http://www.xbitlabs.com/articles/video/display/geforce-fx.html
I haven't seen this posted in B3D before - so apologies if its old news.
I haven't seen this posted in B3D before - so apologies if its old news.
x-bitlabs post 23 page in-depth review of NV30
- Thread starter g__day
- Start date
It's certainly long, and mostly good, but I question their AF comparison. Why did they test such extremes for ATi? I think they should've added a setting of Performance (bilinear) and max quality textures, to compete with nV's Balanced setting. I don't think anyone would play at minimum quality on a $300+ video card, and it certainly doesn't compare fairly to Aggressive.
Yeah, it's detailed, but as far as "good", it's "good for being nvidia slanted".
For example, I think the pro-nvidia slant of the pixel pipeline discussion is quite good...compared to nvidia's label of "8x1". However, I ask you, what is the point of the bottome graph on page 7? It turns off color writes for the GF FX, then increases texture layers up to 4. That's fine for showing that the GF FX isn't affected by texture layer increases when color writes are off, but it includes the 9700 in that graph and then makes the case that the GF FX offers an advantage for Z only passes. Does anyone else maybe think that only the 0 texture part of the graph has any bearing on that "advantage" (the GF FX and 9700 effectively tie for the 0 and 1 texture cases EDIT: at the same clock speeds)?
And the pixel shader slot discussion at the beginning seems rather misleading, ignoring multipass and simply stating the GF FX can do things the R300 can't do because of length restrictions. It ignores that the GF FX's limit is not realtime, and for a gamer oriented review that is a pretty significant distortion, IMO.
The theme seems to be to slant things towards nvidia unless the numbers absolutely preclude presenting things in such a fashion, and if the numbers can support such a thing, to use them even if it requires abusing them. They don't pick and choose only the numbers that favor nvidia, though, but that distinction only seems "good" compared to people like Steve Spence.
Their 3dmark03 article was the same way, including aniso comparison problems, though the GF FX portion of aniso comparison in this article is, IMO, quite a significant improvement over anything I've noticed elsewhere...in fact the GF FX analysis seems pretty excellent, period. The problem comes when they turn down EVERY setting by hand for the 9700 and compare that to just selecting "Aggressive aniso" for the GF FX, and then use that to validate their decision to ignore anything other than that and "Quality" 9700 aniso when comparing performance and image quality with the GF FX. It even reads as if they don't know what the difference between "Performance" and "Quality" aniso is for the R300.
The vertex shader discussion seems to be quite a bit less distorted, but my thought is that is because the GF FX really does have demonstrated vertex processing advantages.
If you consider it valid for a site to pander to the vendor while doing a review, it does a very good job within that expectation.
For example, I think the pro-nvidia slant of the pixel pipeline discussion is quite good...compared to nvidia's label of "8x1". However, I ask you, what is the point of the bottome graph on page 7? It turns off color writes for the GF FX, then increases texture layers up to 4. That's fine for showing that the GF FX isn't affected by texture layer increases when color writes are off, but it includes the 9700 in that graph and then makes the case that the GF FX offers an advantage for Z only passes. Does anyone else maybe think that only the 0 texture part of the graph has any bearing on that "advantage" (the GF FX and 9700 effectively tie for the 0 and 1 texture cases EDIT: at the same clock speeds)?
And the pixel shader slot discussion at the beginning seems rather misleading, ignoring multipass and simply stating the GF FX can do things the R300 can't do because of length restrictions. It ignores that the GF FX's limit is not realtime, and for a gamer oriented review that is a pretty significant distortion, IMO.
The theme seems to be to slant things towards nvidia unless the numbers absolutely preclude presenting things in such a fashion, and if the numbers can support such a thing, to use them even if it requires abusing them. They don't pick and choose only the numbers that favor nvidia, though, but that distinction only seems "good" compared to people like Steve Spence.
Their 3dmark03 article was the same way, including aniso comparison problems, though the GF FX portion of aniso comparison in this article is, IMO, quite a significant improvement over anything I've noticed elsewhere...in fact the GF FX analysis seems pretty excellent, period. The problem comes when they turn down EVERY setting by hand for the 9700 and compare that to just selecting "Aggressive aniso" for the GF FX, and then use that to validate their decision to ignore anything other than that and "Quality" 9700 aniso when comparing performance and image quality with the GF FX. It even reads as if they don't know what the difference between "Performance" and "Quality" aniso is for the R300.
The vertex shader discussion seems to be quite a bit less distorted, but my thought is that is because the GF FX really does have demonstrated vertex processing advantages.
If you consider it valid for a site to pander to the vendor while doing a review, it does a very good job within that expectation.
OpenGL guy
Veteran
I have to disagree with you here. From http://www.xbitlabs.com/articles/video/display/geforce-fx_9.htmldemalion said:
http://www.xbitlabs.com/images/video/geforce-fx/vertex.gif
http://www.xbitlabs.com/images/video/geforce-fx/vertex-speed.gif
http://www.xbitlabs.com/images/video/geforce-fx/ragtroll.gif
The FX loses in every vertex shader test.
Hmm...I didn't say the GF FX beat the 9700 in vertex shading performance, I said it had advantages. That doesn't mean the 9700 doesn't have any itself.
Advantages include fixed function T&L performance and vertex shader instruction functionality. If you want to nitpick, I guess you could say fixed function T&L isn't vertex "shading", but I'd disagree with that considering the body of games and professional applications that exist, and the idea of looking at a GPU as consisting of a vertex and pixel processing portion.
