Anand's 9800XT and FX5950 review, part 2

[BenSkywalker]

Oldfart-

Got some links? Looking at the links rever posted, the nV and Ati parts were pretty much equal in Halo without AA/AF with nV taking a bit of a backseat when AA and AF were used. I hope you dont mean the 1 FPS advantage that nV has over ATi in the unplayable 16 x 12 resolution. That would be really reaching.

Check out the Halo bench sometime and compare it to what you get in game. It will give you a good comparison to look at relative performance of the game, but don't put faith in to the actual framerate number. The bench runs a series of different segments, and the transition and load times between the segments are factored in(something like if you ran all the SS benches in order and counted the time it took for the whole thing including loading up each level). As far as calling a 1FPS edge reaching, 70% faster is the standing claimed advantage by some in this thread. If you were to say that the Red Sox obliterated the Yankees in the ALCS this year would someone pointing out they one be a stretch?

The benches in these reviews do not support that statement. The 2 are pretty much neck and neck.

With nV edging it out at the highest resolution. If you want to talk about playability, look at the TRAoD benches that people like to point out. Does the difference between 11FPS and 20FPS really matter that much?

BFG

I have evidence in the form of Unwinder's publically available comments and application that knows nVidia's cheating drivers inside out.

You have evidence that nV was using app detection, and that it caused issues with some games with older drivers. Everyone jumped to the 'cheat' conclusion. Show me which issue is still present that has the IQ problems, or where performance is down by any meaningful margin versus their 'cheat' drivers.

Yes the application is requesting trilinear, not trilinear anisotropic. If you enable trilinear in an application do you expect anisotropic filtering to be enabled along with it? I sure as hell don't.

Anisotropic filtering is seperate from bilinear or trilinear, it is another tehnique. If you enable AF you still need to decide whether to run bi or tri. With ATi's drivers you don't get that choice unless you are running one of the small handful of games that allow you to select AF in game. ATi is doing something very much like nVidia here.

The application isn't requesting AF and therefore they aren't ignoring anything.

If you are requesting trilinear and you get bilinear how can you so much as imply that you aren't getting what you requested?

Right, so you're back to your comment about Gabe being a liar. Yet again you make outrageous claims without any proof and you expect to use them to form the basis of a solid argument for your case?

I've been using Valve's former statements compared to their more recent ones.

I suppose next you'll ironically claim that the burden of proof is on me to prove that Gabe is trustworthy?

Absolutely not. Why would you need to prove that someone who was paid millions by a company that has something bad to say about the companies direct competitor need to have his credibility questioned for a second? If you want to say do you need to offer some proof behind his accusations then I would say yes without a doubt. Even nV's PR spinfest they had some 'proof' for the claims they were making.

In fact I encourage you to use the other thread where they're being brought up and most of them have already been dismissed.

They have been dismissed by the ATi faithful, how shocking. I don't believe any of them anyway(nV or ATi), they are as absurd as what you have been saying for the last few months. I don't swallow a companies PR load no matter who it is, if you want to for one and not the other all the power to you.

What tangent might that be? Making comments that are provable by widespread evidence? Of course I do.

Now we are on to ~300 posts(this may be 300) and you still haven't linked this evidence. Show me where nV was cheating, not had bugs, to improve performance outside of 3DM2K3. This discussion is getting quite old now and so far you haven't had much to offer at all, particularly since it is so widespread. All of these issues that you point out of course need to have performance and 'cheating' cross referenced with the latest WHQL drivers to see if they were bugs or cheats. Point some out.

Of course, shader subsitution and other such cheats will do that. Both Anand and 3DCenter have verified such shader subsitution.

Code substitution? Of course, DX doesn't compile to machine level.

But how can you claim that if they did everything in their power to make the fastest path on the nVidia cards on the basis that they didn't do the same for the slower and irrelevant path?

They tried to make it look like nV needed more special attention then they do, and they obviously grabbed you hook line and sinker.

Do you deny the comments from dozens of developers and reviewers that the FX line has problems running full precision code?

Superior precission if you want to compare it to the R300 remember. It is slower, that is something I have never argued.

My issue is with you using Halo to "disprove" all other findings that don't support your claims. My other issue is that everything non-Halo to you appears to be irrelevent, badly coded or paid for by ATi.

You want me to start using Doom3 like you use HL2? Would it make you feel better if I used another unreleased not complete game running on drivers no one is going to use? Which title that I have issues with do you disagree with my problems for? One of the synthetic benches, one of the unreleased games, or is it TRAoD?

Great. Does that mean Carmack, Gabe, [insert all other developers] along with 3DCenter, [insert other reviewers here] are all wrong?

I already pointed out to you what you quoted Carmack saying. Are you removing the FX59, 59U, 595U and 57 from your comments or not? You have to clarify that first. I've also quoted Sweeney stating that PP is perfectly fine for this generation and that higher precission will be a real issue when the vendors switch to FP32(and that was mainly due to shared resources).

Except precision isn't the only issue here. The other issue is the architecture and how it relies on a ridiculously unrealistic methods of instruction scheduling in order to have any reasonable chance of competing.

And now the drivers handle the scheduling, not a big need for the developers to sweat over it like they did with the older drivers.

 

[oldfart]

Ben

Check out the Halo bench sometime and compare it to what you get in game. It will give you a good comparison to look at relative performance of the game, but don't put faith in to the actual framerate number. The bench runs a series of different segments, and the transition and load times between the segments are factored in(something like if you ran all the SS benches in order and counted the time it took for the whole thing including loading up each level). As far as calling a 1FPS edge reaching, 70% faster is the standing claimed advantage by some in this thread. If you were to say that the Red Sox obliterated the Yankees in the ALCS this year would someone pointing out they one be a stretch?

I've downloaded and played the demo. I don't get your point here Ben. You are saying that Halo is the only PS game that matters and has some sort of significant advantage on nV hardware (not exact words).

I have not seen a review/bench that supports that.

If you want to prove that statement as correct, please post numbers, a bench, something to back that up. Until you do so, the reviews showing that it plays basically the same on both platforms is what is fact.

This particular point is not about the playoff games or TRoD. Its about your statement of the advantage that Halo has when played on nV hardware. All available data shows this to be a false statement.

FWIW, the framerate that game gets on high end hardware for the level of detail it has is pretty sad anyway.

 

[BenSkywalker]

You are saying that Halo is the only PS game that matters and has some sort of significant advantage on nV hardware (not exact words).

Not exact words is one way to say it, not even close it another. I said the game runs faster on nV hardware, I didn't even hint towards the implication that it had a significant advantage. It is the same type of thing I was saying about the ALCS, it was kneck and kneck down to the very end(bottom of the 11th game seven, can't get much closer then that). The R3x0 is supposed to dominate in PS heavy games, obliterate and embarass nVidia, a 70% performance rift is what we are supposed to be seeing. Where is all of that?

FWIW, the framerate that game gets on high end hardware for the level of detail it has is pretty sad anyway.

Pixel shader heavy games, this is what many have been screaming about being the next big thing. Are you surprised by what you are seeing? What things have you found the most impressive in terms of HL2's engine so far? Have you tried TRAoD? It perfoms worse and I would say doesn't look nearly as good.

I've been branded by many to be a zealot because I've been of the mind that Doom3 is a better path for the moment then heavy PS 2.0 useage. It's hard to compare directly to HL2 as so much of what they are doing with that game that looks stunning has absolutely nothing to do with PS 2.0 at all. I've heard a lot about the visuals in MP2 being killer but from what I have seen so far they only use PS for skinning effects(I'm not stating that as a certainty, that is what I have seen so far). Actually, most people are saying that MP2 kills Halo's visuals.

 

[oldfart]

Not exact words is one way to say it, not even close it another. I said the game runs faster on nV hardware, I didn't even hint towards the implication that it had a significant advantage

Well, good we have it straight then. Halo does not run faster on nV hardware significantly, or otherwise. Why bother even posting that it runs faster at all when the truth is, it does not? This is why I used the word "significant". Unless there is some significant difference between two platforms, whey even make a point about it? 1 FPS means nothing, especially at a res no one would even think of using. ATi was ahead of nV in other settings by more than that, but those are not even worth a mention in this game.

 

[BenSkywalker]

Halo does not run faster on nV hardware significantly, or otherwise.

Didn't say that either. Look around at most benches and nV is slightly faster at the higest settings.

Why bother even posting that it runs faster at all when the truth is, it does not?

It does when you set it to the highest resolutions. Should we compare Q3 @640x480 when looking at the 5950 and 98XT?

I was wondering if you thought about the questions I posed about PixelShader games and Halo's performance? First numbers up, all vid settings maxed(there is a problem with the report saying paritcle effects aren't on in the timedemo, they are, this is happening on ATi and nV vid cards with the current patch) -

Date / Time: 10/25/2003 9:48:13 AM (113257281ms)
2000MHz, 512MB, 128M nVidia GeForce4 Ti4200 (DeviceID=0x0253) Driver=6.14.10.5216 Shader=Fixed Function
C:\Program Files\Microsoft Games\Halo\halo.exe -useff -timedemo -vidmode 1600,1200,72 (Version=1.0.2.581)
Frames=4700
Total Time=134.59s
Average frame rate=34.92fps
Below 5fps= 8% (time) 0% (frames) (11.995s spent in 14 frames)
Below 10fps= 9% (time) 0% (frames)
Below 15fps= 9% (time) 0% (frames)
Below 20fps= 13% (time) 2% (frames)
Below 25fps= 20% (time) 7% (frames)
Below 30fps= 42% (time) 24% (frames)
Below 40fps= 59% (time) 41% (frames)
Below 50fps= 83% (time) 70% (frames)
Below 60fps= 94% (time) 88% (frames)
###Sound Options###
Hardware Acceleration= No
Sound Quality= Low
Environmental Sound= No
Sound Variety= Low
###Video Options###
Resolution= 1600 x 1200
Refresh rate= 85 Hz
Framerate throttle= No Vsync
Specular= Yes
Shadows= No
Decals= Yes
Particles= Off
Texture Quality= High

Halo's numbers are a bit off as it figures in the loading time, but in this bench 93% of frames rendered were at or above 25FPS, this is all settings at their highest for vid cards, running 1600x1200 with one exception, I disabled shaders. Now enabling shaders-

Date / Time: 10/25/2003 9:54:41 AM (113431000ms)
2000MHz, 512MB, 128M nVidia GeForce4 Ti4200 (DeviceID=0x0253) Driver=6.14.10.5216 Shader=1.3
C:\Program Files\Microsoft Games\Halo\halo.exe -timedemo -vidmode 1600,1200,72 (Version=1.0.2.581)
Frames=4700
Total Time=348.49s
Average frame rate=13.49fps
Below 5fps= 10% (time) 2% (frames) (37.019s spent in 100 frames)
Below 10fps= 35% (time) 14% (frames)
Below 15fps= 58% (time) 36% (frames)
Below 20fps= 82% (time) 68% (frames)
Below 25fps= 96% (time) 90% (frames)
Below 30fps= 98% (time) 92% (frames)
Below 40fps= 98% (time) 94% (frames)
Below 50fps= 99% (time) 95% (frames)
Below 60fps= 99% (time) 97% (frames)
###Sound Options###
Hardware Acceleration= No
Sound Quality= Low
Environmental Sound= No
Sound Variety= Low
###Video Options###
Resolution= 1600 x 1200
Refresh rate= 85 Hz
Framerate throttle= No Vsync
Specular= Yes
Shadows= Yes
Decals= Yes
Particles= Off
Texture Quality= High

Simply enabling shaders 90% of the frames rendered are now below 25FPS. 258% comparing the non shader with the shader numbers(in terms of average, the difference is much higher if we eliminate the load times). Halo runs just fine on high end hardware without shaders, hell it runs just fine on outdated hardware without shaders.

 

[oldfart]

I fail to see how the above post proves Halo runs better on nV hardware.

For people who are thinking of which (current generation) of card to buy, and Halo performance is important to you, read the reviews. ATi and nVidia cards provide the same performance.

To say Halo runs better or faster on nV hardware requires green colored glasses.

 

[BenSkywalker]

I fail to see how the above post proves Halo runs better on nV hardware.

The above post wasn't really about nV v ATi, it was about Halo more then anything. This has more to do with all the hype we are hearing about non stop involving PS performance. Here is a very popular title that is very pixel shader intensive, as made extremely obvious by the above numbers, and I'm asking where is the 70% performance rift? nV actually came out ahead, even if it was by a tiny margin, when they were supposed to be blown out of the water by an enormous margin. What went wrong here? We have been hearing it non stop for months now, what is the issue?

To say Halo runs better or faster on nV hardware requires green colored glasses.

I could post links to all the different benches that nV came out ahead on, and again I made no implication that Halo had some big edge over nV, I'm pointing out that that 70% edge for ATi turned out to be a deficit by many benches in the first game people want to play that uses shaders very heavily.

We have-

ATi is going to kill nVidia in shader intensive situations

Not always

Yes they are, you are the biggest nvidiot on the web

What about Halo

Halo performs too slow on high end hardware, it isn't a good example

Halo performs slow because of the shaders, and it actually edges out ATi in performance, here are some numbers

That doesn't matter, ATi wins some benches and they are real close anyway, so you can't say they have won accurately

What about ATi killing nVidia?-

Which is where we are now. Been listening for a long time to the PS 2.0 talk, have bit my tongue for the most part until this thread. Now we have drivers that fix the bugs, and a good game that is very shader heavy and I'm wondering where is all the rhetoric about the huge performance rift, and why aren't people gushing over Halo's stellar visuals? This is what people wanted, I've seen more hype over TRAoD which is a steaming pile of sh!t by any reasonable standards as a game, and people bashing Halo saying it's too slow(despite looking better and running faster then TRAoD).

All the talk about pixel shaders revolutionizing the gaming market, why aren't people all hyped up about a killer game that uses what they have been asking for? It is here, now, and now people are bashing it. I likely should have approached the topic from a different angle, looking at the highest settings now isn't really valid(I still don't understand why, but I am the biggest nVidiot on the web so perhaps it was in some memo that doesn't get sent to my type?).

 

[oldfart]

You are straying off the subject again. My point is very simple.

You said

There is one game (Halo) on the market that has heavy PS useage that includes DX9 shaders that I have any interest in, and it runs faster on nV hardware.

That is a false statement.

Halo does not run faster on nV hardware. According to actual reviews and benches on the web, it runs about the same on nVidia and ATi hardware. You haven't posted a thing to prove otherwise.

Yeah, you can always turn off detail to make a game run faster such as PS2.0 in the example of Halo. Here is what Anantech said about Halo PS 2.0 effects:

Of course, we have gotten plenty of emails pointing out the hack that allows AA to work with Halo. This involves adding a flag to tell Halo to DisableRenderTargets for your particular video card. The problem with this is that rendered textures are used all throughout the game, and many of the awesome PS2.0 effects are lost. Water effects disappear, walls go from dirty and grimy looking to plastic and shiny, and the image quality we get from the hack takes away too much to be worth it in our opinion.

 

[Rogodin2]

Max Payne two is using heavy pixel shaders and NV get's drubbed by ati.

rogo

 

[oldfart]

Originally posted by: Rogodin2
Max Payne two is using heavy pixel shaders and NV get's drubbed by ati.

rogo

I wonder which catagory MP2 performance will fall into?
biased review?
flawed review?
poorly coded game?
irrelevent since I dont like that game?
part of the general anti nVidia conspiracy?

 

[Rogodin2]

I'm curious too

rogo

 

[ZimZum]

Neither ATI nor nVidia were given the opportunity to optimise for Max Payne 2 (yet). Which makes its performance numbers all the more interesting.

 

[BenSkywalker]

According to actual reviews and benches on the web, it runs about the same on nVidia and ATi hardware. You haven't posted a thing to prove otherwise.

About the same or exactly? If it doesn't run exactly the same, then it runs faster on one or the other. I looked at the highest resolution settings and nV was faster, should we now start using the lowest resolution to compare games, or is your issue that in order to be considered faster you need to run a certain given percentage?

Max Payne two is using heavy pixel shaders and NV get's drubbed by ati.

I was mistaken earlier about the pixel shader useage, apparently they have no DX9 level pixel shaders in the game. The mirrors require PS 1.4 though.

I wonder which catagory MP2 performance will fall into?
biased review?
flawed review?
poorly coded game?
irrelevent since I dont like that game?
part of the general anti nVidia conspiracy?

I'll quote one of the guys from Remedy-

Max Payne 2 can't be used as a benchmark.

We enable and disable features "completely under the hood", so you will not be able to get comparable results for varying hardware.

Also, the game doesn't have any built-it timedemo capabilitites. Any kind of "showextendedfps" hacks are not benchmarks - their error margins are HUGE.

We recommend using established benchmarks such as 3DMark03 for reliable testing...

SamiV.

Link. I saw the quote over at B3D but no links(that is why the word benchmark shows up bolded), that is from Remedy.

 

[Rogodin2]

That is bunk Ben and you know it.

All you had to do is quote the whole reply and you'd have posted what was straight out of the PR piehole.

"We recommend using established benchmarks such as 3DMark03 for reliable testing..."


What do you expect them to say? "yea go benchmark the hell out of maxpayne 2 and when nvidia cards choke just tell everyone that it's a GAME LIMITATION and we need to give nvidia some time optimize for it." That's like telling all the gourmet coffee buyers that robusta is just as tasty as arabica, jeez.

Fraps Txt log timed runs are valid regardless of "what's under the hood" a crappy metaphor for "we are neutral and won't take a stance on the whole benchmarking mess."

 

[Rogodin2]

"I was mistaken earlier about the pixel shader useage, apparently they have no DX9 level pixel shaders in the game. The mirrors require PS 1.4 though."

This just means that ATI runs pixel shader 1.4 much faster than nvidia.

You can probably draw the conclusions about 2.0 without even using shadermark.

rogo

 

[BenSkywalker]

All you had to do is quote the whole reply and you'd have posted what was straight out of the PR piehole.

Great reading skills you have there. The rest of his comments seem to be over your head, ask around exactly what it means.

This just means that ATI runs pixel shader 1.4 much faster than nvidia.

Because those mirrors are just everywhere. You assume that MaxPayne is limited by Pixel Shaders, provide some numbers to back it up. I made the assertion with Halo, had it questioned, so I provided numbers. I am questioning you on the point now.

You can probably draw the conclusions about 2.0 without even using shadermark.

I'm sure you can. When is ShaderMark 2.1 going to come out and use Dawn BTW?

 

[reever]

I made the assertion with Halo, had it questioned, so I provided numbers. I am questioning you on the point now.

Seriously, what is with you and Halo? You say it runs faster on Nvidia hardware, i provide numbers which show it doesnt with 1 exception, and you still say it runs faster on nv hardware?

 

[BenSkywalker]

You say it runs faster on Nvidia hardware, i provide numbers which show it doesnt with 1 exception

The one setting being 16x12.

 

[reever]

Originally posted by: BenSkywalker

You say it runs faster on Nvidia hardware, i provide numbers which show it doesnt with 1 exception

The one setting being 16x12.

And? Is that THE most important resolution to the entire gaming community?

 

[Rogodin2]

Great social skills you have there friend.

"Max Payne 2 can't be used as a benchmark.
We enable and disable features "completely under the hood", so you will not be able to get comparable results for varying hardware. "

This is the only quote that is ambiguous-and I posted my thoughts.

"You assume that MaxPayne is limited by Pixel Shaders."

I implied that maxpayne2 runs much better on ati cards+, and that using shadermark, 3dmark2k3, and aquamark3 the radeons run advanced pixel shaders much more efficiently.

rogo

 

[oldfart]

Sorry, Ben, you wonder why you get that label of being so nVidia biased? Your 16 x 12 Halo nonsense is exactly what brings on that type of comment. It has been earned.

I'll go through the trouble of cut and pasting some benches from the reviews for all who read this forum to see. This is with the 52.16 drivers

Anandtech
10 x 7 x 0AA x 0AF
5700U 34.6
9600XT 34.3

HardOCP
10 x 7 x 0AA x 8AF
Framerate Min - Ave - Max
5950U: 14 - 32.8 - 79 <----notice no mention of the large performance difference when AF is used
9800XT 20 - 45.1 - 89

TechReport
10 x 7
5700U 32.6 <------nVidia SMOKES ATi by 1.1 FPS here!!! I'm surprised you missed this one!
9600XT 31.5

12 x 10
5700U 22
9600XT 21.2

16 x 12
5700U 15.6
9600XT 15
5950U 28.32 <------Here is the massive 0.91 FPS nVidia advantage @ 16 x 12!!!
9800XT 27.41

Hothardware
10 x 7
5950U 57.82
9800XT 58.8

12 x 10
5950U 38.9
9800XT 39.04

THG
10 x 7
5950U 58.6
9800XT 60

12 x 10
5950U 39.3
9800XT 39.4

16 x 10
5950U 28.3 <------Again, nVidia crushes ATi by 0.9 FPS in the highly used 16 x 12 resolution!!!
9800XT 27.4

GamePC
10 x 7
5950U 58.15
9800XT 60.9

16 x 12
5950U 28.1
9800XT 27.7 <---whopping 0.4 FPS here


XBit
10 x 7
5950U 46.5
9800XT 45.9

12 x 10
5950U 38.2
9800XT 37.3

16 x 10
5950U 28
9800XT 27.1 <-----Yes it's true. That massive 0.9 FPS again.
Well, the results of applied optimizations seem to be more than evident. NVIDIA graphics cards perform as fast as ATI based ones. However, they owe these good results not only to the shaders optimizing compiler. On NVIDIA GeForce FX graphics cards, the Halo graphics engine uses half-precision pixel shaders 2.0, which provides the NVIDIA GPUs with some performance advantages.

However, I didn?t reveal any issues caused by these pixel shaders optimizations or any calculations precision worsening. The only remark here is that sometimes some objects of the scenes appear to have no fog around them.

10 x 7 x 8AF
5950U 38.7
9800XT 43.6

12 x 10 x 8AF
5950U 28.3
9800XT 33.1

16 x 10 X 8AF
5950U 20.6 <-------GASP! We have a problem!
9800XT 24




Any sane unbiased person would conclude the cards perform remarkably equally. Only one who is incredibly biased would call one or the other faster. The view through green colored glasses is very apparent here.

16 x 12 difference IS LESS THAN 1 FPS FOR PETE'S SAKE This is just too sad. Is this the best you can do??

So once again:

There is one game (Halo) on the market that has heavy PS useage that includes DX9 shaders that I have any interest in, and it runs faster on nV hardware.

Is a joke. Its not true and just more nVidia bias showing through.

 

[BFG10K]

Ben:

Show me which issue is still present that has the IQ problems, or where performance is down by any meaningful margin versus their 'cheat' drivers.

How about the forced bilinear AF issue, an issue which you either continually deny as a cheat or conveniently forget when it suits you?

Anisotropic filtering is seperate from bilinear or trilinear, it is another tehnique.

Thankyou Captain Obvious.

If you enable AF you still need to decide whether to run bi or tri.

And since the application isn't even requesting AF to begin with ATi's drivers decide for it by doing a partial trilinear AF.

Tell me Ben, how is that the same thing as nVidia's drivers forcing bilinear in every situation not matter what the app, control panel or user asks?

With ATi's drivers you don't get that choice unless you are running one of the small handful of games that allow you to select AF in game.

And with nVidia's drivers you don't get that choice at all. Besides, ATi is still giving you trilinear AF to some degree so the setting is both doing something and you still have a choice in the matter regardless of the application or the API it runs on.

If you are requesting trilinear and you get bilinear how can you so much as imply that you aren't getting what you requested?

Because you aren't requesting AF, that's why. You're mixing two things up and trying to draw them together, totally ignoring the fact that nVidia has you clamped down to bilinear AF no matter what you or the application wants to do. Yet again you continue to demonstrate an infuriating desire to excuse nVidia whenever possible while slamming ATi for something that's completely different.

I've been using Valve's former statements compared to their more recent ones.

And you've continually failed to explain in what way Valve's mixed mode path was creating a innacurate picture of things. Do you deny it's 100% optimal? If you do then I'd like to hear why. If not then stop posting this ridiculous argument.

I expect that you'll never answer this though because doing so would punch a hole into your illogical argument.

Why would you need to prove that someone who was paid millions by a company that has something bad to say about the companies direct competitor need to have his credibility questioned for a second?

So in other words yet again you have nothing but invented theories to back your claims.

They have been dismissed by the ATi faithful, how shocking

If you have something worthwile to add about them then by all means, post it.

Now we are on to ~300 posts(this may be 300) and you still haven't linked this evidence.

I have, you just continue to deny it and/or put some ridiculous spin onto it .

Gabe is a liar.
App X is poorly coded/paid for by ATi/I don't like it/etc.
nVidia driver compression is not targetting anti-cheat programs even though it came out at exactly the right time to do so (ie right after they were discovered to be cheating).

I'm quite frankly really sick of constantly going around in circles with issues that have already been covered and then re-covered repeatedly.

Code substitution? Of course, DX doesn't compile to machine level.

Here we go again, back to this same old crap again.

I quite clearly demonstrated to you about ten posts back as to what the difference between realtime subsitution and compilation is. But because you constantly appear to "forget" points that have already been proven time and again this argument never gets anyhere.

They tried to make it look like nV needed more special attention then they do, and they obviously grabbed you hook line and sinker.

So again I'll ask whether doing nothing else except compiling the full precision path with Microsoft's compiler would've created greater performance than not only nVidia's mixed mode path but also to beat ATi's path. Because if not your argument has absolutely no basis to even exist since you still haven't been able to prove by any stretch of the imagination that Valve's mixed mode path is not completely optimal.

Yet again you fail to answer the question and you continue to jump around the issue into whatever angle that suits you, and this is why we're still here after nine pages.

You want me to start using Doom3 like you use HL2?

But ironically Doom III is doing exactly the same thing as the rest of the applications we've discussed! Carmack is making a special path for NV3x class hardware because it runs so slow. I guess he's just another ATi PR monkey like Gabe and/or ID is trying to... what was it you said Valve did:

They tried to make it look like nV needed more special attention then they do, and they obviously grabbed you hook line and sinker

I guess ID is also trying to make it look like nVidia needs special attention when they don't, huh?

Are you removing the FX59, 59U, 595U and 57 from your comments or not? You have to clarify that first.

No because while the NV30 core is weaker than the NV3x core it stills shares a lot of flaws with it. Valve benchmarked a 5900, not a 5800. Of course you'll continue offer the claim that Valve are lying as your "proof" against this point which is why after another nine pages of this crap we'll still be in exactly the same place as we are now.

And now the drivers handle the scheduling, not a big need for the developers to sweat over it like they did with the older drivers.

I agree, this should make things better.

 

[BenSkywalker]

Reever-

And? Is that THE most important resolution to the entire gaming community?

Yes.

Oldfart-

Any sane unbiased person would conclude the cards perform remarkably equally. Only one who is incredibly biased would call one or the other faster. The view through green colored glasses is very apparent here.

Where is the 70% edge that your type have been talking about when playing shader heavy games? nVidia ends up coming out ahead in the most shader intensive settings they use. No matter how small the victory, considering that your people have been banging away at this 70% edge line, how is it possible for that to happen.

BFG-

How about the forced bilinear AF issue, an issue which you either continually deny as a cheat or conveniently forget when it suits you?

I don't think either ATi or nV are cheating with their forced filtering. I can quote myself on non nVidia parts from years ago on this same issue if you would like, I've never considered filtering optimizations a cheat, you do if nVidia does them but not if ATi does.

And with nVidia's drivers you don't get that choice at all. Besides, ATi is still giving you trilinear AF to some degree so the setting is both doing something and you still have a choice in the matter regardless of the application or the API it runs on.

You are much closer to trilinear on nV's parts then you are on ATi's when they are both running their optimizations. nVidia isn't running bilinear on any of their mip levels based on anything I have seen, they are running a 'brilinear' hack. That said, ATi is retaining a higher level of anisotropy on the further mip levels then nV.

Because you aren't requesting AF, that's why. You're mixing two things up and trying to draw them together, totally ignoring the fact that nVidia has you clamped down to bilinear AF no matter what you or the application wants to do. Yet again you continue to demonstrate an infuriating desire to excuse nVidia whenever possible while slamming ATi for something that's completely different.

You are really stretching on this one. Are you next going to say if you enable AA then point filtering is OK? You do not get bilinear with nV no matter what, you don't get full trilinear under D3D applications when you enable AF. For ATi, the same is true outside of a small handful of games.

And you've continually failed to explain in what way Valve's mixed mode path was creating a innacurate picture of things.

Actually I've explained it multiple times. The way the test was set up it gave lower levels of performance for 'pure' DX9 performance then it would have if it was compiled for both nV cards the same way.

If you have something worthwile to add about them then by all means, post it.

And act like you have been? No thanks.

I have, you just continue to deny it and/or put some ridiculous spin onto it .

Gabe is a liar.
App X is poorly coded/paid for by ATi/I don't like it/etc.
nVidia driver compression is not targetting anti-cheat programs even though it came out at exactly the right time to do so (ie right after they were discovered to be cheating).

I'm quite frankly really sick of constantly going around in circles with issues that have already been covered and then re-covered repeatedly.

Your first point, I said I don't trust him and according to Dave he set the test up to show a performance gap between 'pure' DX9 code and optimized DX9 code for nV parts. He did not compile for nV boards on the pure path.

Point 2, ShaderMark uses a made by ATi demo and doesn't even work on DirectX's reference rasterizer for all the tests. Where is anyone disputing those points? If Dawn was used for a ShaderMark 2.1 test would you consider that legit? 3DMark2K3 is poorly optimized by their own statements, they try and make the bench run as slow as poosible according to Futuremark, notice that that point hasn't been argued? Also, I don't quote the numbers despite the fact that they show what you seem to like to think is what I 'want' to see. TRAoD all I can say is check it out for yourself and give an honest assesment and see what you have to say. Is there anything else?

Your third point is a conspiracy theory, I don't subscribe to them in the least and never have.

I quite clearly demonstrated to you about ten posts back as to what the difference between realtime subsitution and compilation is. But because you constantly appear to "forget" points that have already been proven time and again this argument never gets anyhere.

There isn't really any 'real time' substitution or compilation. Also notice that Dave didn't take issue with that line of thought? Why do you think that is?

Because if not your argument has absolutely no basis to even exist since you still haven't been able to prove by any stretch of the imagination that Valve's mixed mode path is not completely optimal.

You have used those numbers repeatedly to "show" how much more work needs to be done to get nVidia's performance up. My entire issue throughout this entire thread on that point has been that Valve was showing a larger performance rift then there actually was between nV's hardware and nV's hardware(no, that wasn't a typo). You have been trying to spin that in to something else.

But ironically Doom III is doing exactly the same thing as the rest of the applications we've discussed! Carmack is making a special path for NV3x class hardware because it runs so slow.

Carmack is making a rendering path for exactly what he wanted for Doom3, ATi simply can't run it. nVidia hardware was most definitely designed with Carmack's upcoming engine in mind, and he is exploiting that. Carmack has already stated that if ATi had comparable extensions to what nVidia has that he would use them. Trying to spin a shortcoming of ATi's part into a negative statement about nV is a bid off.

I guess ID is also trying to make it look like nVidia needs special attention when they don't, huh?

No discernable quality difference. id is using the path they asked for from hardware vendors, ATi doesn't support it.

No because while the NV30 core is weaker than the NV3x core it stills shares a lot of flaws with it. Valve benchmarked a 5900, not a 5800. Of course you'll continue offer the claim that Valve are lying as your "proof" against this point which is why after another nine pages of this crap we'll still be in exactly the same place as we are now.

That question was about Carmack's comments. You are trying to use his comment about needing to drop to INT12 in some cases about the NV30 against the NV3X line.

Of course you'll continue offer the claim that Valve are lying as your "proof" against this point which is why after another nine pages of this crap we'll still be in exactly the same place as we are now.

Then you are saying that using the same compiler for both of the nV paths is less honest the using different ones?

