Aquamark3 Preview at 3DGPU

[Joe DeFuria]

http://www.3dgpu.com/modules/wfsection/article.php?articleid=74&page=0

Only immeditaely disappointing thing is that there is a comparison between a 5600 and a 9800, so of course we don't have a good idea of how the different architectures directly stack up in this test.

They did test 5600 with both Det 40s and Det 50s.

I'll have to read this a few times and come to grips with all the features this offers...but AM3 looks to be functionally a nice tool.

On a related note: are we surprised...

...and you can very easily see that ATI's Radeon 9800 Pro looks much better over both Detonator drivers on the GeForce FX 5600 Ultra. Areas to note are the ground textures to the left-foreground of the green landrover, more lighting shown on the orange/green small ship in the background, and especially the wheels of the green landrover in the foreground and the one on the far bottom right.

The Detonator 51.75 drivers loses even more details over the 45.33 drivers, and it seems color precision was lowered quite visibly.

Time for someone to really do an in-dpeth analysis of the Det 50s

 

[Matt]

I haven't been able to really do an in-depth analysis of the Detonator 50 as much as I'd like. Between juggling a review of a 5600 Ultra and getting this preview out on AquaMark 3, Real Lifeâ„¢, and doing the graphic card shuffle more times than I'd care to admit, I haven't had much time to really delve into the new Det 50's.

I can't say it has to do with color precision, so I took that out and am going to ask NVIDIA about it, to be sure. All I do know is that the Det 51.75 drivers look worse than the 45.33 drivers, and the R9800 Pro with the CAT 3.7 looks much better overall, no questions asked. I mean it's like night and day to me.

I wish I had a 5900 Ultra to test, and I would've if I had one, but I don't. I just tested all 3 cards I have lying here.

 

[Joe DeFuria]

I haven't been able to really do an in-depth analysis of the Detonator 50 as much as I'd like. Between juggling a review of a 5600 Ultra and getting this preview out on AquaMark 3, Real Lifeâ„¢, and doing the graphic card shuffle more times than I'd care to admit, I haven't had much time to really delve into the new Det 50's.

First, thanks for the article. It was a good read.

I'm not really expecting you to do it. I really meant "someone", not in particular you. The "rolleyes" was not directed at you...it was just because of the fact that someone has to do it. It's a shame now that "driver release scrutiny" is almost now forced upon us as "something else that has to be done."

I can't say it has to do with color precision, so I took that out and am going to ask NVIDIA about it, to be sure.

Good call. Color precision is one possible explanation, but it could be lots of other things. I would attempt to may run it through the refreast, for comparison...and also ask the AM3 developers for their opinion on what's going on. They would be as equally qualified (and more trustworthy) than nVidia.

All I do know is that the Det 51.75 drivers look worse than the 45.33 drivers, and the R9800 Pro with the CAT 3.7 looks much better overall, no questions asked. I mean it's like night and day to me.

That's a shame, and it quite possibly explains Gabe Newell's "reaction" to the Det 50's.

I wish I had a 5900 Ultra to test, and I would've if I had one, but I don't. I just tested all 3 cards I have lying here.

Heh...check on E-Bay....there might be a rather sudden surge of used 5900 sales.

 

[Matt]

I know the rolleye wasn't directed at me, just thought I'd explain what's on my plate here. Good call on asking my contact at Massive about the lowered image quality, I'll do just that (and ask NVIDIA for kicks to see what they say too). eBay may have a flood of auctions up on those 5900U's, but will there be any bids?

Glad you enjoyed the article.

 

[Veridian3]

Driverheavens Aquamark3 article...

click http://www.driverheaven.net/articles/aquamark3/index.htm to view our preview. The article features 96/97/98 series Radeon cards and a FX 5900Ultra as well as driver comparisons for each manufacturer. Additionally there is a very interesting image quality comparison.

Stu

 

[Joe DeFuria]

Re: Driverheavens Aquamark3 article...

click http://www.driverheaven.net/articles/aquamark3/index.htm to view our preview. The article features 96/97/98 series Radeon cards and a FX 5900Ultra as well as driver comparisons for each manufacturer. Additionally there is a very interesting image quality comparison.

Stu

Yes, interesting indeed. Another "Det 50's obviously reduce image quality" observation.

Question: does DH use 3DMark03 as a standard benchmarking tool? I don't honestly know, and I'm too lazy to check myself.

 

[Veridian3]

In the past, yes plus 01, glexcess codecreatures etc.

As of our next review we are changing the benchmarking and will be using a few different things.

AM3
Splinter Cell
UT2003
TR:AOD
3DMark03 (the people like to see it )
Elite Force 2 (custom timedemo)

And a few more to be added...

