*Game Development Issues*

[Rockster]

It's just easier to reach peak performance on Xenos, moreover an unified ALUs GPU is very good at automatically fix certain art/contnet problems (too dense meshes for example, missing LODs, etc..)

And this is bad because? nAo, given what you know about NT's staff and the state of both Microsoft's and Sony's development environments, would you say that it would have taken more or less time, resources, and money to create Heavenly Sword on the 360.

 

[nAo]

And this is bad because?

It's not bad per se, it's good to be able to produce good graphics with less effort but every cycle (cleverly) spent by the GPU to fix art/content problems could be spent elsewhere.

 

[Dr. Nick]

Along side typical geometry lod's now you have to write shader lod's. A pain in the ass which isn't 100% foolproof either.
.

I know you are just giving an example but wouldn't using something like this help on the 360 as well.That being said has ATI(AMD) or Microsoft released updates for Load Balancing on the Xenos.

I doubt they have very complex shaders, they are master at faking stuff, and this is all what matters in the end.

Yep, no matter what technical feat other dev's working on racing games claim to have accomplished GT's artistry beats them out every time.

 

[MrFloopy]

What did you see? Did you write an allocator for it? IIRC on MS's optimization guide for 360 there was "don't use linked list" or something like that... Instancing templates can lead to unexpected code bloat, I doubt it is the best solution for the 256KB memory.

Avoiding linked lists for performance reasons is not in anyway specific to the 360. They're pretty evil generally. Although I imagine, coincidently, not too bad on an SPU with LS.

 

[scificube]

Avoiding linked lists for performance reasons is not in anyway specific to the 360. They're pretty evil generally. Although I imagine, coincidently, not too bad on an SPU with LS.

Indeed. If one ties lists or tree like structures with "free" lists and can likewise fit them in an LS/cache then they will perform quite well. Basically the pointer chasing is all on chip, you gain data locality (which makes DMAs more predictable/simpler), and memory fragmentation is 100% impossible.

 

[joker454]

You are describing a situation at the extreme where RSX may have some difficulty.

Not really extreme, actually fairly common for all sports games, and probably reasonably common in any game that gives the user camera control. It's amazing the amount of havoc a user controlled camera can cause with respect to optimization.

What happens in the case where both heavy vertex and pixel processing demands must be met concurrently? In this case both the vetex and fragment pipelines are available on RSX so the demand can be met. On Xenos the pipes can only be split as the ALUs can't handle both jobs at the same point in time.

RSX's vertex abilities at full tilt can't keep pace with Xenos, even if split. Maybe even if quartered It's been argued here that pixel side RSX should be better but we haven't seen that, at least not in our shop. I suppose if we really hollowed out our pixel shaders then the split bandwidth on PS3 could prove to be an edge. But thus far, our pixel shaders still have enough instructions in them to give Xenos the edge. Mated with it's automatic load balancing, it's tough for RSX to keep pace.

This is a fidelity issue rather than extracting good performance.

It's a fidelity issue resulting from a performance issue. Believe me, we'd like to be able to just use alpha, in some cases it just plain looks better. But we can't on PS3 in this particular case due to performance issues, which in turn has now led to a fidelity issue. You would probably never notice in a shooter or faster moving games. But in baseball where it can just sit there in your face for a while, it is noticeable.

Sometimes the performance issues lead to more than just fidelity problems. Our 360/PS3 versions are almost identical, but there is for example one replay-only effect (30fps mode for replay) where we really push edram to create a nice volumetric grass. We have not been able to recreate this on PS3 at a reasonable framerate. So in this case the problem extends beyond just a visual issue. We had to cut that effect totally on PS3.

You could try different MSAA levels on the fence or perhaps dynamically alter it.

I'm not a fan of changing msaa to 0x/2x/4x on the fly, the visual incongruity it causes in this case would to me would be worse that the artifacts we're seeing in the first place!

I know you are just giving an example but wouldn't using something like this help on the 360 as well.

Not necessarily! It's like the obsolete art of assembly coding. Ok, that may offend a few here But the reality is that for the most part, given the complexity of todays hardware you are better off coding in a slightly more abstracted language and just let the compiler arrange the microcode optimally for you. Overall it's likely to do a better job. That's why for example it's always strongly recommended to write shaders in hlsl and just let the compiler do the rest.

In this case, the gpu ultimately has intimate knowledge of whats going on at any given moment and is in a better position to determine what load needs to be distributed where at a given frame. We can try to help it a bit on our side, but only at a more macro level. We are better off letting Xenos do it's thing.

 

[Mintmaster]

Really that would be fine by me

Sure, but it invalidates the point you were making.

PD isn't using complex shaders. They simply aren't pissing away shader cycles on negligible improvements, and are taking photos and tweaking the colours/textures to match reality.

You can even see how their "HDR" is limited when a white license plate has the same bloom as the sun reflecting off the body or wheels (which points to it not being HDR at all). However, because they don't abuse the range and distort effects way beyond reality, overall it looks better than many games that use real HDR.

The variance in dev and artist talent is much greater than that of the hardware.

 

[betan]

Here is what Dennis Dyack (Silicon Knights) says:

The 360 and the PS 3 are equal in power in my eyes. Maybe the PS 3 has more processing power. The 360 has more available memory. It’s pretty much a net, net. The public perception of the PS 3 was that it was much more powerful. To developers, they look even.

There is potentially more power in the PS 3, he acknowledged, but it is harder to tap.

Nothing new of course.

Take alpha for another example. Alpha as most know is terrible on RSX. Fine, so we ditch it and go with alpha coverage. It is after all claimed to look as good as alpha to the common user.

