Why does the Source DX8 codepath look so poor?
- Thread starter Fodder
- Start date
One thing I've noticed is one metric of "better" that seems like it is going to be haunting us for a while: "stands out more" = "better".
I think this is a result of the perspective of being in a transition from "low detail unreleastic" rendering to adding more detail...by looking especially for detail, finding it more prominently becomes a measure of quality by standing out more visibly from "low detail unrealistic" rendering.
What I think this results in here is a phenomenon where emphasis of 3D effects, in still screenshots especially, strikes the viewer as being "better" over more realistic and subtle effects, and the interaction of the lighting and realistic limitations in the visibility of details in objects aren't considered subjectively.
...
Looking at the gaps in between the tiles, the DX 8 shot seems to suffer by an over-emphasis of black areas and rougher depth, giving the appearance of high ridges in the grout casting a large second shadow in some of the tile gaps. This also holds true on the tile surface, which starts to diverge from an appearance of typical tile due to the shininess being at odds with the deep pitting and distortions that goes with it (such that the shininess seems like it should have gone as well) due to the coarseness presented in the shadowing.
It does look unusual, but it doesn't look grossly unrealistic for the most part due to actual variance you can expect in real tile and grouting. Given that, the more emphasized shadows seem to satisfy a desire to find evidence of progress in graphical detail, unless attention is paid to the (fairly faint) unrealistic step transitions to "grey" that occur in the full screenshot.
The only major problem in the DX 8 tile surface is that it doesn't fit my recollection of the sounds and behavior of objects undergoing physics and friction with the tile...it fits the more subtle and smoother tile in my mind's eye, but not the more deeply pitted, and apparently thicker, one.
...
For the stone detail, I do think that extra detail is a successful fit for the surface, and that it could even be classified when taken by itself as "better" in the small size preview shot present in the thread before you click on it to follow it. However, looking at the shots at the increased size after following them, there is nasty artifacting and unrealistic "ghosting" around the details that isn't present in the DX 9 rendering.
In fact, from experience, I know the DX 9 shot, even with its less apparent detail in the shot, is detail rich while being, in my evaluation, more realistic (for the given surface detail resolution). This is because the detail isn't fully apparent in naked daylight illumination from above, but can be made more apparent by turning on the (more sophisticated of the two available) flashlight, and viewing the impact on the surface of its lighting.
The problem is that while such subtleties are more realistic (a pitted rock of fairly consistent coloration with such shallow and wide pits would tend to look less detailed in bright daylight, with details standing out when lighting is provided from a specific direction, plus it would have sand in the pits to make them more subtle), they are also less visible, noticeable, and convey a "lack of detail" until realistic circumstances bring the detail out (like cleaning out sand, changing flashlight position, etc).
...
What I wonder is when the evaluation will shift from looking for more universally evident detail (as I feel drove things like Far Cry's shiny pipes, the original HL2 roof video, and the details in the Source VST), to how realistically details are and are not made evident to the viewer? One thing I note is that subtle detail would benefit from higher resolution textures in magnification...is it too soon for Valve to (apparently) be trying this type of subtlety for some surfaces, or is it just that still screenshot evaluation is leading to an early verdict in the absence of the full atmosphere in which the surfaces occur?
I think this is a result of the perspective of being in a transition from "low detail unreleastic" rendering to adding more detail...by looking especially for detail, finding it more prominently becomes a measure of quality by standing out more visibly from "low detail unrealistic" rendering.
What I think this results in here is a phenomenon where emphasis of 3D effects, in still screenshots especially, strikes the viewer as being "better" over more realistic and subtle effects, and the interaction of the lighting and realistic limitations in the visibility of details in objects aren't considered subjectively.
...
Looking at the gaps in between the tiles, the DX 8 shot seems to suffer by an over-emphasis of black areas and rougher depth, giving the appearance of high ridges in the grout casting a large second shadow in some of the tile gaps. This also holds true on the tile surface, which starts to diverge from an appearance of typical tile due to the shininess being at odds with the deep pitting and distortions that goes with it (such that the shininess seems like it should have gone as well) due to the coarseness presented in the shadowing.
It does look unusual, but it doesn't look grossly unrealistic for the most part due to actual variance you can expect in real tile and grouting. Given that, the more emphasized shadows seem to satisfy a desire to find evidence of progress in graphical detail, unless attention is paid to the (fairly faint) unrealistic step transitions to "grey" that occur in the full screenshot.