Stu

 

[John Reynolds]

Wow, from the Det 44.03 to the new 51.75 the 5900U saw basically a 100% improvement.

Now, im sure most of you have read Gabes recent comments regarding the detonator 51.75s, and Nvidia's offical response but I really do have to say, having seen this first hand it confirms to both myself and Veridian that the new detonators are far from a high quality IQ set. Alot of negative publicity is currently surrounding Nvidia, and here at driverheaven we like to remain as impartial and open minded as we possibly can, but after reading all the articles recently such as here coming from good sources and experiencing this ourselves first hand, I can no longer recommend an nvidia card to anyone.

Ouchies.

 

[Dave Baumann]

Well, it would seem that Shader replacement has taken place here as well - unless the runtime optimser now optimises to use lower precision. I'd hope that others treat Aquamark similarily to how they have treated 3DMark03.

 

[Joe DeFuria]

Well, it would seem that Shader replacement has taken place here as well - unless the runtime optimser now optimises to use lower precision. I'd hope that others treat Aquamark similarily to how they have treated 3DMark03.

Principally, I hope Massive treats nVidia like Valve did, and unlike FutureMark has in recent past. (Pending FutureMark's forthcoming statements.)

 

[demonic]

And a few more to be added...

Please don't forget Halo which has recently gone gold.

 

[Sazar]

Well, it would seem that Shader replacement has taken place here as well - unless the runtime optimser now optimises to use lower precision. I'd hope that others treat Aquamark similarily to how they have treated 3DMark03.

Principally, I hope Massive treats nVidia like Valve did, and unlike FutureMark has in recent past. (Pending FutureMark's forthcoming statements.)

it would be nice but considering massive's fawning @ nvidia's amazing ability to gain performance through spectacular driver's (they said something to that effect) I find it hard to believe massive will treat nvidia any different than FM...

I could well be wrong but there is no indication I have seen that would suggest otherwise..

 

[g__day]

Matt, Veridan3 and of course Massive Develoment thanks for the great reviews done so far. I can't wait for Monday and I'm please you trialled the Det 51.75

I am very pleased with what this benchmark appears to have delivered and thank you guys for excellent reviews in the time available.

BTW Dave when you say shader replacement you are referring to a driver shader alteration by NVidia - not Aquamark 3 itself down grading a PS 2.0 shader to a PS 1.4 or lower shader as Ingo Frick pointed out is built in

http://www.guru3d.com/article.php?cat=article&id=76&pagenumber=10

My concern is this, if a shader operation can fallback from 2.0 towards 1.4, 1.3 or even lower. Then this is not a 100% DX9 benchmark and such a fallback would be very hard to see in performance. It's as much DX9 compatible as it can be as it will utilize DX9 as much as it can. Also a small side-note, AM3 does not rely so extremly on 2.0 Shaders as for example Half-Live will, AM-3 approximately generates 30% of it's screen pixels (without overdraw) through Pixel Shader 2.0 It's measures performance in DirectX 9 very well though. I've raised my concerns towards Massive here though and this was the response from Ingo, Technical Director.

The AM3 shaders have been designed with the fallback possibility in mind. For every shader which has been written in 2.0, 1.4, 1.3, 1.1 we have fallbacks, which generate a VERY similar result on the screen. Very similar means, that we do not ignore features like shadow, double caustics, detail maps etc. and therefore the benchmark provides an identical benefit for the user (identical regarding screen content) under different techniques. This concept is frequently used in other industries to compare different techniques. To make them comparable you first have to define the result and afterwards you have to guarantee that every technique generates exactly this result. This concept is often referred as “basket of goods/benefitâ€￾ concept.

But nevertheless we try to create the same screen content with every technique, we face minor differences which arise from the internal accuracy which is smaller when we select a multi-pass technique instead of a multi-texture technique. The fallback mechanisms are optimized as heavily as possible, so we can ensure that the way to achieve the defined result is a near optimal way for all ps/vs versions. For that reason, the AM3 score is comparable, because the only fact that counts is the users benefit (ignoring image quality losses, which are negligible in AM3).

But you still have to pay attenantion regarding other resulting numbers. For instance, if you want to compare the number of triangles rendered per second, you compare apples and oranges because multipass techniques lead to significantly different numbers compared to multitexture techniques.

Best regards,

Ingo

 

[jandar]

I see three things so far:

shader replacement
lower AF (look at the wheels in the images)
and fog rendering is non existant on the DET 50 series. (look at the ships in the background)

no wonder why Valve doesn't like these things.

 

[WaltC]

Not being harsh but just honest...this is all so "nVidia" to me...I can clearly recall years ago the first Detonator driver release with claims of "up to 50% performance improvement" here, or "up to 25% performance improvement" there...etc. Every time a new driver set was released it was more of the same--especially when a competitor was releasing new hardware in between nVidia's hardware releases. I mean, if you add up the cumulative percentage increase claims nVidia made for driver sets released for specific products the ending driver would represent about 400-500% total performance improvement across all the drivers... Almost always, the improvements were strictly found in benchmarks, and/or the result of some kind of IQ compromise.

If it was 30 years ago I'd have said nVidia sounds like a stuck record...but now I'll say they sound like they're stuck in a recursive loop... What is it with these guys? Why do they keep banging their corporate head against the wall with this kind of thing? It's like past habits are so deeply ingrained that the idea of changing things isn't conceivable...

 

[Sazar]

Not being harsh but just honest...this is all so "nVidia" to me...I can clearly recall years ago the first Detonator driver release with claims of "up to 50% performance improvement" here, or "up to 25% performance improvement" there...etc. Every time a new driver set was released it was more of the same--especially when a competitor was releasing new hardware in between nVidia's hardware releases. I mean, if you add up the cumulative percentage increase claims nVidia made for driver sets released for specific products the ending driver would represent about 400-500% total performance improvement across all the drivers... Almost always, the improvements were strictly found in benchmarks, and/or the result of some kind of IQ compromise.

If it was 30 years ago I'd have said nVidia sounds like a stuck record...but now I'll say they sound like they're stuck in a recursive loop... What is it with these guys? Why do they keep banging their corporate head against the wall with this kind of thing? It's like past habits are so deeply ingrained that the idea of changing things isn't conceivable...

its all that evil ati's fault.. curse them and their r3xx lineup..

if it wasn't for that nvidia would be chugging along happily not having to defend themselves @ every step...

 

[Typedef Enum]

This is what I think. I believe that representatives from all the major hardware review sites should get together, and draft a document that essentially states that they will no longer accept this kind of nonsense in their previews/reviews any longer. In other words, as long as you continue to kick out this kind of crap, we will _refuse_ to use those drivers when talking about numbers.

If all the major sites out there agree to this basic concept and simply refuse to use these drivers...forward the declaration to nVidia, I think they will actually think very hard about providing some sort of checkbox in their drivers that will ultimately allow the end user to enable/disable the optimizations so _they_ can choose how they want to play the game.

I have been thinking about this for a long time. I truly believe that nVidia would actually cave if the Anandtechs/HARDOCP's/Toms/B3D's of the world simply flat out refused to use these drivers.

 

[Seiko]

"The Detonator 51.75 drivers loses even more details over the 45.33 drivers. Especially note how the green on the foreground landrover is darker than the other two shots. Now, these drivers are beta, and not available to the public, so let's hope by the time they're released, the lowered image quality won't be in them. Because frankly, I'm just shocked how much better the Radeon 9800 Pro looks over the new Detonator 50 drivers on a GeForce FX 5600 Ultra, and any self-respecting gamer will choose the image quality on the Radeon 9800 Pro anyday. The amount of visual quality loss is not worth the, on average, 2-3fps that are shown to have been gained with these drivers in AquaMark 3."

Well it's a pleasant surprise to see a site publish it as they see it, finally some one calling a spade a spade?

I've said for sometime now it's not until sites start publishing these heavy hitting statements will the likes of Nvidia and any other IHV for that matter realise they simply can't get away with this.

It will be interesting to see if any other sites observe and publish this IQ degradation path Nvidia has continued to traverse!

 

[L233]

What is it with these guys? Why do they keep banging their corporate head against the wall with this kind of thing?

They do it because they don't really have a choice. It's either being like 50% slower when running DX9 shaders or less image quality. Since most reviews don't look at image quality, it's a no-brainer. If I was Nvidia, I'd do the same. Mask your own weakness as good as possible, create a PR smoke screen and hope most customers won't look too closely.

I mean... what do you expect? Nvidia issuing a press statement like "Ok guys, our hardware isn't really competetive in new DX9 games. Sorry, for that. You should better just buy an ATI board."?

 

[Dave Baumann]

I have been thinking about this for a long time. I truly believe that nVidia would actually cave if the Anandtechs/HARDOCP's/Toms/B3D's of the world simply flat out refused to use these drivers.

I've wanted to do a similar thing for a long time, but the issue comes with preview reference hardware - nobody would agree to waiting for official WHQL'ed drivers before reviewing. Unfortunatly though, its these previews that can give the largest misrepresentation of a product as they are conducted with limited time and its very difficult to catch anything.