My idea is, that they did it because the engine uses Xenos' EDRAM.
[360, PS3] Crysis 2
- Thread starter ultragpu
- Start date
My idea is, that they did it because the engine uses Xenos' EDRAM.
dragonelite
Veteran
Prophecy2k
Veteran
Where and when was that quote from?
That was quoted in a magazine way back. Shortly after Crytek bought the remains of Free Radical, I think it was as the journalist was being shown around the office he wasn't allowed in a super secret room, where something amazing was, apparently, being done with the PS3.
Hopefully it's all about TimeSplitters 4!!
Guess we'll really have to wait for the demo.Kind of silly now that they put out the earlier 360 beta, but I suppose it's part of their hype train, and the work was already done from the closed beta. Now they can (hopefully) show something with closer-to-final code.
I'm happy they released all the demos they have. The 360 beta was a mess online half the time, even after the closed alpha/beta demo.
I expect this next demo to play much better and represent what the final game will play like, pending any possible launch day patch they may incorporate.
Out curiosity what exactly did they do because of the EDRAM?
This way they're able to fit three frames' (1152x720 x 4 bytes) worth of data in the EDRAM which is only 10MB big. This gives them enough memory for a deferred renderer, although I'm not entirely sure about the G-buffer layout (and C2 is deferring the lighting only as I recall).
Ok thanks, thought this is what he meant but wasn't sure.
Does the G-buffer reside in memory or the EDRAM? Was wondering if the EDRAM creates any hurdles for developers to use deferred rendering on the 360.
Well, you need some place to put the rendered pixels into while you're rendering, and if all your data doesn't fit into the EDRAM then you'll need to render several passes, re-submitting all geometry and so on. Or you can use tiling and assemble the G-buffer in main memory; I think it has to be copied there for the lighting and final compositing stages anyway.
As far as I know Trials HD is a fully deferred renderer running on the 360, but it's not 1280*720 either, it cuts 20 rows from the picture, but on the other hand it runs at 60fps.
So it can probably be said that full 1280*720 and deferred rendering at the same time with high levels of detail are a bit too much for the Xbox.
As far as I know Trials HD is a fully deferred renderer running on the 360, but it's not 1280*720 either, it cuts 20 rows from the picture, but on the other hand it runs at 60fps.
So it can probably be said that full 1280*720 and deferred rendering at the same time with high levels of detail are a bit too much for the Xbox.
Actually, from the Siggraph10 slides it's only two render targets for their G-Buffer.
RT0: Depth 24bpp, Stencil 8bpp
RT1: 24bpp normals + glossiness 8bpp
In the next pass (light accumulation), they render to ARGB16 in eDRAM then resolve to R11G11B10 in main memory.
During the G-buffer generation there is some room left, however during the light accumulation there isn't.
Depth + RGBA16 @ 1200x720 is just about exactly 10Mb.
Hmm, they do mention the 8 byte G-buffer as the minimum requirement, but some other presentations mention storing more specular information as well.
My uninformed impression is that the 8-byte layout is for the PS3, because of various considerations (bandwidth, memory) and they're using a higher quality layout on the 360 because of the EDRAM. But they're scared of the flame wars it'd ignite so they're silent about it
Checking it in the final version should involve careful examination of specular highlights checking for the amount of color preservation, and maybe also normal map precision (their solution for 24 bit is good, certainly better than normalized normals; but if they had 2x16 bit, it'd probably be even better).
There could of course be a completely different explanation for what they're doing with the remaining 1/3 of the EDRAM on the 360, but there must be something if they went for 1152*720 instead of a full 1280 frame.
My uninformed impression is that the 8-byte layout is for the PS3, because of various considerations (bandwidth, memory) and they're using a higher quality layout on the 360 because of the EDRAM. But they're scared of the flame wars it'd ignite so they're silent about it
Checking it in the final version should involve careful examination of specular highlights checking for the amount of color preservation, and maybe also normal map precision (their solution for 24 bit is good, certainly better than normalized normals; but if they had 2x16 bit, it'd probably be even better).
There could of course be a completely different explanation for what they're doing with the remaining 1/3 of the EDRAM on the 360, but there must be something if they went for 1152*720 instead of a full 1280 frame.
Hmm, that's interesting, however are they really rendering to RGBA16 on the 360?
Presentations suggest R11G11B10 and constantly adjusting the dynamic range based on the previous frame.
Yeah, in various siggraph papers, they mention rendering to signed ARGB16 in eDRAM, which is resolved to R11G11B10 in main memory so they keep the 32bpp external bandwidth.
My uninformed impression is that the 8-byte layout is for the PS3, because of various considerations (bandwidth, memory) and they're using a higher quality layout on the 360 because of the EDRAM. But they're scared of the flame wars it'd ignite so they're silent about it
They may be doing this for more material flexibility, at least going by the Siggraph 09 presentation on LPP renderers, with the downside being a second geometry pass.
Indeed... they did mention the 10-bit per channel as a possible avenue in the initial presentation on their deferred lighting. And then there's also the other work they do with best fit normals & 3Dc Encoding (adv. render techniques @ Crytek site).
Well, I'd still say that if they have free space while filling up the G-buffer then why not use it? Writing out more render targets should not affect their performance that much.
Then again, the entire concept of the engine is to sacrifice quality for more features on current console hardware in order to better prepare their renderer and content creation for the next generation systems...
Then again, the entire concept of the engine is to sacrifice quality for more features on current console hardware in order to better prepare their renderer and content creation for the next generation systems...
Could anyone explain these three slides for me, please?
http://www.crytek.com/sites/default/files/A_bit_more_deferred_-_CryEngine3.ppt, Slides 7-9.
http://www.crytek.com/sites/default/files/A_bit_more_deferred_-_CryEngine3.ppt, Slides 7-9.
Similar threads
