Prey demo on Xbox Live Marketplace

[BenSkywalker]

Of course I am talking about edge aliasing, that's the subject of the conversation. Personally I think you are trying to obfuscate the issue here.

Not at all. Why I mentioned that was that the majority of what appears to be 'edges' in this game is from normal maps, there are only a very small fraction of actual geometric edges compared to what it appears there is on screen.

If there are no blended near horizontal or vertical edges, that says more to me than analysing ~45 degree edges which are significantly more difficult to analyse given the analog filtered screenshots we have.

Analog filtering doesn't care about what degrees the angles are, they are all handled the same(at least as far as what we are dealing with).

A 2x1 OG will blend vertical edges much better than it will blend near horizontal ones (if even not at all), vice versa with 1x2. The statement you made in determining that OG patterns were being used was that it was "having trouble with near vertical and near horizontal edges." At the very least, OG will provide adequate blending for one of those, depending on the orientation of the pattern surely?

Yes, yes and no. No matter which setup you go with they will handle either near vertical or near horizontal better then the other, but for either one the optimal angles are 45, 135, 225 and 315. Either 2x1 or 1x2 OG is less then ideal for handling near vertical OR near horizontal. The reason is if the sampling positions are slightly offset to the edge then you will end up 'missing' the blend where it is needed. With the near 45 degree angles, you nigh can't miss them(relatively speaking obviously).

If these are not blended (as seen clearly in these shots and in game) then it is hard to find any implementation except reaching for the exotic.

They are blended clearly on certain angles.

http://media.teamxbox.com/games/ss/1172/full-res/1150390224.jpg

There is a better screenshot. This one it appears that it is handling near horizontal angles better then near vertical.

Kyle- I don't see any at all, I get a page not available message

 

[Mmmkay]

Analog filtering doesn't care about what degrees the angles are, they are all handled the same(at least as far as what we are dealing with).

I was just making a point that these screenshots are somewhat 'fuzzy' thanks to being captured from the analog output from the 360 as opposed to a digital framebuffer. Although on second thought, it's neither here nor there so ignore that.

Yes, yes and no. No matter which setup you go with they will handle either near vertical or near horizontal better then the other, but for either one the optimal angles are 45, 135, 225 and 315. Either 2x1 or 1x2 OG is less then ideal for handling near vertical OR near horizontal. The reason is if the sampling positions are slightly offset to the edge then you will end up 'missing' the blend where it is needed. With the near 45 degree angles, you nigh can't miss them(relatively speaking obviously).

Right so when we have screenshots which have unblended edges at both near horizontal and near vertical angles, how can it be consistent with an OG pattern? Take this screenshot for example:

http://xbox360media.ign.com/xbox360/image/article/715/715455/prey-20060628052213837.jpg
The near horizontal edges visible in the rock formations are unblended and the near vertical edges on the statues and grandfather are also unblended.

And then there is this one...
http://xbox360media.ign.com/xbox360/image/article/715/715455/prey-20060628052207416.jpg

The one you posted, I'm sorry but I can't spot any blended edges so you'll have to highlight them for me.

 

[Farid]

They are blended clearly on certain angles.

http://media.teamxbox.com/games/ss/1172/full-res/1150390224.jpg

There is a better screenshot. This one it appears that it is handling near horizontal angles better then near vertical.

This screenshot resolution is 1600x1200. So that's surely from the PC.

 

[Bad_Boy]

Rumor has it, the game is pretty short, 4-5 hours. Any truth to that?
Would be dissapointing after the pretty good demo.

 

[kyleb]

Kyle- I don't see any at all, I get a page not available message

Doh, I got carried away cleaning my FTP space a few hours after I uploaded that, but I put it back up. So what do you think is smoothing the edges on the crowbar and such here?

 

[TheChefO]

Doh, I got carried away cleaning my FTP space a few hours after I uploaded that, but I put it back up. So what do you think is smoothing the edges on the crowbar and such here?

The front of the cash register in that shot clearly shows no AA. Other areas are questionable. The top of the fridges appear to have a broken reverse AA where they actually amplify the edges. Nice technique really

 

[Subtlesnake]

Pssh, that's rather hard to sallow no matter who it comes from.

I'm frankly still extremely weirded out at why the Xbox 360 demo is so freaking huge, while the PC demo is tiny, even in comparison to other recent demo's.

Why reduce the textures? Seriously, you're just hurting you're GREATLY by doing something like that, then even more so by not mentioning it to those who download it.

He posted an image from the full game:

http://forums.3drealms.com/vb/attachment.php?attachmentid=5323&d=1152288465

Then why'd the Xbox360 download be more than twice the size of the PC download?

