any equations.

It does remove equations, got to B3D's main page and check out the snippet they have with Carmack. They replace the properly equated look up table with a simpler equation.

I thought the tweak replaced an instruction with an equation. An equation already embedded in the instruction set. The lookup is faster in some cases (nVidia), the equation faster in others (ATI).

 

[LocutusX]

Originally posted by: rbV5
or a CPU more powerful than a mobile barton @ 2.3 GHz.

Mobile Bartons @ 2.4ghz are only about 10% slower than A64 3000+ @ stock, aren't they (in games, at least)? I'd figure they should be enough for D3 unless Carmack had optimized the engine for A64.

 

[rbV5]

Originally posted by: LocutusX

Originally posted by: rbV5
or a CPU more powerful than a mobile barton @ 2.3 GHz.

Mobile Bartons @ 2.4ghz are only about 10% slower than A64 3000+ @ stock, aren't they (in games, at least)? I'd figure they should be enough for D3 unless Carmack had optimized the engine for A64.

Plenty fast for Doom3, just not fast enough to run at 16x12 or really 12x10 with 6800NU on high detail.

 

[BenSkywalker]

I thought the tweak replaced an instruction with an equation. An equation already embedded in the instruction set. The lookup is faster in some cases (nVidia), the equation faster in others (ATI).

Hmmm, Ok- the lookup table is kind of like a cheat sheet- giving you the answers instead of having to do out the maths. To get the answers you need to hit memory which chews up some bandwidth. This is done as the equations to get those results would be too intensive to handle properly.

The equation in this particular example(Humus's) isn't the same as the one that was precalculated for the lookup table. It is a simpler equation that does not give the same results, it's just close enough that most people who think ATi has 'good' texture filtering anyway are certainly not going to notice the imperfections the overwhelming majority of the time. The approximation is "close enough" that people are happy with it.

 

[Drayvn]

From what i read, its a math calculations, and they also said that ATi are better at math calculations than nVidia also, thats why if u do this to a 6800 its detrimental...

Here is something about it from Elite Bastards, who tried it out...

"
As we can see, Humus' tweak has a much greater effect when using ATIs own control panel anisotropic filtering than when in-game filtering is used. The explanation for this is most likely quite simple - In the normal part of the game code altered by Humus, a texture look-up table was used to generate values. Using in-game AF, this is obviously not filtered as it is both unnecessary and computationally expensive. However, when AF is forced by the driver, it applies its filtering to everything, including the aforementioned look-up table, thus contributing to the large performance hits we've seen using this method.

Humus' tweak removes the look-up table code to replace it with code to calculate maths values instead - This offers a small performance improvement when in-game AF is used thanks to the X800 Pros speed at handling such operations, but when AF is forced at driver-level, the performance increase is vastly greater as the look-up table is no longer available to be processed by the anisotropic filtering mechanism, thus saving the amount of time normally required to be spent on filtering it."

You can find the review here... http://www.elitebastards.com/page.php?pageid=6201&head=1&comments=1

 

[BFG10K]

Oh well. I suppose we'll be back to you telling me stuff like "If you would have bought the Ultra model instead of the GT, you'd know what I'm talking about. You have to run those butt ugly 12X10 Doom3 settings!" but congrats on the card. You made the right choice, IMO.

The GT is a nice card for its price and in fact I would have gotten one were it not for the fact that the new shipment was already sold out even before it hit my area. They had one more 6800U so I took the plunge. And I'm glad I did too - it's not that loud and the dual slot-cooling isn't a problem at all. Also I'm running it fine off a 400W PSU.

but I have a 6800NU which dies when you up the resolution or AA/AF

Yeah the 6800 is quite lacking sometimes. When it comes to memory bandwidth I think it has even less than the 5950.

we already know the quality settings only sharpen the game up a little. the sounds don't change.

The difference between 800x600@50 FPS and 1600x1200@65 FPS is simply astonishing.

 

[ronnn]

I think this tweak is interesting just in the fact it works. For the troll team here, just to flog Nvidia products - go make benchmarking threads or play with your drivers.

 

[nRollo]

Originally posted by: ronnn
I think this tweak is interesting just in the fact it works. For the troll team here, just to flog Nvidia products - go make benchmarking threads or play with your drivers.

Of course, it wasn't trolling to reference Shady Days 1,234,987 times as evidence of ATIs superiority, even though it's a year later and no HL2 in sight, right Ronnn?

Or to point at "Wallet Raider, Angel of Games Nobody Bought", before that?

Yeah, there wouldn't be any point in discussing an actual game you can buy that just about everyone in the free world is playing, that will be the basis of many huge licenses.

That would be "trolling". We have to wait for HL2 to come out, see if it's significantly better on ATI, and if it is, we can post about that 1,234,987 times as evidence of ATIs superiority.

Only then will we not be trolling.

:roll:

 

[gururu]

Originally posted by: BenSkywalker

I thought the tweak replaced an instruction with an equation. An equation already embedded in the instruction set. The lookup is faster in some cases (nVidia), the equation faster in others (ATI).

Hmmm, Ok- the lookup table is kind of like a cheat sheet- giving you the answers instead of having to do out the maths. To get the answers you need to hit memory which chews up some bandwidth. This is done as the equations to get those results would be too intensive to handle properly.

The equation in this particular example(Humus's) isn't the same as the one that was precalculated for the lookup table. It is a simpler equation that does not give the same results, it's just close enough that most people who think ATi has 'good' texture filtering anyway are certainly not going to notice the imperfections the overwhelming majority of the time. The approximation is "close enough" that people are happy with it.

