Albatron GeForceFX 5200 Ultra Review

[First]

Originally posted by: DaveBaumann

Originally posted by: Evan Lieb

In any case, I think one of NVIDIA's engineers mentioned to me a while back that trilinear buffering is disabled in certain games (this could apply to certain levels in UT2K3) as it'll significantly degrade fps for no good reason (i.e. no IQ gain). Though I'll have to confirm that info.

Trilinear Buffering?

Yeah, I meant to say filtering. Thanks?

Originally posted by: DaveBaumann
But anyway, regarless - you think its OK to just arbitarily remove Trilinear filtering at will? You did not question this when you were told it? Did you actually investigate any of the maps in UT2003 to see if it was noticable or not? Did you solicit opinions from the developers to ask if they thought it was OK to alter the IQ from which they selected for their application?

I was merely stating what I remembered an NVIDIA engineer telling me a few months ago. Since I cannot remember exactly what was said I will have to call up NVIDIA soon and ask them point blank about trilinear filtering in UT2K3. And to answer your question, yes, I have compared multiple UT2K3 maps (with a 5900 Ultra box and a 9800 Pro box standing side by side, Quality mode, various resolutions) and noticed very tiny variances in overall IQ.

I?ll ask a few developers some questions about this, good idea.

Originally posted by: DaveBaumann
Come on, really, this is 2003 - Trilinear filtering has been here for years (and even for Free on NV10, in terms of fill-rate), should we really be accepting that its just turned off on these new "uber boards" being released which are several generations on from those boards that initially offered it?

I agree, it shouldn?t be expected. But perhaps there?s something you or I don?t know about that makes ?smart? trilinear filtering in UT2K3 useful. And perhaps NVIDIA is just trying to cut a few corners to get higher frame rates. Having seen infinitesimal IQ variance with identically configured 9800 Pro and 5900 Ultra boxes in UT2K3, I really don?t think this is too big a deal. Of course, perhaps I?ve missed something and this ?smart? trilinear filtering (if it even exists, I?ll have to ask NVIDIA and do some independent testing) is reproducible in various maps.

Overall, unless I (or someone) can actually reproduce a scenario (from a gamer's perspective) in UT2K3 where the 44.03 drivers are degrading IQ for the sake of higher fps in UT2K3, this is hardly all that important, though certainly worth some investigation.

 

[Jeff7181]

Originally posted by: Evan Lieb

Originally posted by: shady06
planning on including an overclocking section?

Of course.

I sure hope so... these...

NVIDIA's Back with NV35 - GeForceFX 5900 Ultra
ATI's Radeon 9600 Pro - Another Win for ATI
NVIDIA's GeForce FX 5600/5200 Ultra Performance Review
NVIDIA GeForce FX 5800 Ultra: It's Here, but is it Good? ATI's Radeon 9500 Pro - Now Shipping & Faster
etc. etc.

... were highly disappointing in that respect.

 

[First]

Originally posted by: Jeff7181

Originally posted by: Evan Lieb

Originally posted by: shady06
planning on including an overclocking section?

Of course.

I sure hope so... these...

NVIDIA's Back with NV35 - GeForceFX 5900 Ultra
ATI's Radeon 9600 Pro - Another Win for ATI
NVIDIA's GeForce FX 5600/5200 Ultra Performance Review
NVIDIA GeForce FX 5800 Ultra: It's Here, but is it Good? ATI's Radeon 9500 Pro - Now Shipping & Faster
etc. etc.

... were highly disappointing in that respect.

As with the vast majority of "technology" or "first look" reviews, overclocking is fun but ultimately fairly pointless. For example, if we received a reference VIA PT800 motherboard Anand would write up a review on the chipset itself, with some benchmarks and commentary. He wouldn't perform any overclocking because said chipset is not yet in retail.

Therefore, for individual video card reviews, core clock and memory clock overclocking will be conducted, because all "individual" reviews deal with retail products.

 

[DaveBaumann]

Overall, unless I (or someone) can actually reproduce a scenario (from a gamer's perspective) in UT2K3 where the 44.03 drivers are degrading IQ for the sake of higher fps in UT2K3, this is hardly all that important, though certainly worth some investigation.

UT2003 has two different levels of texture map - the default textures (low resolution) and the detail textures (high resolution), these two two are mixed into the maps. What we've seen is that neither of them have full trilinear applied, but the detail textures have even less - close to Bilinear, or equivelent to their Fasted Performance texture mode. So, all you need to do is look at an area with the high detail textures in use.

http://www.beyond3d.com/forum/viewtopic.php?t=6719&postdays=0&postorder=asc&start=20

Those images there were taken from a map relatively at random, and you can see that the left hand section (the barred floor) uses a detail texture and the mip map transistions are reasonably noticable - they would be moreso in motion.

