*spin-off* idTech Related Discussion

John Carmack himself gave a good example on what might be an Xenos advantage:

"Embedded dram should be a driving force. It is possible to put several
megs of extremely high bandwidth dram on a chip or die with a video
controller, but won’t be possible (for a while) to cram a 64 meg
geforce in."

But again, the supposed higher Shader performance and the EDRAM advantage is all known factors and it´s been like this for a long time. That hasn´t stopped other developers from getting the same performance.

The real question is if id i able to get the Cell to do it´s part.

Carmack was talking about scratchpad in EDRAM, Xenos does not have it.
You can only resolve to EDRAM on Xenos, not from EDRAM.
 
Since the RSX has double the number of fragment shaders of a 7600, as well as more vertex shaders, and other enhancements, I think it's a bit misleading to call it "a 7600-derivative". In fact, I'd be surprised if the 8600 was as fast as RSX. (The 8600 is certainly slower than most 7900's, which are arguably most similar to RSX.) Of course, the 8600 didn't launch until months after the PS3, and probably years after the first prototypes of the PS3 were made available to developers, so it's a moot point...

Well then I stand corrected.
 
Flickering lod's probably have nothing to do with xenos. The dvd drive might be seeking all over the place trying to feed pixel data, and possible falling behind in some situations currently. Installing the game to hdd would solve that, but they still have to make it work on the $199 hdd-less model.

My point is that we don't know this.
 
Carmack was talking about scratchpad in EDRAM, Xenos does not have it.
You can only resolve to EDRAM on Xenos, not from EDRAM.

You have it backwards, resolves happen from edram to main ram. You do all rendering in edram and you resolve it out to main ram when you are done with it. This is why for example you can do full 1280x720 transparencies on 360 because during rendering it can lookup the blend color very fast right from edram (which is the rendering scratchpad), whereas the PS3 will run out of bandwidth very fast hence requiring 1/4 size (or less) transparencies that look very blocky like in Infamous.


My point is that we don't know this.

We don't, but we can make make an educated guess at it, and it sounds like a dvd bandwidth issue to me.
 
We don't, but we can make make an educated guess at it, and it sounds like a dvd bandwidth issue to me.

Why you do think so? From our current understanding of MT, the lower mip map levels are only showed (and faded in) when they are fully loaded. Until then the higher level mip map is displayed. There's no potential for flickering in the streaming whether be it because of slow media read speeds or otherwise, from previous interviews MT even manages streaming over VPN. To me this looks like it's a problem with the shader-based filtering (or surface aliasing - but I don't see a lot of highfrequency noise in the screenshots released thus far).
 
You have it backwards, resolves happen from edram to main ram. You do all rendering in edram and you resolve it out to main ram when you are done with it.

But Carmack is not talking about resolving back the result, he wants to resolve before that, e.g. use EDRAM as texture cache "shader scratchpad".
 
Last edited by a moderator:
Why you do think so? From our current understanding of MT, the lower mip map levels are only showed (and faded in) when they are fully loaded. Until then the higher level mip map is displayed. There's no potential for flickering in the streaming whether be it because of slow media read speeds or otherwise, from previous interviews MT even manages streaming over VPN. To me this looks like it's a problem with the shader-based filtering (or surface aliasing - but I don't see a lot of highfrequency noise in the screenshots released thus far).

Streaming over vpn would be fine if the source on the other end is a hard drive. I think it's the dvd drive seeks which could be problematic in some cases. Apparently they have extended the use of megatexture beyond just terrain and are using it for everything this time around. Given how huge the terrain textures must be on disc, it seems to imply that the optical head must be bouncing all over the place to get pixels for terrain, buildings, characters, etc, since they may be a big distance apart on the disc. If the player is say moving forwards quickly, will it really be able to get all the high mips loaded in time? Will it be able to get all textures loaded at all? Presumably there is a fail safe in there somewhere to default to a color if no texture/mip is ready at all. If I rush fowards into a scene that shows lots of buildings, characters, and terrain far into the distance, will it really have all those textures ready to go just from the optical drive or will it sometimes have to temporarily default to just a color? That kind of visual glitch could be perceived as flickering.

