Digital Foundry Article Technical Discussion Archive [2010]

Status
Not open for further replies.
Infact Split second also has the advantage of 2* MSAA vs 2*Temporal AA of Reach.
I agree that vertical resolution is indeed more noticeable than horizontal reso...even more so in case of racing games where vertical resolution is of more importance than that in other games.
 
There is a lot of talk in various threads about the impact of resolution on IQ.
Granmaster, MazingerDude do you think it would be possible to do a "blind" test in one of your Digital Foundry/eurogamer issue? Or possibly Nebula?
The set-up of the blind test could be:
Captures would be done @1080p, vids @720p (1080p would be better but hosting should prove problematic).
the same games would be run @various settings, the pattern could be something like this:
540p 4xAA, 8xAF, highest texture quality.
640p 2xAA, 4xAF, high texture quality.
720p 2xAA or 0xAA, trilinear filtering, average texture quality.
1080p 0xAA, trilinear filtering, low texture quality.
(I'm letting aside thing like SSAO DoF and motion blur).
People would know nothing of the various setting but would simply be asked which set-up they find the most pleasing.

The tough point would be to have "jumps" in texture quality reasonable in regard to resolution increment. The game could be an old PC games with quiet some mods available so you have some choices when it comes to textures packs. I'm sure Nebula could help choosing the perfect candidates.

I know that's quiet some work but results could prove interesting.
 
It would probably be a good idea to check out the net effect of negative LOD bias, particularly for lower resolution framebuffers. As for scaling, perhaps lanczos or lower quality filter to compare them directly at 720p or other.
 
I think this is the right place to ask for it.

Its obvious that the upcoming RDR article will also include thoughts & analysis over the differences between the two platform so its kinda safe to assume that PS3's transparency/fillrate issues will get a mention here. Now I was thinking, wouldn't it be great for them to have a look at how Farcry 2 (an open world game with heavy tech and lots of transparencies/ extremely high grass density) resolves this issue on the PS3 hardware..?
 
Last edited by a moderator:
I think this is the right place to ask for it.

Its obvious that the upcoming RDR article will also include thoughts & analysis over the differences between the two platform so a its kinda safe to assume that PS3's transparency/fillrate issues will get a mention here. Now I was thinking, wouldn't it be great for them to have a look at how Farcry 2 (an open world game with heavy tech and lots of transparencies/ extremely high grass density) resolves this issue on the PS3 hardware..?

I think I gave DF the config file for FC2 that also contains console settings some year ago. Might be something in there. I can post it if anyone wants to see it.
 
I don't get this comparison:

http://images.eurogamer.net/assets/articles//a/1/1/0/1/8/9/5/LightningChange360_001.jpg.jpg
http://images.eurogamer.net/assets/articles//a/1/1/0/1/8/9/5/LightningChangePS3_001_.jpg.jpg

They said only on ps3 shot you could see the lighting from the door being opened, but it definitely seems there on 360 shot, its just in a different angle so it blends with the door shadow...

For me the strange thing about those pictures is how the middle of the room is being lit as if the light was coming through the window, which seems impossible given the shadow orientation...
 
I don't understand this (atleast in case of GTA4)

It is perhaps not surprising. A cityscape naturally brings about a lot of occluding buildings, masking out the environments beyond, and reducing engine load. Red Dead Redemption is all about making up for the reduced levels of detail in the foreground by presenting a wilderness setting that seems to stretch almost limitlessly into the distance.

I mean shouldn't the fact that you could practically ride a chopper & fly around the city at any give elevation nullify this point ? I remember Laa-Yosh stating the same.
 
I don't understand this (atleast in case of GTA4)



I mean shouldn't the fact that you could practically ride a chopper & fly around the city at any give elevation nullify this point ? I remember Laa-Yosh stating the same.

Yup and game also got a fair amount of low height houses and road crossing. The height is also not the same for whole city so you can end on roads or terrain set higher than other areas and thus get a wide view.
 
Just read the DF comparison+interview for Blur (the only critics I have about the videos: please grandmaster, learn to drive in FPS view, third person view in racing games is for kids only)!
Very good read overall-thanks the dev for giving this great insight!!

It seems that finally Bizzare is a MP dev who cares for the architecture of the PS3? - astounding!
Instead of constantly moaning*** about the difficult PS3 architecture, they see challenges and potential - cool!
Looking forward to see what they get in future!