The only major problem in the DX 8 tile surface is that it doesn't fit my recollection of the sounds and behavior of objects undergoing physics and friction with the tile...it fits the more subtle and smoother tile in my mind's eye, but not the more deeply pitted, and apparently thicker, one.
...
For the stone detail, I do think that extra detail is a successful fit for the surface, and that it could even be classified when taken by itself as "better" in the small size preview shot present in the thread before you click on it to follow it. However, looking at the shots at the increased size after following them, there is nasty artifacting and unrealistic "ghosting" around the details that isn't present in the DX 9 rendering.
In fact, from experience, I know the DX 9 shot, even with its less apparent detail in the shot, is detail rich while being, in my evaluation, more realistic (for the given surface detail resolution). This is because the detail isn't fully apparent in naked daylight illumination from above, but can be made more apparent by turning on the (more sophisticated of the two available) flashlight, and viewing the impact on the surface of its lighting.
The problem is that while such subtleties are more realistic (a pitted rock of fairly consistent coloration with such shallow and wide pits would tend to look less detailed in bright daylight, with details standing out when lighting is provided from a specific direction, plus it would have sand in the pits to make them more subtle), they are also less visible, noticeable, and convey a "lack of detail" until realistic circumstances bring the detail out (like cleaning out sand, changing flashlight position, etc).
...
What I wonder is when the evaluation will shift from looking for more universally evident detail (as I feel drove things like Far Cry's shiny pipes, the original HL2 roof video, and the details in the Source VST), to how realistically details are and are not made evident to the viewer? One thing I note is that subtle detail would benefit from higher resolution textures in magnification...is it too soon for Valve to (apparently) be trying this type of subtlety for some surfaces, or is it just that still screenshot evaluation is leading to an early verdict in the absence of the full atmosphere in which the surfaces occur?
It seems to me the effects in the DX8 path are more exaggerated in terms of contrast and depth. They look bump-mapped, where as the DX9 path is more subtle and realistic. I think it's going to be hard to say which looks better until you see these effects in context; if they blend in with the textures and general look of the game then they'll be fine. However, they may also stick out like a sore thumb.
Overall, though, I'm sure I could play the game with either path and enjoy it equally (which I am sure is the case for the vast majority of gamers).
Overall, though, I'm sure I could play the game with either path and enjoy it equally (which I am sure is the case for the vast majority of gamers).
overclocked
Veteran
Very good post demalion.
For another relevant issue with the DX8.1/DX9 path´s where i think most
NV35 owner´s mind is whas that you want the option of the DX9 path but
that "first" in the VST you couldn´t force the path but later there was an easy way to run it(was it Firingsquad?). If i remember nearly accurate the framerate dropped from 100fps to 50fps give or take on a FX5950.
If VST is an extreme test on shader´s and represant of HL2 i think if you
will you can play the DX9 path pretty good at 1024*768, and/or if you think the speedloss is not worth the extra quality you can throw in little AA/AF and be fine. just my thought´s and it depends on as we know how the final game´s different path´s in quality turn out to be. Either way Valve has done a great path for the FX series as it seem´s and it´s always nice to have the option.
For another relevant issue with the DX8.1/DX9 path´s where i think most
NV35 owner´s mind is whas that you want the option of the DX9 path but
that "first" in the VST you couldn´t force the path but later there was an easy way to run it(was it Firingsquad?). If i remember nearly accurate the framerate dropped from 100fps to 50fps give or take on a FX5950.
If VST is an extreme test on shader´s and represant of HL2 i think if you
will you can play the DX9 path pretty good at 1024*768, and/or if you think the speedloss is not worth the extra quality you can throw in little AA/AF and be fine. just my thought´s and it depends on as we know how the final game´s different path´s in quality turn out to be. Either way Valve has done a great path for the FX series as it seem´s and it´s always nice to have the option.
To me the question itself is akin to also being baffled at rendering differences between DX7-8 graphics programming, or between DX6-7 graphics. I thought it was pretty much understood that advances in the API structures when it comes to 3d are done to increase the rendering realism of 3d at acceptable performance levels--otherwise, what would be the sense of establishing a newer API version over an older one or buying newer hardware? I hardly think it suprising that such rendering differences are observable--unless you want to proceed from the unlikely supposition that DX8 was already "perfect" and cannot be improved upon....
