[Hellbinder]

I am actually a little upset at this release.... more like upset in general..

As with most CATALYST™ releases, the performance has increased in numerous situations. Here are some examples observed in CATALYST™ version 3.5.
Improvements in the ATI OpenGL drivers have experienced performance gains such as:
• GL Excess performance improves 12-14% on the RADEON™ 9700 PRO and 9800 PRO. Modest benefits are noticed on our other graphic products as well
• Unigraphics performance (ViewPerf Ugs-03) improves 27% on the RADEON™ 9800 PRO, 52% on the RADEON™ 9700 PRO, and 79% on the RADEON™ 9500 PRO
• Data Explorer performance (ViewPerf DV-08 ) improves 9% on RADEON™ 9800 PRO, and the RADEON™9700 PRO. The RADEON™ 9500 PRO improves by 14.5%

How can this not be *cheating* ???? ( otherwise known as *optimizations*)

Only improvemetns in these specific *BENCHMARKS*. Means they are application detecting and tweaking for a higher Benchmark score. With little effect on real games.

I am honestly a little Irritated right now about the whole thing. I am sick to the core of all companies and And review sites emphasizing Prefabed Benchmarks. I am begining to find myself in some agreement with Kyle except for different motivations.

Screw all this 3dmark, Shadermark, Glexcess, Villagemark, Specview etc etc etc junk. Its a giant Meaningless pissing contest that has ZERO effect on the real world of gaming.

[Saem]

Calm down.

YOu have no evidence that this is due to benchmark specific optimisations. They could be general optimisations that happened to have a dramatic effect in these benchmarks.

[Hellbinder]

ok...

So tweaking for benchmark improvements is *ok* then. fixes that only affect Glexcess and Specview. Yet have no affect on actual games..

Tell me.. Given that they are 45-80% increases in these specific benchmarks yet dont affect anything else.. how is it possible they are not application specific?

Look, I just want to know. Either this kind of thing is *Ok* or its not. If its *OK* then unwinder needs to back the truck up with his utility that negates all application specific tweaks. Becuae he is calling it Cheating.

[dbeard]

Another interpretation could be that it is across the board but the tools available to demonstrate a reproducible increase to the public are the commonly known "prefabed" benchmarks.

Remember that they make these statements knowing that they could be sued for fraud if the results are somehow ambigously generated (i.e. using some game but having to do it manually since the game does not support a benchmarking function) that can be construed to represent something they may not have intended.

I'll bet that anyone can reproduce these results given a similar system.

OTOH, for VS2.0 and PS2.0, only the "prefabed" benchmarks are available to show the performance improvements at all.

cya,

[digitalwanderer]

Let's not hang 'em until the jury is in, I haven't even had a chance to install these yet me ownself.

(But I gotta agree that I'm a bit nervous by how specific they said they were removing all the optimizations that were "found" in 3dm2k3, kind of implying that they may just have left in some unfound ones. )

[Mark]

OpenGL performance is up a bit all around... why they didn't just say that I have no idea. Marketing crap I would assume, but I sure hope they've learned their lesson and aren't doing any app detection

[demalion]

Simplification leads to lots of confusion. I'm going to address this by unsimplifying your commentary (no jokes from the peanut gallery about confusing wording, please :P) and offering my related opinions.

Originally Posted by Hellbinder

ok...

So tweaking for benchmark improvements is *ok* then.

Depends on the meaning of that phrase...there are several distinct meanings possible.

You could mean improvements that only manifest for a specific application, because it is the only thing that is doing things a certain way. This tends towards being completely valid, determined by the equivalency of output and how special case the optimization is.

You could mean improvements that only manifest for a specific application and depend on characteristics that only manifest while actually in the act of benchmarking. This tends towards being pure deception, regardless of quality equivalency.

Note what this specificies, and what it doesn't. There are other distinct issues beyond this, however, though you aren't recognizing that in your posts.

One example: Application detection is a method of activating optimizations of the type above, and I have many reasons to consider this a bad thing when the application being detected is purely for benchmarking. If the application isn't only a benchmark, I think evaluations such as those above are the determining factors.