 

[Hanners]

Originally posted by: Evan Lieb Overall, unless I (or someone) can actually reproduce a scenario (from a gamer's perspective) in UT2K3 where the 44.03 drivers are degrading IQ for the sake of higher fps in UT2K3, this is hardly all that important, though certainly worth some investigation.

You'll at least be mentioning that nVidia disables trilinear filering for UT2003 in the text of the review, right?

On another technical note, if you are doing any Splinter Cell benchmarks, make sure you don't forget about the AA issues.

 

[Jeff7181]

As with the vast majority of "technology" or "first look" reviews, overclocking is fun but ultimately fairly pointless.

I don't know if I'd agree with that... I realize overclocking ability varies from card to card and isn't totally dependant on the GPU. It does give at least some indication of how much headroom the GPU has. An example of that is the 9600 Pro... it's highly overclockable, which would have been nice to know. Now if only a manufacturer would give one of them some really good RAM like Asus did with the V8420S. An overclocked 9600 Pro could probably give a 9700 Pro a run for it's money despite it's pipeline handicap if the RAM was capable of 900+ Mhz.

 

[BrentJ]

Originally posted by: Evan Lieb

I agree, it shouldn?t be expected. But perhaps there?s something you or I don?t know about that makes ?smart? trilinear filtering in UT2K3 useful. And perhaps NVIDIA is just trying to cut a few corners to get higher frame rates.

So you are saying that its ok to cut corners to gain fps? At what cost, IQ?

 

[First]

Originally posted by: DaveBaumann

UT2003 has two different levels of texture map - the default textures (low resolution) and the detail textures (high resolution), these two two are mixed into the maps. What we've seen is that neither of them have full trilinear applied, but the detail textures have even less - close to Bilinear, or equivelent to their Fasted Performance texture mode. So, all you need to do is look at an area with the high detail textures in use.

http://www.beyond3d.com/forum/viewtopic.php?t=6719&postdays=0&postorder=asc&start=20

Those images there were taken from a map relatively at random, and you can see that the left hand section (the barred floor) uses a detail texture and the mip map transistions are reasonably noticable - they would be moreso in motion.

Great find, it's definitely noticable. Now, the only question is if it's noticable in actual in-game experience.

Originally posted by: BrentJ

Originally posted by: Evan Lieb

I agree, it shouldn?t be expected. But perhaps there?s something you or I don?t know about that makes ?smart? trilinear filtering in UT2K3 useful. And perhaps NVIDIA is just trying to cut a few corners to get higher frame rates.

So you are saying that its ok to cut corners to gain fps? At what cost, IQ?

No, I'm saying that as long as IQ is not sacrified in any noticable way, optimizing for fps gain is not a bad thing.

Originally posted by: Hanners

You'll at least be mentioning that nVidia disables trilinear filering for UT2003 in the text of the review, right?

Yes, but first I'll have to see what NVIDIA has to say about all of this.

Originally posted by: Hanners
On another technical note, if you are doing any Splinter Cell benchmarks, make sure you don't forget about the AA issues.

We won't even be running SC with AA enabled, so yes we will mention the issue.

 

[First]

Looks like HardOCP officially puts this UT2K3 trilinear filtering issue to rest. As usual, Brent and Kyle do an excellent job outlining the IQ differences.

 

[fell8]

It may be too late for this, but I would love to see some minimum FPS rates as I feel it better reveals GPU limitations and performance when truely taxed.

I would also very much like to see some 5600/9600 numbers thrown in, mostly because nobody seems to be doing reviews on these cards (while you're at it, maybe some solid info on what exactly is the differernce between the 5600 & 5600U, I mean is it just the core/mem clock speeds' or is there further castration like with the 5200 & 5200U?).

Thanks!

Doug

 

[Dean]

Originally posted by: Evan Lieb
Looks like HardOCP officially puts this UT2K3 trilinear filtering issue to rest. As usual, Brent and Kyle do an excellent job outlining the IQ differences.

Put it to rest in what way? They basically said there is no noticable difference when playing the game UT2K3 between Bilinear and Trilinear. Ok that is fine(that is their opinion), but they are still benching the nvidia card using Bilinear Filtering and the ATI card using Trilinear Filtering.