At the end, one of the devs mentioned that they are looking into the GOW3 MLAA stuff and other related tech for AA on PS3 as this is obviously one of the Xbox360 advantages up to now.
So I thought about this:

They already state that they are using the SPUs for lot's of different stuff to help out the renderer. As MLAA is not free, they somehow need to free up SPU time. On the other hand, you could free RSX from his MSAA burden when using MLAA on the SPUs.
How to free up SPU time? Maybe, RSX can use its new freedom (without MSAA) to take some rendering stuff back? Am I right to assume that RSX can do SPU rendering stuff at least as fast?
Thinking about this SPU-RSX ping-pong relationship, I somehow thought that this is funny.

They also talked about that they are doing a lot of stuff in parallel on the PS3...in the sense that SPUs work in parallel to RSX for the renderer.
So is the PS3 a system with 2 "weak" unsymmetric graphics card compared to the Xbox360 which is a system with 1 "strong" graphics card?




***Alan Kertz (EA DICE)
Looking forward to the PS4, in hopes it doesn't suck as much to develop for as the PS3.
5:37 AM May 24th via Digsby

Source: http://twitter.com/Demize99
 
As for 4x MSAA on PS3, this simply wasn't doable because of the additional memory requirements. A back buffer of 1280x720x4AA with depth buffer is nearly 30MB on PS3, whereas this is around 7MB on 360 because the multi samples are only ever used in EDRAM and combined as you copy to main memory, so AA doesn't require any additional memory. We simply couldn't afford such a large memory discrepancy between the PS3 and 360 versions because other systems all had tight budgets.

I did not know this, can someone explain to me in more detail how can the 360 manage to only use 7MB back buffer compared to the 30MB the PS3 uses for 1280X720 4xAA.

http://www.eurogamer.net/articles/digitalfoundry-blur-tech-interview?page=2
 
I did not know this, can someone explain to me in more detail how can the 360 manage to only use 7MB back buffer compared to the 30MB the PS3 uses for 1280X720 4xAA.

http://www.eurogamer.net/articles/digitalfoundry-blur-tech-interview?page=2

I presume it's done fully in EDRAM, but in two or four tiles. The EDRAM being 10MB, you'd never go above that figure, and the 7MB probably comes form there. I don't know though what part of the 10 render targets are compiled before that and what part of the render targets during or after - it would have been an interesting quesiton.

That's just a rough guess though - I'm not really a tech-head. ;)
 
What about MLAA though? How much memory does that take?

I think it also uses a manner of tiling, and I think it doesn't take any additional memory, because the framebuffer is streamed through the SPUs. Again, not 100% sure though, as again, I'm not a tech-head.
 
I think it also uses a manner of tiling, and I think it doesn't take any additional memory, because the framebuffer is streamed through the SPUs. Again, not 100% sure though, as again, I'm not a tech-head.
That sounds quite complicated. I thought the RSX was the one resolving the framebuffer, not the SPUs. Neither I am a techie though, so....

To me, I don't find anything muse-ery, before the whole FB ithing is resolved I half wonder if it's done in the GPU or the CPU. It's quite confusing
 
There is a lot of talk in various threads about the impact of resolution on IQ.
Granmaster, MazingerDude do you think it would be possible to do a "blind" test in one of your Digital Foundry/eurogamer issue? Or possibly Nebula?

The set-up of the blind test could be:
Captures would be done @1080p, vids @720p (1080p would be better but hosting should prove problematic).
the same games would be run @various settings, the pattern could be something like this:
540p 4xAA, 8xAF, highest texture quality.
640p 2xAA, 4xAF, high texture quality.
720p 2xAA or 0xAA, trilinear filtering, average texture quality.
1080p 0xAA, trilinear filtering, low texture quality.
(I'm letting aside thing like SSAO DoF and motion blur).
People would know nothing of the various setting but would simply be asked which set-up they find the most pleasing.

The tough point would be to have "jumps" in texture quality reasonable in regard to resolution increment. The game could be an old PC games with quiet some mods available so you have some choices when it comes to textures packs. I'm sure Nebula could help choosing the perfect candidates.

I know that's quiet some work but results could prove interesting.
I am up for it, great idea. A joint collaboration between Mazingerdude, grandmaster -yes, DF knows all and tells all....- and Nebula would be truly nice.

Also the framerate differences between each of those different settings and with what settings it's easier to obtain the awe-inspiring 60 fps in games without affecting visual fidelity that much!
 