Of course, some developers will overdo some of the newer rendering capabilities, such as "shiny pipes" everywhere when obviously "shiny" isn't needed everywhere, but only needed occasionally, just as some games will overdo the "rubbery" look simply because it's new, or overdo HDR, or motion blur, or what have you. But those are developer issues pertinent to specific development efforts as opposed to overall commentaries on API versions, imo.
Of course, some developers will overdo some of the newer rendering capabilities, such as "shiny pipes" everywhere when obviously "shiny" isn't needed everywhere, but only needed occasionally, just as some games will overdo the "rubbery" look simply because it's new, or overdo HDR, or motion blur, or what have you. But those are developer issues pertinent to specific development efforts as opposed to overall commentaries on API versions, imo.
IMO, the DX8 looks terrible. The contrast is far too high and looks nothing like anything real.
Colourless, yeah. I asked Gabe and he said that for the DX9 renders, his mother had an influence (she forced him to scrub the floors).
On a more serious note, and I have not "studied" those screens like some of you had, the first thing I thought of was some exaggerated LOD bias for the DX8 renders. Which really has little to do with "DX8 vs DX9" of course but there you go when it comes to my first impression.
[edit]In the first set of screens, we've got white-vs-black (DX9-vs-DX8) for the stuff between tiles. Don't know about you guys, but in my home, I use "white stuff" tiling the floor.
On a more serious note, and I have not "studied" those screens like some of you had, the first thing I thought of was some exaggerated LOD bias for the DX8 renders. Which really has little to do with "DX8 vs DX9" of course but there you go when it comes to my first impression.
[edit]In the first set of screens, we've got white-vs-black (DX9-vs-DX8) for the stuff between tiles. Don't know about you guys, but in my home, I use "white stuff" tiling the floor.
I agree with Demalion! (Probably, I might have gotten lost somwhere.
)I just can't understand how someone can think the DX8 wood/stones/whatever it is (I can't tell from either of the pictures) looks better. It's covered with horrible artifacts that looks like someone has gone mad with edge sharpening and "oil painting" filters.
Or maybe I do understand that people think it's better, just not why.
It might be a similar reason as a friend gave why he thought his TV had so much better colors. I couldn't understand what he meant until he showed that he always ran it with chrominance turned up to max, making everyone look like they've eaten a couple of habaneros. Hey lookie, bright colors!
Here's a hint: more isn't nessesarily better, the right amount is best.
1. "Flagstones" != wood.
2. You're all crazy.
Seriously, how can so many of you think DX8 looks better? Is it less than ideal viewing conditions (old monitors, bad lighting/brightness/contrast)? It just looks like past a certain brightness the "highlights" default to black. Blech. Yeah, the DX9 tiles look like someone spilled the Vaseline, and they also look slightly blurred relative to the DX8 path, but the latter's liberal black errors don't so much look like greater detail than ... well, errors. It's like some 3D effect gone horribly wrong, with the shifted red and green colors shunted to black.
Is the blurriness more impressive in motion? Is that also why the DX9
"bumps" look shallower? Is this a compromise for older cards that might not support higher-res bump maps (chunky/obvious maps for older cards, blurred to give the appearance of more subtle detail on DX9 cards)?
2. You're all crazy.
Seriously, how can so many of you think DX8 looks better? Is it less than ideal viewing conditions (old monitors, bad lighting/brightness/contrast)? It just looks like past a certain brightness the "highlights" default to black. Blech. Yeah, the DX9 tiles look like someone spilled the Vaseline, and they also look slightly blurred relative to the DX8 path, but the latter's liberal black errors don't so much look like greater detail than ... well, errors. It's like some 3D effect gone horribly wrong, with the shifted red and green colors shunted to black.
Is the blurriness more impressive in motion? Is that also why the DX9
"bumps" look shallower? Is this a compromise for older cards that might not support higher-res bump maps (chunky/obvious maps for older cards, blurred to give the appearance of more subtle detail on DX9 cards)?
Spaceman-Spiff
Regular
More screenies: http://www.firingsquad.com/hardware/half_life_2_fx/page2.asp (compares DX 7, 8.0, 8.1, and 9)
.Similar threads