If you are going to compare two video cards you have to bench them in an apples to apples comparison. If you want the ATI cards to use Bilinear you will have to bench them in PERFORMANCE mode to achieve Bilinear filtering.

They also made no mention of games where bilinear mipmap transitions are hugely noticable like Star Trek Elite Force II. Nvidia is NOT forcing trilinear in their control panel mainly to achieve higher FPS while ATI is forcing trilinear. They then compare the two card with their respective FPS while the ATI card is being FORCED to do more work than their Nvidia counterpart.


Since Brent and Kyle are saying Bilinear is ok I fully expect them and anyone who agrees with them to Bench the ATI cards in PERFORMANCE mode. Why bench the ATI in quality and force it to lose FPS if the difference between bilinear and trilinear is not noticable? Their article is nothing more than a defensive posture taken because ATI questioned the way they are benchmarking.

 

[BrentJ]

Originally posted by: Dean

Originally posted by: Evan Lieb
Looks like HardOCP officially puts this UT2K3 trilinear filtering issue to rest. As usual, Brent and Kyle do an excellent job outlining the IQ differences.

Put it to rest in what way? They basically said there is no noticable difference when playing the game UT2K3 between Bilinear and Trilinear. Ok that is fine(that is their opinion), but they are still benching the nvidia card using Bilinear Filtering and the ATI card using Trilinear Filtering.

If you are going to compare two video cards you have to bench them in an apples to apples comparison. If you want the ATI cards to use Bilinear you will have to bench them in PERFORMANCE mode to achieve Bilinear filtering.

They also made no mention of games where bilinear mipmap transitions are hugely noticable like Star Trek Elite Force II. Nvidia is NOT forcing trilinear in their control panel mainly to achieve higher FPS while ATI is forcing trilinear. They then compare the two card with their respective FPS while the ATI card is being FORCED to do more work than their Nvidia counterpart.


Since Brent and Kyle are saying Bilinear is ok I fully expect them and anyone who agrees with them to Bench the ATI cards in PERFORMANCE mode. Why bench the ATI in quality and force it to lose FPS if the difference between bilinear and trilinear is not noticable? Their article is nothing more than a defensive posture taken because ATI questioned the way they are benchmarking.

With AF enabled both cards use their own algorithim and are not comparable exactly apples-to-apples.

The main point was to look at Non-AF Trilinear as that was the question at the forefront. And in that situation no IQ differences were noted with the mip-map boundaries, and in fact the 5900u has a slightly sharper texture with no AF. Now, with AF the 9800 Pro has a sharper texture as was seen in the screenshots.

 

[Dean]

Well Brent did you test the ATI card with PERFORMANCE set in the control panel to force Bilinear? The ATI card is still taking a performance hit using trilinear while the Nvidia card is not. Even though you say FPS is unimportant you still managed to post the numbers in your original review which creates an illusion for your readers.

 

[BrentJ]

Originally posted by: Dean
Well Brent did you test the ATI card with PERFORMANCE set in the control panel to force Bilinear? The ATI card is still taking a performance hit using trilinear while the Nvidia card is not. Even though you say FPS is unimportant you still managed to post the numbers in your original review which creates an illusion for your readers.

Why would set the card to Performance AF?

The main point was to see its default Non-AF Trilinear quality, as that was the main question brought up. Default on the ATI control panel is Application Preference, and on the NVIDIA control panel is Quality.

At those default settings there is no noticeable mip-map boundary differences, and in fact the 5900u has a SLIGHTLY sharper texture.

8XAF was thrown in there as a secondary measure to see if there were any differences with AF quality.

That is the order I went in with the article.

 

[Rand]

I was left with the impression the article was more of a justification of their previous stance then any distinct effort to seperate the image quality provides by the respective boards.

In the first comparative image utilized the scene has relatively poor contrast throughout with very dark textures layered on one another.
You state that there is virtually no difference in the images rendered by the graphics cards, but you've chosen a scene that would tend to minimize any differences there may have been.

Many of the scenes chosen to compare follow the same trend of being those in which it would be most difficult to individually detect a mip-map boundary, whether due to lighting, and the selection of contrasting colors or excessively complex scenes.


With all due respect this was not in any way one of Brent Justice's better articles in my mind.
For the most part I've come to consider him one of the better graphics card reviewers, but this article seemed designed to provide a selection of images specifically chosen to ensure the least likelyhood of easy differentiation in image quality.

I don't mean to imply Mr. Justice is intentionally being misleading as I've seen few overt signs of such in the past, and as I stated previously I do in general quite enjoy his work.
In this instance however I believe he has made a very poor choice in terms of selection of images to compare.