If you guys at ATI want to completely hold the performance crown, you could maybe get out the R350 256MB refresh with boosted clock speeds.....
*sigh*, I'm so transparent.
Unless the R350 and/or new drivers already has some vertex processing performance and functionality enhancements I just didn't notice?
Advantages include fixed function T&L performance and vertex shader instruction functionality. If you want to nitpick, I guess you could say fixed function T&L isn't vertex "shading", but I'd disagree with that considering the body of games and professional applications that exist, and the idea of looking at a GPU as consisting of a vertex and pixel processing portion.
If you guys at ATI want to completely hold the performance crown, you could maybe get out the R350 256MB refresh with boosted clock speeds.....
*sigh*, I'm so transparent.Unless the R350 and/or new drivers already has some vertex processing performance and functionality enhancements I just didn't notice?
chavvdarrr
Veteran
?
Am I right?
In same time FX's "aggresive" although using really 'aggresive' aniso&trilinear combination (limited to mip-map boundary), does use trilinear. And IMHO 8xAgg gives better quality than 16xP - judging from seen shots of course.
AFAIK "performance" aniso is using bi-linear filtering only.demalion said:It even reads as if they don't know what the difference between "Performance" and "Quality" aniso is for the R300.
Am I right?
In same time FX's "aggresive" although using really 'aggresive' aniso&trilinear combination (limited to mip-map boundary), does use trilinear. And IMHO 8xAgg gives better quality than 16xP - judging from seen shots of course.
Re: ?
It appears to sample from two map levels over an extremely short interval, so discontinuities in the filtering are probably marginally less apparent than a bilinear filter, but only marginally.
This would certainly still be an improvement over a purely bilinearly filtered performance mode if the mipmap detail calculation were left alone, but at the same time the mipmap boundaries seem to be very different from the 'Application' mode - in shots that I've seen they appear to bulge towards the eye on inclined planes, instead of the expected behaviour where they would bend away - the effect of this is to move the transitions to lower mip-levels on these surfaces much closer to the eye than is usual, which will increase blurring. It appears from the shots that have been shown from test applications that the first and second mipmap transitions are moved significantly forward.
chavvdarrr said:AFAIK "performance" aniso is using bi-linear filtering only.demalion said:It even reads as if they don't know what the difference between "Performance" and "Quality" aniso is for the R300.
Am I right?
In same time FX's "aggresive" although using really 'aggresive' aniso&trilinear combination (limited to mip-map boundary), does use trilinear. And IMHO 8xAgg gives better quality than 16xP - judging from seen shots of course.
It appears to sample from two map levels over an extremely short interval, so discontinuities in the filtering are probably marginally less apparent than a bilinear filter, but only marginally.
This would certainly still be an improvement over a purely bilinearly filtered performance mode if the mipmap detail calculation were left alone, but at the same time the mipmap boundaries seem to be very different from the 'Application' mode - in shots that I've seen they appear to bulge towards the eye on inclined planes, instead of the expected behaviour where they would bend away - the effect of this is to move the transitions to lower mip-levels on these surfaces much closer to the eye than is usual, which will increase blurring. It appears from the shots that have been shown from test applications that the first and second mipmap transitions are moved significantly forward.
Re: ?
In the Serious Sam shot it did, but Serious Sam was using the "“GFX: Extreme Quality†add on. The ATI "Speed" shots weren't just Performance aniso, they were performance aniso with texture and mip map detail set to override to the lowest quality possible. The GF FX Aggressive seems to substitute mip map detail level for aniso processing, and mip map detail didn't seem to be lowered, atleast in Serious Sam with those settings. Were there any screen shot comparisons for a game besides Serious Sam? I'd look, but they seem to have omitted an index for the article, and I don't feel like rewarding that with more ad hits.
I'm not sure if the R300 gets a significant performance advantage for lowering the image quality so drastically...ATI seems to concentrate on full quality speed (That's even true on my R200, which doesn't even have the monster bandwidth of the R300).
There were lots of things wrong with the way the "Speed" comparison was done, IMO.
chavvdarrr said:AFAIK "performance" aniso is using bi-linear filtering only.demalion said:It even reads as if they don't know what the difference between "Performance" and "Quality" aniso is for the R300.
Am I right?
In same time FX's "aggresive" although using really 'aggresive' aniso&trilinear combination (limited to mip-map boundary), does use trilinear. And IMHO 8xAgg gives better quality than 16xP - judging from seen shots of course.
In the Serious Sam shot it did, but Serious Sam was using the "“GFX: Extreme Quality†add on. The ATI "Speed" shots weren't just Performance aniso, they were performance aniso with texture and mip map detail set to override to the lowest quality possible. The GF FX Aggressive seems to substitute mip map detail level for aniso processing, and mip map detail didn't seem to be lowered, atleast in Serious Sam with those settings. Were there any screen shot comparisons for a game besides Serious Sam? I'd look, but they seem to have omitted an index for the article, and I don't feel like rewarding that with more ad hits.
I'm not sure if the R300 gets a significant performance advantage for lowering the image quality so drastically...ATI seems to concentrate on full quality speed (That's even true on my R200, which doesn't even have the monster bandwidth of the R300).
There were lots of things wrong with the way the "Speed" comparison was done, IMO.
Similar threads