Additionally, if they are indeed showing the highest mip level all the time until the others are loaded, then that could lead to lots of on screen shimmering that could be perceived as flickering. For example, with some textures if you use the high mip for something far in the distance, it will shimmer really bad as you move around. If lower mips are piece meal loaded in and happen to be ready staggered all over the screen then the result could be a mish mash of on screen shimmering which might also look like flickering.

When it comes to loading pixels the PS3 should have the advantage there both because data is likely to be less far apart on the blu-ray (which means shorter seeks), but also because they can stream from optical disc and hdd at the same time. They can even require a 5gb mandatory install (which I think is very likely) and install non terrain art to hdd and leave terrain art on hdd. That way the optical drive can be focus on just loading terrain pixels which are likely always close by since you aren't teleporting around, and let the hdd deal with the rest. That would further minimize the seeks the optical drive would need to do. I think that's why the PS3 version doesn't have the flickering issue.

I suppose there could also be an issue with shader based filtering if they are doing heavy pixel approximations in the vertex shaders on the 360 version for stuff that is far away. I don't know if that would come across as "flickering" though, but true it could lead to artifacts.
 
IIRC....
G70:
Fragment/Pixel Shader ALU dual-issue vec4, (so it's essentially equivalent to 48 single-issue Vec4)
Vertex Shader ALU co-issue is Vec4 + Scalar
5 instructions per pixel


Xenos: Vec4 + Scalar
apparently "6 ALU instructions per pixel" according to Wavey's document.

It may be worth noting that Xenos' thread scheduling should eliminate texture lookup stalls... something that'd be an issue for RSX (bye-bye half of fragment shading?).


IIRC, 64 pixels vs 1024 pixels.

edit: G70 might be lower than 1024... fuzzy memory is fuzzy. :oops:

So (please enlight me/corret me if i wrong) RSX/G70 cannot put "on the fly" all 136 shaders instrutions/cycle(or 384 flops Vec4 + free) texture at same time?

If i understand...If RSX can only 50% shaders + texture ,so this gpu can offer something like 70/80 shaders instruction/cycle versus 96 + Xenos Gpu?

(but sometime ago wee see developers declaration about RSX is "shader monster"...)
 
Last edited by a moderator:
Have we seen console screenshots yet? Or is this something we'd see on PC screenshots too?

The shots I've seen are oversampled bullshots so we won't be able to judge filtering or anything like that based on them. Id will probably not release 1:1 shots until very close to the release, if at all.
 
Given how huge the terrain textures must be on disc, it seems to imply that the optical head must be bouncing all over the place to get pixels for terrain, buildings, characters, etc, since they may be a big distance apart on the disc. If the player is say moving forwards quickly, will it really be able to get all the high mips loaded in time?

Not to mention using a sniper rifle ;)
 
Have we seen console screenshots yet? Or is this something we'd see on PC screenshots too?

I believe any compression differences between XBOX/PC wouldn't be enough to remove high frequency noise (not to the point of one platform showing flickering because of it).

Additionally, if they are indeed showing the highest mip level all the time until the others are loaded, then that could lead to lots of on screen shimmering that could be perceived as flickering. For example, with some textures if you use the high mip for something far in the distance, it will shimmer really bad as you move around.

In ETQW the highest mip map level is always loaded into memory. The game let's you test various bandwidth constricted scenarios through CVars and this artefact never showed up. The MT in id Tech 5 is different without a doubt but if they still pre-load the highest mm level and continue to fade in the lower levels when they're done loading I don't believe flickering can occur.

If lower mips are piece meal loaded in and happen to be ready staggered all over the screen then the result could be a mish mash of on screen shimmering which might also look like flickering.

IMO this is a lot more likely than the above situation. In ETQW the tiles were loaded asynchronosly in a concentric ring around the player as he moved. By mapping arbitrary 3D geometry to a 2D surface and stored on 1D media you do get a lot more discrepancy between physical location of the camera, read position on the texture and sector location on the drive.