To minimise seek time on the DVD, data on the 360 is organised into three PAK files for each level which contain all the models, textures and sound data for that level (those loaded at the start and all those which might be needed during the level, as we also have to pre-cache everything on 360 to avoid stalls caused by mid-level loads).

This means three DVD seeks instead of hundreds when loading a level (seeks are expensive time-wise) and was a necessary change to bring the load time down under 45 seconds- it was around 5 minutes a level from DVD before this was done! However it means textures, sounds and models are stored again in the PAKs for any levels they are needed for."

http://forums.3drealms.com/vb/showthread.php?t=19012&

 

[Skrying]

Thanks for the links SubtleSnake, that goes a long way in explaining it, and actually does make sense.

I'm still of the thought that it would have made a lot more sense for a shorter demo with the larger textures, or just a flat out larger download with the larger textures. They really hurt themselves by not going with the full textures, stupid stupid mistake.

But here's my question, wouldnt the larger textures hurt performance even more? From what others are saying its already not to great.

Oh, and a load time of 45 seconds? I'm glad I'll be playing this game on a PC.

 

[Jawed]

To minimise seek time on the DVD, data on the 360 is organised into three PAK files for each level which contain all the models, textures and sound data for that level (those loaded at the start and all those which might be needed during the level, as we also have to pre-cache everything on 360 to avoid stalls caused by mid-level loads).

This means three DVD seeks instead of hundreds when loading a level (seeks are expensive time-wise) and was a necessary change to bring the load time down under 45 seconds- it was around 5 minutes a level from DVD before this was done! However it means textures, sounds and models are stored again in the PAKs for any levels they are needed for."

http://forums.3drealms.com/vb/showthread.php?t=19012&

That needs its own thread. That really is a big deal for all sorts of reasons.

Jawed

 

[swaaye]

Sounds like they are running up against problems caused by streaming data from the DVD instead of from a hard disk. Seek times on an optical drive are rather shockingly horrible compared to the already horrible hard drive.

And as for whether bigger textures would hurt performance; that depends on whether the game is graphics or CPU limited. I'm learning towards CPU, personally. Of course, available RAM could also be a problem. The high graphics setting on PC suggests a 256 MB graphics card. Xbox 360 has 512 MB RAM, shared, so maybe there's just not enough RAM for the bigger textures? I haven't checked to see how much system RAM the game uses when running on PC.

 

[Jawed]

Sounds like they are running up against problems caused by streaming data from the DVD instead of from a hard disk. Seek times on an optical drive are rather shockingly horrible compared to the already horrible hard drive.

And Prey/Q4 are the first games that use the PAK format for such a large amount of console game data

This perhaps implies that PAK has reached the end of the road and perhaps it's another reason why JC is so keen on megatexture, which I guess is more explicitly designed for streaming and will work better off DVD.

And as for whether bigger textures would hurt performance; that depends on whether the game is graphics or CPU limited. I'm learning towards CPU, personally. Of course, available RAM could also be a problem. The high graphics setting on PC suggests a 256 MB graphics card. Xbox 360 has 512 MB RAM, shared, so maybe there's just not enough RAM for the bigger textures? I haven't checked to see how much system RAM the game uses when running on PC.

I think it's prolly RAM.

First indications are that performance of Prey on PC is much lower than Q4:

http://www.xbitlabs.com/articles/video/display/powercolor-x1900gt_13.html

but Prey doesn't have built in Timedemo capabilities (so they used FRAPs) and is obviously very new so drivers aren't necessarily up to scratch.

Jawed

 

[Richard]

This perhaps implies that PAK has reached the end of the road and perhaps it's another reason why JC is so keen on megatexture, which I guess is more explicitly designed for streaming and will work better off DVD.

The PAK "format" v3 and upwards (used from quake 3 onwards) is simply renamed .ZIPs. Where megatexture helps consoles is a reduction on texture memory consumption, all things beings equal.

but Prey doesn't have built in Timedemo capabilities (so they used FRAPs) and is obviously very new so drivers aren't necessarily up to scratch.

Prey does have timedemo capabilities, that's what I used here: http://www.beyond3d.com/forum/showpost.php?p=782433&postcount=4

Ratchet also posted some Prey performance review using timedemos : http://www.rage3d.com/articles/preydemoperf/

Having said that, I agree with you that memory is the most likely culprit for this texture problem.