...

The release note blurb you quoted does not indicate this is application detection...believing it indicates this, or that it doesn't for that matter, is an assumption at the moment.

My analysis:

The wording presented tends to convey that it is a general optimization, but for techniques used prominently in those benchmarks. In case you forgot in the midst of all the FUD nVidia is putting out about benchmarks (with help from people such as Kyle No, that's not "hostile" Russ, that's an observation), that's the way IHVs are supposed to use benchmarks. ATI has used benchmarks in this exact way before ("high poly bug") in a way that is established as beneficial to consumers.

However, the amount of the increases tend to provide a reason to doubt that. That is purely circumstantial evidence right now (as opposed to evidence that is circumstantial as well as being substantiated by other factors, which I feel is a distinction that the aforementioned FUD tries to blur). This could change quickly, shifting likelihood one way or the other, by Unwinder's work and/or some other analysis.

fixes that only affect Glexcess and Specview. Yet have no affect on actual games..

That is an assumption. Lent credence by your decision for suspicion (which I consider valid) and the large percentage figures quoted...but that's all that lends them credence (again, right now...that's what competent reviewers and technical minded enthusiasts like ourselves are for ). What they actually said was "Here are some examples observed", not "Here are the only things we achieved".

Tell me.. Given that they are 45-80% increases in these specific benchmarks yet dont affect anything else.. how is it possible they are not application specific?

This is the problem with your post. You are blindly assuming the worst. It is the "blindly" that is the problem...just assuming the worst (not blindly) would recognize that something else is possible, and be more focused on investigating instead of assuming application specific cheats.

Look, I just want to know.

Just wanting to know whether application specific cheats are going on or not seems fine to me, and even working from the assumption that the improvements are useless outside of these benchmarks seems reasonable to me, as long as you recognize that further testing and reasoning are required to be conclusive.

Either this kind of thing is *Ok* or its not.

Here is the problem with your post...you're simplifying things immensely, and missing a lot of very important details. While you might not be wrong, in this specific case, with your assumptions, note that it is easily possible for you to be, and that in atleast the specific case of the "high poly bug" your simplification seems to be established as clearly unsuitable.

f its *OK* then unwinder needs to back the truck up with his utility that negates all application specific tweaks.

Actually, at the moment I sort of universally support defeating application specific detections...if the application is a benchmark. I do have a problem that he made no distinction between benchmarks and games, and that the digit-life article sounds like it will be making some similar simplifications with a huge potential to mislead and distort based on it. I really hope we get more details and access to the exposure tools for discussion of the issues here.

Becuae he is calling it Cheating.

Well, it might be, but it might not be too. That's the problem with simplifications.

[andypski]

Originally Posted by Hellbinder

Improvements in the ATI OpenGL drivers have experienced performance gains such as:

Please note the use of the words 'such as'. This means these are examples of the gains achieved.

How can this not be *cheating* ???? ( otherwise known as *optimizations*)

Very easily. Maybe you'd prefer it if we ran every OpenGL application out there and collated all the results to see what the performance increases in them are, and then listed them for your edification, but this would be a lot of work for us. General optimisation will also result in gains that can be demonstrated with specific applications.

Only improvemetns in these specific *BENCHMARKS*. Means they are application detecting and tweaking for a higher Benchmark score. With little effect on real games.

Nobody said that only these benchmarks were affected, except for you.

I am honestly a little Irritated right now about the whole thing. I am sick to the core of all companies and And review sites emphasizing Prefabed Benchmarks. I am begining to find myself in some agreement with Kyle except for different motivations.

Please think about this. We have to have some way of quantifying our performance increases and improvements to our drivers. To do this we run benchmarks (amazingly enough). We then sometimes report these figures as part of the information with our driver releases. Who'd have thought it...

[Saem]

Hellbinder, you're flying off the handle.

Again, those were examples, not the ONLY improvement(s). COMPREHEND WHAT THEY STATED!

Furthermore, IIRC, those benchmarks are VERY specific, as in the majority of their performance is dependent upon a small sub-system. For example, the sub-systems that are heavily employed here might only be employed 5% or less for other games. This means, you'll see very small increases for games that don't rely on this sub-system that much but will see very dramatic gains in things that are basically totally dependent on it.