It seems relatively clear their are some rather harsh transitions in mip-map boundaries in the images rendered by the GeForceFX.
There is doubtlessly a degradation in image quality, now whether or not the difference is easily visible is debateable.

The end result in my mind is that a notation should most definitely be inclueded in any article benchmarking the GF FX that proper trilinear filtering is not being done, and as such will degrade image quality to some extent. Such a statement would then be followed by the individual others opinion as to whether the degredation is clearly visible during active gameplay.

 

[Dean]

Its becoming pretty sad these days when reviewers have to try to justify an IHV lowering image quality(noticable or not) on $400 and $500 video cards so they can have higher FPS. This whole thing stinks and throws "Cinematic Quality" out the window. Reviews should be demanding the highest of IQ on top end cards.

 

[hjs]

So if the FX doens't do Tri-linear in UT 2003, test the Radeon 9800 also at bi-linear instead of tri-linear.
Seem more then fair to me.
2 bars, one the Radeon at Bi and one the Radeon at Tri in youre graphs against a FX bar in only Bi. It's a simple
and some screen shots of the difference in IQ

 

[hjs]

Originally posted by: BrentJ
The main point was to see its default Non-AF Trilinear quality, as that was the main question brought up. Default on the ATI control panel is Application Preference, and on the NVIDIA control panel is Quality.

If even the IQ is the same, the speed won't be, and all of us read the FPS they reach in the test.
and if because of this the FX get a lot more FPS the test is unfair.
And how is the IQ of the R9800 in bi-linear ?

 

[Pete]

Brent, thanks for the article. I don't so much care about still image comparison as I do about how the two cards compare in motion, and you indicated you saw no such difference, which I'll accept. But now I'm curious: if nV's "barely there" (per colored MIP-map shots) trilinear was indistinguishable from ATi's more "obvious" trilinear, would there be an IQ difference between actual bilinear and trilinear? If not, would you consider ditching Quality AF numbers for UT2K3? There's no sense in wasting GPU horsepower on a feature that seems to offer no appreciable benefit.

Edit: I also don't see what texture "sharpness" has to do with the issue at hand. It's an extraneous fact, IMO polluting the main investigation.

 

[Mingon]

A bit late in the day i Know, but what about comparing generation differences via identical clock / memory speeds. This would be interesting as it would show were nvdia improved each series or reduced unneeded capabilities. e.g. a ti500 gef3 had (supposedly) larger cache sizes compared to ti4200 which in turn had twice the vertex power.

 

[Rand]

Another analysis of AF in UT2003 on the GF FX

This time it's done by AMDMB.com
Conclusion was the opposite of that stated by HardOCP.
BTW, does anyone know what happened to the HardOCP article? It seems to have been removed.

Havent read the AMDMB article yet so I can't comment, though I hope it's better then the rather poor (IMO) presentation on HardOCP.

 

[Pete]

Kyle said that it's been temporarily removed due to server problems. AMDMB's article is somewhat different, in that they expose the fact that the 44.03 drivers appears to not AF every texture. That may be a bug, though, and nV has obviously subsequently released many newer drivers.

 

[BrentJ]

Originally posted by: Pete
Brent, thanks for the article. I don't so much care about still image comparison as I do about how the two cards compare in motion, and you indicated you saw no such difference, which I'll accept. But now I'm curious: if nV's "barely there" (per colored MIP-map shots) trilinear was indistinguishable from ATi's more "obvious" trilinear, would there be an IQ difference between actual bilinear and trilinear? If not, would you consider ditching Quality AF numbers for UT2K3? There's no sense in wasting GPU horsepower on a feature that seems to offer no appreciable benefit.

Edit: I also don't see what texture "sharpness" has to do with the issue at hand. It's an extraneous fact, IMO polluting the main investigation.

There is a noticeable difference between Bilinear AF on ATI's cards and "Quality" AF on nvidia's cards, the bilinear AF looks worse than nvidia's current quality setting, the banding is noticeable, so that comparison really wouldn't work

NVIDIA's Quality setting in ut2k3 seems to be somewhere between Bi and Tri filtering

 

[Pete]

Thanks, Brent. I'm guessing nVidia's Performance AF also shows banding? If so, how does it compare to ATi's Performance AF?

 

[BrentJ]

Originally posted by: Pete
Thanks, Brent. I'm guessing nVidia's Performance AF also shows banding? If so, how does it compare to ATi's Performance AF?

I dunno, haven't compared performance af