 

[BenSkywalker]

Right so when we have screenshots which have unblended edges at both near horizontal and near vertical angles, how can it be consistent with an OG pattern? Take this screenshot for example:

http://xbox360media.ign.com/xbox360/image/article/715/715455/prey-20060628052213837.jpg
The near horizontal edges visible in the rock formations are unblended and the near vertical edges on the statues and grandfather are also unblended.

Now this shot is really odd. Take a look at the statue closest to the viewpoint on the right hand side. The front of the statue is clearly aliased while the backside it quite smooth- both near vertical. That is an odd shot.

And then there is this one...
http://xbox360media.ign.com/xbox360/image/article/715/715455/prey-20060628052207416.jpg

The one you posted, I'm sorry but I can't spot any blended edges so you'll have to highlight them for me.

That one has the hand blended, the jukebox not. The shot I posted check out the railings in particular.

So what do you think is smoothing the edges on the crowbar and such here?

Looks like the whole shot was resampled. Not criticizing you or anything, but it actually looks like the image has been scaled twice.

This screenshot resolution is 1600x1200. So that's surely from the PC.

Could be, or it could be from an analog source.

 

[kyleb]

Now this shot is really odd. Take a look at the statue closest to the viewpoint on the right hand side. The front of the statue is clearly aliased while the backside it quite smooth- both near vertical. That is an odd shot.



That one has the hand blended, the jukebox not. The shot I posted check out the railings in particular.

Again you are looking at the effects of heavy JPEG compression in both those examples.

Looks like the whole shot was resampled. Not criticizing you or anything, but it actually looks like the image has been scaled twice.

Your eyes and lack of understanding of compression artifacts deceive you yet again, the shot in question is simply a 50% quality optimized and ICC profiled copy of this full quality capture.

Could be, or it could be from an analog source.

What source rendering at 1600x1200? It's obviously not resized, as can be clearly seen in this x8 zoom, red dot representing a single pixel added by me for comparison sake:

Opps had the wrong shot linked.

 

[Subtlesnake]

But here's my question, wouldnt the larger textures hurt performance even more? From what others are saying its already not to great.

Apparently not.

Increasing texture resolution does not lower the frame rate, because (in general) bigger textures consume storage space, not processing power. You're still filling in the same number of screen pixels, you just have better source material to work with.

http://forums.3drealms.com/vb/showthread.php?t=19343&page=2

Perhaps someone more knowledgeable could elaborate on how the texturing demands scale with resolution.

Also, HH have admitted that memory was the reason for the texture reductions (the full version will still feature reduced texture resolution in certain areas).

I'm sorry I even mentioned the medium thing. I was really only referring to texture size, and that was done for memory reasons, not power. They were also done with an eye toward making them as invisible to the player as possible (things like if you never go near an object that we had a huge texture on changing to a lower resolution because you won't be able to tell the difference at the resolutions we run at). I just didn't want to overstate things and say that no eye was taken toward optimization, as that would be untrue.

http://forums.3drealms.com/vb/showthread.php?p=338137#post338137

 

[Mmmkay]

That one has the hand blended, the jukebox not. The shot I posted check out the railings in particular.

So I took a closer look at the railings:

I don't see a single blended geometric edge. I do see clever usage of lighting and texturing to mask aliasing but I couldn't possibly conclude that there is any MSAA from those shots.

And with regards to the hand in the jukebox shot, as kyleb stated it is clearly down to compression artefacts:

 

[Jawed]

The PAK "format" v3 and upwards (used from quake 3 onwards) is simply renamed .ZIPs. Where megatexture helps consoles is a reduction on texture memory consumption, all things beings equal.

I presume the tiling for streaming is while the gameplayer is moving within the level:

http://www.beyond3d.com/interviews/etqw/ - you did the interview didn't you?

MegaGen outputs two entirely unique 4GB textures - a diffuse map containing colour data, and a normal map - and then combines them into a single 5GB data file. This data file is then split into unique tiles suitable for streaming, and then compressed to reduce disk space usage.

The resulting unique MegaTexture is around 500MB in size. This represents a reasonable tradeoff between ETQW’s visual quality and disk space usage (maintaining a shippable size for the game).

All I'm getting at is that MT hopefully improves loading times and is fast enough to stream tiles directly off DVD while the player moves through a level. But I don't know if PAK actually is unsuited to streaming or tiled/streaming

Jawed

 

[Richard]

It's not a case of either or. ETQW will use megatextures and will still use PAK files. Megatexture is not a replacement for PAK. That's why I mentioned, all things being equal, megatexture will only reduce texture memory usage.

 

[zed]

i think u mean increase texture memory usage (due to the unique texturing)

 

[kyleb]

He means decreased in comparison to what it would take to accomplish the same thing with conventional methods.