I did not know this, can someone explain to me in more detail how can the 360 manage to only use 7MB back buffer compared to the 30MB the PS3 uses for 1280X720 4xAA.

http://www.eurogamer.net/articles/digitalfoundry-blur-tech-interview?page=2

At 4xmsaa, you need to store data for 1280x720x4 pixels, or 3686400 pixels total. For each of those pixels you need both color and z data each of which is typically 4 bytes/pixel. So if you need 8 bytes/pixel then it comes to 3686400 * 8 = 29491200 bytes total, or around ~28mb needed to store the temporary msaa data. On PS3 there is no scratch memory space so that 28mb eats up part of the 512mb of memory that it has.

On 360 the overall total amount of memory needed to complete the 4xmsaa process is identical, 28mb. The difference is that it takes 0mb of the main 512mb of memory because the 10 edram memory is used. You render it as multiple tiles, so instead of rendering it as 1280x720x4, you render it in three parts:

1) 1280x256x4
2) 1280x256x4
3) 1280x208x4

Assuming again 8 bytes per pixel means that tile #1 above needs 1280x256*4*8 = 10485760 bytes, which is exactly 10mb. Because it's all done in edram, that means that 0mb of the 512mb of memory is needed to do msaa on 360. In otherwords, all things being equal if you use 4xmsaa on both PS3 and 360 versions of the game, then you will need to free up an additional ~28mb on the PS3 version to make it fit into memory.

I'm not quite sure why the article says it takes 7mb on 360 though. On both machines you need a final 1280x720 buffer to resolve to, which would be around 7mb in size, but I don't know why they include that figure on the 360 memory total but not on the PS3's. If you go by just memory needed for 4xmsaa, then it's ~28mb on PS3 and 0mb on 360. If you include the final resolve buffers into that then it's ~35mb on PS3 and ~7mb on 360.


What about MLAA though? How much memory does that take?

Not sure on this, but worse case I think you need two 1280x720 color buffers, or best case probably just one. Because the mlaa process is done in parallel on spu's (while the gpu does other stuff) means that you have to keep a single 1280x720 source color buffer around for the spu's to munch on, and make sure nothing else writes to it during that process, so that memory needs to be kept around. However It's possible that other post process steps are using that same that pre-mlaa'd color buffer for other stuff in which case you could consider that memory free as far as mlaa goes. For the second buffer, presumably you need a destination 1280x720 mlaa'd output buffer somewhere in memory, and again because it's happening in parallel with the gpu I would assume that it's a secondary backbuffer. In other words, while the gpu is writing to the main backbuffer doing whatever it needs to do, the spu's are in parallel writing to another backbuffer. Hence why I figure it's worse case two 1280x720 color buffers of memory, and best case jsut one. Just a guess though.
 
At 4xmsaa, you need to store data for 1280x720x4 pixels, or 3686400 pixels total. For each of those pixels you need both color and z data each of which is typically 4 bytes/pixel. So if you need 8 bytes/pixel then it comes to 3686400 * 8 = 29491200 bytes total, or around ~28mb needed to store the temporary msaa data. On PS3 there is no scratch memory space so that 28mb eats up part of the 512mb of memory that it has.

On 360 the overall total amount of memory needed to complete the 4xmsaa process is identical, 28mb. The difference is that it takes 0mb of the main 512mb of memory because the 10 edram memory is used. You render it as multiple tiles, so instead of rendering it as 1280x720x4, you render it in three parts:

1) 1280x256x4
2) 1280x256x4
3) 1280x208x4

Assuming again 8 bytes per pixel means that tile #1 above needs 1280x256*4*8 = 10485760 bytes, which is exactly 10mb. Because it's all done in edram, that means that 0mb of the 512mb of memory is needed to do msaa on 360. In otherwords, all things being equal if you use 4xmsaa on both PS3 and 360 versions of the game, then you will need to free up an additional ~28mb on the PS3 version to make it fit into memory.

I'm not quite sure why the article says it takes 7mb on 360 though. On both machines you need a final 1280x720 buffer to resolve to, which would be around 7mb in size, but I don't know why they include that figure on the 360 memory total but not on the PS3's. If you go by just memory needed for 4xmsaa, then it's ~28mb on PS3 and 0mb on 360. If you include the final resolve buffers into that then it's ~35mb on PS3 and ~7mb on 360.

Perfect, thanks Joker. :eek:
 
Status
Not open for further replies.
Back
Top