Thanks Benskywalker, that was a clear explanation. The article didn't really talk about the precise differences the tweak employed. If Humus added his own simpler equation, does that decrease the possibility of Id including it in a patch?

 

[Acanthus]

Originally posted by: BFG10K
My main reason was ATi's driver issues, although having the fastest card for Doom III is a nice bonus.

Didnt you specifically argue with me about how NVIDIA has just as many issues as ATi does? :roll:

 

[Drayvn]

Originally posted by: gururu

Originally posted by: BenSkywalker

I thought the tweak replaced an instruction with an equation. An equation already embedded in the instruction set. The lookup is faster in some cases (nVidia), the equation faster in others (ATI).

Hmmm, Ok- the lookup table is kind of like a cheat sheet- giving you the answers instead of having to do out the maths. To get the answers you need to hit memory which chews up some bandwidth. This is done as the equations to get those results would be too intensive to handle properly.

The equation in this particular example(Humus's) isn't the same as the one that was precalculated for the lookup table. It is a simpler equation that does not give the same results, it's just close enough that most people who think ATi has 'good' texture filtering anyway are certainly not going to notice the imperfections the overwhelming majority of the time. The approximation is "close enough" that people are happy with it.

Thanks Benskywalker, that was a clear explanation. The article didn't really talk about the precise differences the tweak employed. If Humus added his own simpler equation, does that decrease the possibility of Id including it in a patch?

Gururu, i already answered his statement, ill post it again here, its from Elite Bastards, and i think it tells u what the tweak is about...

"As we can see, Humus' tweak has a much greater effect when using ATIs own control panel anisotropic filtering than when in-game filtering is used. The explanation for this is most likely quite simple - In the normal part of the game code altered by Humus, a texture look-up table was used to generate values. Using in-game AF, this is obviously not filtered as it is both unnecessary and computationally expensive. However, when AF is forced by the driver, it applies its filtering to everything, including the aforementioned look-up table, thus contributing to the large performance hits we've seen using this method.

Humus' tweak removes the look-up table code to replace it with code to calculate maths values instead - This offers a small performance improvement when in-game AF is used thanks to the X800 Pros speed at handling such operations, but when AF is forced at driver-level, the performance increase is vastly greater as the look-up table is no longer available to be processed by the anisotropic filtering mechanism, thus saving the amount of time normally required to be spent on filtering it."

 

[otispunkmeyer]

im playin @ 800x600 med quality, on my 9500pro (325/300) and tbh its not the slide show i predicted infact most of the time im over 30fps and i like it.....plays smooth, sound options bit limited no eax? using the humus tweak an although tyhe flash light seems bit faitner....the gamply feels some how more smooth

gettin a 6800GT probably think ive bout decided on it so 1280x1024 on high will be winging its way to me soon.

 

[Creig]

You go, NvRollo!

 

[nRollo]

Originally posted by: Creig
You go, NvRollo!

Only if you'll favor us with more "news" like:

Will the tide shift to the X800 series when HL2 is released? Probably. We'll see when (if) Valve ever gets it out the door.

Time will tell, eh, wise one? LOL

Nvidia had the spotlight all the way from the TNT series to the GF4 line. Then ATI stole the crown with the 9500 to 9800 lineup and now the two are battling for supremacy between the 6800/X800.

So wait a minute- you're saying nVidia used to be on top, then ATI, now they're evenly matched? You must be a professor or something Creig- I couldn't have figured that out! LOL

Remember, competition is a good thing.

Where did you get that? A fortune cookie? LOL

I try to post useful benches and give my opinions, Creig, but how do you expect me to live up to your genious when you post stuff like:

It's a good card, decent price, nice set of features, is (somewhat) available and is the card of choice to play the latest game with (Doom III). It's only logical that it would be praised by the masses.

Why don't you cut those less BRILLIANT than you are some slack buddy?

 

[ronnn]

Originally posted by: Rollo

Originally posted by: ronnn
I think this tweak is interesting just in the fact it works. For the troll team here, just to flog Nvidia products - go make benchmarking threads or play with your drivers.

Of course, it wasn't trolling to reference Shady Days 1,234,987 times as evidence of ATIs superiority, even though it's a year later and no HL2 in sight, right Ronnn?

Or to point at "Wallet Raider, Angel of Games Nobody Bought", before that?

Yeah, there wouldn't be any point in discussing an actual game you can buy that just about everyone in the free world is playing, that will be the basis of many huge licenses.

That would be "trolling". We have to wait for HL2 to come out, see if it's significantly better on ATI, and if it is, we can post about that 1,234,987 times as evidence of ATIs superiority.

Only then will we not be trolling.

:roll:

Did I crap on Nvidia threads referencing shady days? Go troll in a benchmarking thread.

 

[rbV5]

When it comes to memory bandwidth I think it has even less than the 5950.

I believe your're correct, I read on NVnews you have to overclock the memory to 850MHz to equal 5950 memory bandwidth.

The difference between 800x600@50 FPS and 1600x1200@65 FPS is simply astonishing

Thats probably why I'll be stepping up to a GT in my NV rig.

 

[nRollo]

Did I crap on Nvidia threads referencing shady days? Go troll in a benchmarking thread.

Oh no.

Ronnn said I'm a troll. I can never come back here, as Ronnn is the voice of the board.

Please oh please Ronnn. I won't troll anymore! I'll pimp the Humpus tweaks and their pathetic 2fps increases! I'll say "DX9b is good enough!" I'll say "Who needs that UltraShadow or video encoding!" I'll say "who needs linux?"

For the love of God Ronnn, I have a family. Don't say I'm a troll.









LOL :roll: