Next Gen Graphic Effects Are Amazing (Xbox 360, PS3)

[Dave Baumann]

I'm quite surprised you didn't ask for the article, since this was even such a point of speculation regarding their then upcoming cards (R520), and everyone was wondering when ATi would start supporting it.

I thought that I had asked it and that the answer was that it could, but I can’t recall for definite now whether I explicitly asked about FP16 Blending. There’s also the case that what is supported by hardware may not absolutely be supported by the API yet.

Yes, my question is what that cost with a 32-bit buffer is, once whatever impact compression has is factored in (be it little or large).

Framebuffer compression ratios aren’t actually that high without MSAA, so the savings probably aren’t that great; also notice Bob’s statement on G70 that there is insufficient bandwidth to blend 8 FP16 pixels per cycle – that’s in reference to a 256-bit solution.

Again, though, unless NVidia made changes for their 90nm parts and RSX, this isn't even an option, so why consider it?

It was an example in relation to Xenos.

 

[Megadrive1988]

what really excites me is not Xbox2/PS3 graphical output but that of the next-next gen Xbox3/PS4 - right now with the new Xbox2/PS3 it costs alot of performance to do HDTV resolutions. both fillrate and bandwidth. Xbox2/PS3 can do HDTV pretty easily, but that does not leave a hell of alot of performance left over to really blow away what the current gen consoles can do at 480. there's not a order of magnitude difference in graphics between GS and RSX or NV2A and Xenos (although there is in CPU power between current and new generation). so that brings me back to Xbox3/PS4. by then, doing HDTV resolution will cost almost nothing, and these consoles should really be able to MOVE graphics forward in the same way that PS2-Cube-Xbox did over PS1-N64.

 

[Titanio]

I thought that I had asked it and that the answer was that it could, but I can’t recall for definite now whether I explicitly asked about FP16 Blending. There’s also the case that what is supported by hardware may not absolutely be supported by the API yet.

I think the latter seems a little bit of a stretch. I'm not sure what would prioritise something like FX16 over FP16 as far as the API goes, for example. It's possible, but I think it very unlikely.

It was an example in relation to Xenos.

Yeah, but you were talking about available RAM "for other things" on both systems, wherein that example does not apply on PS3 (or likely won't).

 

[nAo]

Xenos doesn't support blending when using fp16 AFAIK, and it's likely that it can't 'automatically' resolve a MSAAed fp16 buffer for the same reason since it needs blending ALUs to do that.
From a quality standpoint FP16 is likely to be better than FP10 and it should be easy to build a patological case where even a child can spot the difference, even if FP16 is not worth the costs on Xenos, IMHO.

 

[Asher]

One thing we should keep in mind is that when we're actually playing these games, we won't be comparing the HDR implementations side by side.

I've completed PGR3 (well, all of the single player career races up to medium difficulty), and I'm very satisfied with its HDR implementation: especially the effects of coming out of a tunnel and having the exposure adjust (like pupil's responding). If that's fp10 only, I'm more than satisfied with it -- to me it looked very immersive and realistic in that regard.

 

[Skrying]

I'm hoping HDR in console games will help us see real HDR. I'm sorry, but what I've seen first hand with HDR is actually a drawback. The only HDR I've seen is FP16 though, it just seemed far to extreme and unrealistic. Maybe a lower quality will keep Dev's from screwing it up to much.

 

[nAo]

As Dave has already stated in his article about Xenos most developers will just use fp10 cause it's fast, give you less headaches and has got a good quality.

 

[nAo]

I'm hoping HDR in console games will help us see real HDR. I'm sorry, but what I've seen first hand with HDR is actually a drawback. The only HDR I've seen is FP16 though, it just seemed far to extreme and unrealistic. Maybe a lower quality will keep Dev's from screwing it up to much.

There's no such thing as 'real HDR'.

 

[Titanio]

One thing we should keep in mind is that when we're actually playing these games, we won't be comparing the HDR implementations side by side.

That argument could be applied to anything though - "if you don't know something better is possible, you'll be satisfied" - but it's a pretty weak one.

I'm hoping HDR in console games will help us see real HDR. I'm sorry, but what I've seen first hand with HDR is actually a drawback. The only HDR I've seen is FP16 though, it just seemed far to extreme and unrealistic. Maybe a lower quality will keep Dev's from screwing it up to much.

As said above, everything is an approximation. FP16 is just a better one than FP10, like FP32 is a better one than FP16 (although I think I remember reading somewhere that FP32 comes close to the range the eye can detect, though I'm not 100% sure on that. Regardless, there's always the display anyway..). The number of HDR implementations on the PC side has been pretty small to date, but if you want to see a really judicious use of HDR, I'd recommend the MGS4 trailer. Nothing I've seen thusfar touches the quality of the darker scenes with the little robot - the contrast and retention of detail is really very good. It's something of a layman's observation (so, someone can correct me if I'm wrong), but it seems logical that the HDR is making a large contribution there. It's not just about bloom and highlights and the like, but the subtlety of light and shade, the great range of that, and capturing that as best is possible.

 

[aaronspink]

We're going round in circles now. Simple support for the framebuffer format is not in question. It's whether you can effectively do HDR on it or not. Simply having the framebuffer format is not enough. You need to support blending on that format, for example (which is where I guess the problem is on Xenos).

Some would argue that G70 can't effectively do HDR either since their is such a significant performance hit doing FP16 blending vs I8 blending.

IIRC, the slide from way back on the Xenos that leaked out of MS showed support for FP16 blending but abeit at half performance. This is quite reasonable since FP16 blending requires 2x the bandwidth.

Aaron Spink
speaking for myself inc.

 

[Titanio]

Some would argue that G70 can't effectively do HDR either since their is such a significant performance hit doing FP16 blending vs I8 blending.

Doesn't mean a jot if "half the performance" is acceptable performance. We already know of PS3 games using FP16, so that's "effective" enough for me.

IIRC, the slide from way back on the Xenos that leaked out of MS showed support for FP16 blending but abeit at half performance.

Are you thinking of the reference to FX16?

 

[aaronspink]

True, but there's much headroom on both for "creative solutions". I'm not sure how far you can go to escape a lack of FP16 functionality, but either way, workarounds are rarely standard. FP10 usage will be that standard I think.

You are confusing the issue. You are viewing the use of FP10 as a detriment rather than a benefit. A more reasonable arguement is that FP10 allows HDR rending with little to no performance impact and therefore is a benefit.

It'll always be twice as costly, but twice what is the question..

Bandwidth, frame buffer size.

You mean just storing the framebuffer in RAM vs storing some or all of it in eDram. So you'll save a few MBs, maybe? It's not a one way street, though, on the flipside you have to consider the memory requirements of the display list when using tiling..

If a few is being defined as 5-10% of the memory space. Xenos doesn't use a memory based display list. It is a two pass design with the first pass merely setup up the Z compare logic.

And FWIW, I've seen X360 demos that have better lighting and enviromental effects than anything yet demo'd for PS3. Basing your opinion on games that are targetted for 2005 and early 2006 vs games targetted for sometime in 2007 is pointless.

Aaron Spink
speaking for myself

 

[aaronspink]

Again, though, unless NVidia made changes for their 90nm parts and RSX, this isn't even an option, so why consider it? You'll be typically talking about a 720p 64bit frame without MSAA. That, with a 32-bit framebuffer, would be ~10MB?

So your are admitting that the RSX won't support full featured HDR! Thanks, I guess we can end the discussion now.

Aaron Spink
speaking for myself inc.

 

[nAo]

I think people should just stop to think 'real HDR' = FP16. It's just plain wrong

 

[Titanio]

You are confusing the issue. You are viewing the use of FP10 as a detriment rather than a benefit. A more reasonable arguement is that FP10 allows HDR rending with little to no performance impact and therefore is a benefit.

Bandwidth, frame buffer size.

If you follow the thread, you'll see these points have been discussed. FP10 has no performance impact over the regular 32-bit FB, but it's also not as precise, it is a tradeoff.

If a few is being defined as 5-10% of the memory space. Xenos doesn't use a memory based display list.

I'd heard differently from a dev here. As for the 5-10% figure, again, it's been discussed. A FB on RSX isn't going to take 10% of RAM unless we see a change in the MSAA/HDR issue. A 1080p FP16 FB might take closer to 5%, but I think we can agree that's not very likely either.

And FWIW, I've seen X360 demos that have better lighting and enviromental effects than anything yet demo'd for PS3. Basing your opinion on games that are targetted for 2005 and early 2006 vs games targetted for sometime in 2007 is pointless.

This will still be an issue with 2007 games vs 2007 games, methinks. And I've seen that quality in 2006 candidates too.

So your are admitting that the RSX won't support full featured HDR! Thanks, I guess we can end the discussion now.

I don't consider simultaneous MSAA to be necessary, but if you can live with less precision with AA (FP10 is less precision too remember), you can get that on RSX too. If you want higher precision, it's there. I think most devs will opt for the latter any day.

 

[nAo]

Xenos doesn't use a memory based display list.

Umh? display list(s) has to be stored somewhere

It is a two pass design with the first pass merely setup up the Z compare logic.

That's not mandatory

 

[Dave Baumann]

I'd heard differently from a dev here. As for the 5-10% figure, again, it's been discussed.

I assume you are referencing this comment, where he state "may or may not" have memory impacts.

 

[Titanio]

I assume you are referencing this comment, where he state "may or may not" have memory impacts.

Yes, indeed.

The whole argument seems rather silly when you're talking about ~10MB for a 720p FB/32bit Z-buffer. Xenos will be copying its framebuffer out into memory, so the difference is going to be smaller still in most instances.

 

[mj77]

Hi, I have been reading on up on these forums for a while. Sorry i dont have anything to add to this discussion, but I have been wanting to ask this for a while now. If you feel like this question derails the topic at hand, ill gladly delete it and make a separate topic for it.

For the most part from what I have read it appears as if rsx doesn't have any advantages over xenos. So what is the common belief here on the rsx vs xenos? Which at the moment seems to be capable of better visuals in your oppinion?

 

[zidane1strife]

Hi, I have been reading on up on these forums for a while. Sorry i dont have anything to add to this discussion, but I have been wanting to ask this for a while now. If you feel like this question derails the topic at hand, ill gladly delete it and make a separate topic for it.

For the most part from what I have read it appears as if rsx doesn't have any advantages over xenos. So what is the common belief here on the rsx vs xenos? Which at the moment seems to be capable of better visuals in your oppinion?

From what I've heard the RSX may offer slightly more in terms of quantity, while the xenos makes up for it in efficiency and certain features it has. It's still not entirely known which would come out on top[given that RSX is still partially an unknown], but they're expected to be close.

For what it's worth, I personally suspect there's something funny going on with the xenos, and there may be real performance reasons for why many games are running at 30fps and even future games are being aimed at that framerate, and for the lack of Aniso in some.