It's like having an application which is totally CPU bound. If you put in more RAM a better video card, the increase will be slim to none, however if you improve the CPU sub-system of your machine it'll run MUCH faster.

[Snyder]

Hellbinders emotional reaction to the alledgedly application-specific speed gains advertised in the Cat 3.5 readme is just all too characteristic for the consequence that the cheating debacle caused, which is not too different to the Counterstrike cheating problems:
If you are a good player, chances are you will be accused of cheating, even if you really have e.g. a nightmarish good accuracy and use no cheats whatsoever. This will again happen with ATi in the future, and it will also happen with nVidia. We can only hope that it teaches them to learn their lesson.

[Doomtrooper]

Hellbinder seems to have fallen off the fence, for the past 5 years all the IHVs post synthetic improvements but latley everyone appears to be over evaluating these comments.

I'm running them and all my games were running fine before, but it appears the Nforce 2 bump mapping issue has been rectified which is good.

Fixes:

Fixed in this driver:
The system is not responding when playing Quake II with the OpenGL option set to default, and the video mode set to 1024x768 is now resolved
Setting the display resolution to 1027x768 and running the Unreal Tournament 2003 benchmark with Anti-Aliasing enabled, no longer results in the system rebooting
The system rebooting when attempting to start a new game of Ballistic Forces is now resolved
Enabling Direct3D as the rendering device and setting the display resolution to 1600x1200 in the game Descent 3 no longer results in the game failing to play
The shadow flickering beneath the car when choosing Quick Race in Nascar Racing 2003 is now resolved
Setting the display resolution to 1024x768 32bpp and enabling OpenGL no longer results in the ground being discolored in the game IL2 Forgotten Battles
Connecting a secondary display and enabling extended desktop, followed by running Soldier of Fortune 2 no longer results in a flickering display
The game Unreal Tournament 2003 no longer hangs when loading the CTF-December level
The game Unreal Tournament 2003 no longer displays corruption when Anti-Alasing is enabled
Setting the display options to 1600x1200 32bpp in the Hot Wheels Velocity X game no longer results in the menu screen not being displayed
Display corruption is no longer noticed when playing the game Earth 2150 and the game resolution is set to 1024x768
Playing the game Nascar Racing 2003 in OpenGL mode no longer results in the display going black
Exiting the game Age of Empire II no longer results in display not being restored properly
Playing the game Serious Sam 2 with the display resolution set to 1024x768 32bpp no longer results in the wrong OpenGL renderer being listed when setting the graphics to OpenGL
The game A Tale in the Desert no longer pauses during game play and no longer causes the system to randomly stop responding
Enabling Pixel Shaders in the game Asheron's Call 2 no longer results in display corruption appearing as the game is played
Playing the game Project IGI2: Covert Strike with a RADEON™ 9700 series card installed in the system no longer results in the ground terrain disappearing and re-appearing as the character is moved.
The system no longer stops responding when running 3ds max 2001
The 3DMark2003 shader optimizations found in previous CATALYST™ releases have been removed
Increased CPU performance is now noticed when enabling the use accelerated post processing in the DivX player
Running Maya or SoftImage no longer results in the overly front and backbuffers not being cleared at allocation time
Dragging the WinDVD player to the extended desktop when playing a DVD no longer results in a system hang
Running certain OpenGL application under Windows XP with a RADEON™ 7200 series card installed no longer results in display corruption
Running the Windows Movie Maker program no longer results in the display corruption appearing on the preview screen
Running the ATI Tranquility screen saver found at http://www.driverheaven.net, on an ATI RADEON™ 9000 series card, no longer results in the system not responding under Windows XP
When installing the EAZYLOOK™ TV upgrade the installer now upgrades the Watch TV scheduled event property in the registry correctly
Playing a DVD with TheaterMode enabled no longer results in the display not being maximized on the secondary LCD
Enabling or disabling the extended desktop when the Media player is running no longer results in the system not responding
Previewing the 3D Pipes screen saver with latest video driver installed no longer results in a slow start and exit of the screen saver
The monitor shutting down when booting to the Windows XP service pack 1 desktop when a RADEON™ 9000 Pro is installed in the system is now resolved
The display resolution is now correct on the first startup after the display driver install
Playing an AVI no longer results in a screen delay. This issue was known to occur under Windows 2000 with the RADEON™ 9600 series card installed
Enabling or disabling the extended desktop when the Media player is running no longer results in the system not responding
The OpenGL hardware overlay is now working correctly in panning mode
The memory size on the RADEON™ 9800 SE not being reported accurately is now resolved
Windows 2000 no longer detects the secondary CRT display adapter as the primary display adapter
The HDTV icon is no longer displayed when Disable Component Output Support is selected
The information tab found in the advanced display properties for the secondary monitor now provides information on the driver version installed
The Bios information is no longer listed incorrectly on the Adapter Information

[RussSchultz]

TOnight I'll check and see if it fixes my 'change the forced AA settings and the graphics card goes wonky' bug.

[Ante P]

HB, you're missing the obvious.
As ATi has stated before Catalyst 3.5 is the first new set of their Workstation + Desktop unified drivers.
They carried over a lot of the work done on the FireGL drivers to the Catalyst drivers.
I know that you know that, a lot, in fact almost everyting, depends on drivers when it comes to the difference between say a Fire GL 8800 and a Radeon 8500 for an example.

As for boosting benchmark scores, did you even consider that they might have removed some bug that caused unnaturally low scores?
If you remember "The HPG" you'd know that when the bug was fixes we saw a pretty dramatic increase in GL Excess and Serious Sam scores etc.

You're just off your rocker this time..

[Quitch]

Benchmarks are merely a way of showing improved performance. You can't list every game under the sun, ATI wouldn't have them all for one (why didn't they tell me that performance improved in game X, but did for game Y? I hate ATI!). This is simply a very easy way of saying "Look, performance has increased".

Of course, after the nVidia farce these claims need to be checked out. Not for accuracy, but for whether these are application specific (would some reshuffled code wipe them out for example?).

[fallguy]

Fact is, benchmarks sell video cards. If they can make a benchmark run faster, without any visual loss, or any other "cheat" then more power to them, or anyone one else.

The problems start when they cross the line to get better performance. Which there is ZERO evidence of here.

[K.I.L.E.R]

Originally Posted by Saem

Calm down.

YOu have no evidence that this is due to benchmark specific optimisations. They could be general optimisations that happened to have a dramatic effect in these benchmarks.

You are correct, all my GL games/apps had a noticeable increase in performance.

[Evildeus]

Originally Posted by K.I.L.E.R

You are correct, all my GL games/apps had a noticeable increase in performance.

Well seems you are the only one, some people don't find any increase or even some little decrease...
http://www.amd-insight.de/tests/graf...ich/index.html

[karlotta]

Originally Posted by Evildeus

Well seems you are the only one, some people don't find any increase or even some little decrease...
http://www.amd-insight.de/tests/graf...ich/index.html

no not the only one... i have found that where i used 2xAA i can run 4xAA and 6xAA instead of 4xAA. this is on all Quake based games... never slower at any time ...? well your milage may very.

[Heathen]

I got improvements across the board in OpenGL as well.

[Hellbinder]

Look this is all I am saying...

There are about 100,000 Words written in the last 3 weeks regarding how Nvidia is Aweful because they artificially inflate or Write Specific code to enhance Benchmarks. Wether or not the IQ is affected is a side issue.

I am not flying off the handle here. It is not an oversimplification either. It is either A. Ok to write specific code into drivers that ONLY enhance Benchmark scores or. B. It is not ok. This has been the core of the entire debate about Nvidia. dont try to tell me that the 80% increase in speed is accross the board in OpenGL. Becuase its NOT. We are talkng about Specific applications with no other function than to BENCHMARK. What have we found out here over the last couple weeks?? That *it seems* like every time a claim to suddenly higher scores in a well known Benchmark happens... then specific application detection has occured. example 3dmark 2001 GT4.

I Completely understand that and am for application detection and special coding to enhance Game play. But I thought it was the general understanding after the recent fiasco that doing this for Benchmarks is not good. maybe i am wrong and everyone needs to rethink what ha been going on recently. If it is ok and completely acceptable then a lot of people owe Nvidia a Big appology.

If you are going to enhance OpenGL performance then do something about this

http://firingsquad.gamers.com/hardwa...view/page8.asp

What good does it do anyone to spend time writing specialized code for a pointless Benchmark. ?

[Hellbinder]

No one said it was a bug that was fixed. Glexcess scores have been similar for a long demonstatable time.

[Quitch]

Originally Posted by Hellbinder

Look this is all I am saying...

There are about 100,000 Words written in the last 3 weeks regarding how Nvidia is Aweful because they artificially inflate or Write Specific code to enhance Benchmarks. Wether or not the IQ is affected is a side issue.

I am not flying off the handle here. It is not an oversimplification either. It is either A. Ok to write specific code into drivers that ONLY enhance Benchmark scores or. B. It is not ok. This has been the core of the entire debate about Nvidia. dont try to tell me that the 80% increase in speed is accross the board in OpenGL. Becuase its NOT. We are talkng about Specific applications with no other function than to BENCHMARK. What have we found out here over the last couple weeks?? That *it seems* like every time a claim to suddenly higher scores in a well known Benchmark happens... then specific application detection has occured. example 3dmark 2001 GT4.

I Completely understand that and am for application detection and special coding to enhance Game play. But I thought it was the general understanding after the recent fiasco that doing this for Benchmarks is not good. maybe i am wrong and everyone needs to rethink what ha been going on recently. If it is ok and completely acceptable then a lot of people owe Nvidia a Big appology.

If you are going to enhance OpenGL performance then do something about this

http://firingsquad.gamers.com/hardwa...view/page8.asp

What good does it do anyone to spend time writing specialized code for a pointless Benchmark. ?

While you may have read the events of the last few weeks, I don't think you fully understand them. These programs are designed to evaluate small functions of the graphics card, and it is quite possible to achieve a large boost by making improvements solely in the area that benchmark looks to cover WITHOUT doing anything with that benchmark in mind.

There is currently no evidence that benchmark optimisation has happened, though I'm keeping my eyes open. What I've seen in 90% of feedback is a boost to OpenGL scores across the board though.

[Dave Baumann]

HB - SPECviewperf is a collection of different benchmarks based on a number of Workstation packages - you might like to find out if the speed of rendering within those packages have improved as well.

[jb]

If only the benchmarks scores have improved then thats not good. But if what ever tweakes they did also effect the program by making that a bit faster then its fine. nV issue with me is that they seem to have targeted only benchmarks demos and not the game iteself and that is very bad.

[demalion]

HB, please read some of the replies to you again. You are indeed simplifying, because you persist in not recognizing that it is not unexpected or inappropriate that benchmarks highlight specific factors more than games, and that it is therefore not necessarily indication of application detection but simply that benchmark performance increases more for them than for most applications. That's why application based detection is bad when it comes to benchmarks...you've let nVidia's simplification color your thinking and transform it to "big changes in benchmark performance are bad". They're not, they're just reason for suspicion...suspicion confirmed extensively for nVidia by other factors, analysis, and specific observations. You're skipping that step, and nVidia's attack on synthetic benchmarking seeks to do the same thing to avoid the inconvenient issues involved.

You're treating synthetic benchmark increases as if they are an inherent "evil" despite my attempt to discuss just that with you before...that's part of exactly what nVidia is setting out to achieve. I think you are missing that because you are only concentrating on the other issue: that they appear to cheat in games too. Did you forget that they are attacking API's they don't control as well? Please consider why that is, and how your reaction fits their desires. Then try to address the technical reasons being proposed in disagreement to your statements.

The same steps are necessary in the case of ATI's increases here, and the performance increases in other OpenGL applications already seem to provide a significant indication in the opposite direction from the indications for nVidia.


Please, read my post, or other similar ones, a bit more closely...your response seem to indicate you skimmed them.