I could swear there's was an interview where JC or someone at id mentioned how they already had id tech 5 MT working smoothly on DVD but google/bing isn't throwing up anything useful. Maybe it was in video interview.

I think that's why the PS3 version doesn't have the flickering issue.

If this is physical location on disc problem I'd say the PS3 also wouldn't have this because the extra storage space would allow redundancy.
 
I could swear there's was an interview where JC or someone at id mentioned how they already had id tech 5 MT working smoothly on DVD but google/bing isn't throwing up anything useful. Maybe it was in video interview.

Hmm, well if you can find it post it, I wouldn't mind taking a peek at it. I can see how they could get dvd working fine for the older tech which just MT'd the terrain, since it's fairly sequential access to terrain data that can be well optimized. But if they can also pull it off on their current tech which MT's everything then that would be pretty neat.

I suppose they could get clever and intermix textures. So say if there is a building sitting on a piece of terrain, then that buildings textures could be layed out on disc within that part of the terrain texture at that same world location, which would keep texture access relatively sequential as you walk around the world. They just need a comprehensive table of contents of sorts to let them extract everything into useful data. Then again it doesn't help with movable stuff like characters, and it would duplicate a lot of texture data as well like if two buildings at separate locations share the same textures but each gets embedded uniquely into their respective areas of terrain texture. That's fine for blu-ray but not ideal for dvd. I guess we need to wait until Id reveals more.


If this is physical location on disc problem I'd say the PS3 also wouldn't have this because the extra storage space would allow redundancy.

It's also because there's no layer change penalty. Presumably a single layer 25gb blu-ray would be enough, whereas they have no choice but to go with dual layer dvd's, which take a performance hit when you hop between the layers. They could engineer the levels to minimize that hit though. It will be really interesting to benchmark this game to see how it runs just with dvd compared to running off the 360's hdd. I know Fable 2 had fairly dramatic performance differences in that regard, I suspect this game will as well.
 
IIRC....
G70:
Fragment/Pixel Shader ALU dual-issue vec4, (so it's essentially equivalent to 48 single-issue Vec4)

Right, except that the "vec4" can be a "vec3 + scalar" or a "vec2 + vec2", so, either way, 96 instructions per cycle for RSX.

Then, IIRC, there's some sort of "mini-ALU" (or special function unit?) that adds another instruction per shader per clock, so, possibly, if these extra instructions are useful, that raises RSX's fragment shading capabilities to 120 instructions per clock.

Vertex Shader ALU co-issue is Vec4 + Scalar
5 instructions per pixel

I don't follow...

Xenos: Vec4 + Scalar

Correct, so 96 instructions per cycle for Xenos.

Also, we can't forget that Xenos also benefits from having 16 dedicated texture units, so, if they're fully utilized, then Xenos peaks at 112 instructions per cycle.

apparently "6 ALU instructions per pixel" according to Wavey's document.

What? :?: Where can I find Wavey's document? This sounds interesting...

It may be worth noting that Xenos' thread scheduling should eliminate texture lookup stalls... something that'd be an issue for RSX (bye-bye half of fragment shading?).

Agreed, but I hope it's not as bad as wasting 50% of RSX's peak performance.

IIRC, 64 pixels vs 1024 pixels.

I think that G70 is 880 pixels (because I always wondered how an unusual number like 880 came about).

Anyway, I believe Carmack if he says that megatexturing runs faster on Xenos, but I'd like to understand why.
 
Sounds like joker was on the right track. LOD snap in virtual textures because of high latency between when it is needed and available. Optical disc read is noted as a culprit.

Have to say, those images in the presentation look fantastic. I'm assuming high high high high end PC.

Oh, and the job model sounds pretty cool. I'm becoming more and more interested in parallelism, but I don't really have any programs to work on that would require it.
 
Sounds like joker was on the right track. LOD snap in virtual textures because of high latency between when it is needed and available. Optical disc read is noted as a culprit.

The culprit is seeking, not read speeds, which means the PS3 would also suffer from this. Note also that J.P. says they use blending to get rid of the snap (which would be easier to do if they used trilinear filtering).
 
Back
Top