 

[oldfart]

Ben

Where is the 70% edge that your type have been talking about when playing shader heavy games? nVidia ends up coming out ahead in the most shader intensive settings they use. No matter how small the victory, considering that your people have been banging away at this 70% edge line, how is it possible for that to happen.

Who said anything about 70%?? I and every web review on the planet says they perform the same. 70% is your number you keep spewing, not mine.

Once again, stay on topic. We are talking about your Halo comments here right?

You are in here calling evey game nVidia benches slower on biased, invalid, flawed, rigged, screwed up, etc etc. Then you find one PS heavy game that performs identically to ATi and you are all over it for less that one FPS @ 16 x 12? That is just too sad.

And of course, no mention that ATI is faster (if you want to split hairs) on the resolutions people actually use, and is much faster when using AF.

How do you expect people to take you seriously. It doesn't take a rocket scientist to see through it.

BTW, what is "my type"? The type that doesn't believe the BS that is posted here? The type that believes the countless reviews on the web instead of discounting them all for some imaginary reasons? What is interesting is how you are pretty much alone in "your type" of completely one sided biased opinion.

Anyway, I'M DONE with this topic. I've made my point and backed it up with factual reviews for all to see. The numbers are there and dont lie. This could go on forever and I've wasted too much time on it already.

 

[BFG10K]

Ben:

you do if nVidia does them but not if ATi does.

If ATi did straight bilinear filtering when trilinear was set in the control panel then I'd agree with you. The point is that they don't and because the application doesn't request AF then it's their choice as to how it does it. The other point is that they do true trilinear AF when the application requests it, unlike nVidia who does not.

Are you next going to say if you enable AA then point filtering is OK?

No because there's no option in the control panel or the game for point filtering. Point filtering doesn't even enter into the discussion at all.

The way the test was set up it gave lower levels of performance for 'pure' DX9 performance then it would have if it was compiled for both nV cards the same way.

You've finally quantified (see the bold) what you mean but you didn't do this before. Up until now you simply stated that Valve was deceiving the public in general by creating an unrealistic performance rift when that is not the case since their mixed mode path is fully optimal.

However for the full precision path I completely agree with you that the performance delta is larger than what it should be. However I'd also like to add that Microsoft's compiler may not have been available at the time the path was benchmarked and also the fact that Valve have gone out of their way to make the mixed mode path as optimal as possible which basically makes the full precision path somewhat irrelevant.

If the mixed mode path wasn't compiled with Microsoft's compiler then you'd have a much stronger case (and I'd agree with you) but as it stands now I don't believe Valve did anything under-handed. So they didn't completely optimise the irrelevent path. Big deal. ATi is still ahead when nVidia is completely optimal and I'd believe that was the idea Valve were trying to get across.

Point 2, ShaderMark uses a made by ATi demo and doesn't even work on DirectX's reference rasterizer for all the tests. Where is anyone disputing those points?

I wasn't aware of that and I agree with you that it's a bit strange that the reference renderer doesn't work with some of the tests.

3DMark2K3 is poorly optimized by their own statements, they try and make the bench run as slow as poosible according to Futuremark, notice that that point hasn't been argued?

Yes but you've also argued against using Max Payne 2 as a benchmark on the grounds of what the developers say, the same developers that then turn around and say that using 3DMark is a much better option. Who are we to believe then? You can't selectively mix and match statements when it suits you.

There isn't really any 'real time' substitution or compilation.

Not for compilation as by definition it can't be realtime. However for subsitution there certainly is and it's one of the cheats that FutureMark detailed. Basically it's when the drivers detect the presence of a shader and completely ignore the whole thing and replace it with a pre-compiled version that may or may not produce the exact same output. That is not an optimisation by any stretch of the imagination.

Also notice that Dave didn't take issue with that line of thought? Why do you think that is?

Because it's correct and it comes directly from FutureMark's findings? Because the shader subsitution has been listed many times in other reviews that discuss nVidia's dubious optimisations?

Also please note that I don't think the realtime code reordering in the 5x.xx drivers is cheating; it's a genuine optimisation, provided it doesn't change the output or the work requested by the program. I just thought I'd clear that up and also point out that this is not what I mean when I talk about shader subsitution.

Carmack is making a rendering path for exactly what he wanted for Doom3,

He has also explained, like all of the other developers, that the path is necessary to get any reasonable form of performance from the NV3x. Quite clearly this isn't a conspiracy theory or a Valve lie.

Also it appears that even your holy grail application Halo is also using a special reduced mode path to get reasonable performance on nVidia's boards. What do you say to that Ben? Whether the lower precision looks as good as ATi's full precision is debatable; what isn't debatable is that the NV3x core has severe architectural problems when running at full precision.