I think you need at least 4xAA to make such a claim. I expect it to be noisy at 2xAA. This is what you do on PS3 right?.

But I noticed that GT5 demo has very good fence even at 1080p, and it's possible to put it in front of the whole view so it's certainly not art related.

Anyways, I can go on and on, but that's just two examples on many such graphics related issues that arise when you try getting a Xenos pushing game over to RSX.

Please do so as I find that more interesting than how CELL is difficult for noobs.

 

[scificube]

Sure, but it invalidates the point you were making.

PD isn't using complex shaders. They simply aren't pissing away shader cycles on negligible improvements, and are taking photos and tweaking the colours/textures to match reality.

You can even see how their "HDR" is limited when a white license plate has the same bloom as the sun reflecting off the body or wheels (which points to it not being HDR at all). However, because they don't abuse the range and distort effects way beyond reality, overall it looks better than many games that use real HDR.

The variance in dev and artist talent is much greater than that of the hardware.

My point is in the scenario I outlined. I had a bad example and was corrected on it.

My point a little more generally is that there are extremes RSX handles better than Xenos and visa versa. That was it.

 

[scificube]

Not really extreme, actually fairly common for all sports games, and probably reasonably common in any game that gives the user camera control. It's amazing the amount of havoc a user controlled camera can cause with respect to optimization.

I was speaking to the extremes from the performance point of view. I understand that is not an uncommon situation in games.

Still the solutions are relatively straight forward and Cell support is not a must to implement them.

RSX's vertex abilities at full tilt can't keep pace with Xenos, even if split. Maybe even if quartered It's been argued here that pixel side RSX should be better but we haven't seen that, at least not in our shop. I suppose if we really hollowed out our pixel shaders then the split bandwidth on PS3 could prove to be an edge. But thus far, our pixel shaders still have enough instructions in them to give Xenos the edge. Mated with it's automatic load balancing, it's tough for RSX to keep pace.

I want to be clear here. I am not debating whether Xenos has it's advantages over RSX and especially not vertex processing. There is nothing to quibble about there.

I'm not really in this discussion to establish which part is better than the other.

What interested me was your saying Cell intervention is needed for decent performance out of RSX. I do not feel this is how it is and that is all I wanted or wish to speak to.

Frankly I do not see RSX making PS3 development untenable. I accept that there are issues. I do not accept that RSX is the source of widespread problems for everyone.

It's a fidelity issue resulting from a performance issue. Believe me, we'd like to be able to just use alpha, in some cases it just plain looks better. But we can't on PS3 in this particular case due to performance issues, which in turn has now led to a fidelity issue. You would probably never notice in a shooter or faster moving games. But in baseball where it can just sit there in your face for a while, it is noticeable.

Sometimes the performance issues lead to more than just fidelity problems. Our 360/PS3 versions are almost identical, but there is for example one replay-only effect (30fps mode for replay) where we really push edram to create a nice volumetric grass. We have not been able to recreate this on PS3 at a reasonable framerate. So in this case the problem extends beyond just a visual issue. We had to cut that effect totally on PS3.

If you reeeeally push edram is it surprising that you would have issues reproducing exact results elsewhere?

Anyhow maybe that chapter in GPU gems 2 on Photorealism in Botany could help you here. (haven't looked myself...or other literature perhaps)

I'm not a fan of changing msaa to 0x/2x/4x on the fly, the visual incongruity it causes in this case would to me would be worse that the artifacts we're seeing in the first place!

Only a quick suggestion. Don't change it on the fly then or try something else.

 

[Rockster]

It's not bad per se, it's good to be able to produce good graphics with less effort but every cycle (cleverly) spent by the GPU to fix art/content problems could be spent elsewhere.

Sure, but isn't the same true of artist, designer, and programmer time? No comment on my second question then?

 

[assen]

Joker454's point is "to match a fully utilized Xenos, you need SPUs to help the RSX".

Please take a few seconds to think why the following aren't counter-arguments:

"My favorite first-party XXX, which may or may not be seen yet by the public, looks great, therefore joker454 isn't right."

"A few senior people could abstract away all the difficulties in SPU programming in three weeks."

A valid counter-argument to joker454's point would be:

"I work on a multiplatform title, we max out Xenos and we match it in visual quality and framerate with the RSX without using any SPUs for rendering."

I would like to see "moderatable" offense the quoting of titles which have not been yet released (and in KZ2's case, more than a year away, if you believe the latest rumors) as proof of this or that. Uncharted is the current favorite, and for COD4 everyone is blindly parroting the piece of PR from IW which states that "both versions look the same" - notice the missing "and run at the same framerate".

 

[-tkf-]

notice the missing "and run at the same framerate".

And the same resolution.

 

[Betanumerical]

I would like to see "moderatable" offense the quoting of titles which have not been yet released (and in KZ2's case, more than a year away, if you believe the latest rumors) as proof of this or that. Uncharted is the current favorite, and for COD4 everyone is blindly parroting the piece of PR from IW which states that "both versions look the same" - notice the missing "and run at the same framerate".

They run at the same framerate

http://www.neogaf.com/forum/showpost.php?p=7410767&postcount=69

Same level of AA between 360/PS3, both @ 60fps.

Hate on the PS3 all you want guys, but we've done did it up right for the Sony lovers. Nothing has been 360 footage yet touted as PS3, you crack-heads! I'm not sure whats up with what you guys might have seen at E3, but we do have AA on the PS3.

As for the low res on those shots - I have no idea why these are so low. The full 720p versions have been available for a few days now.

Finally, the beta doesn't have a start date yet - just keep an eye on www.charlieoscardelta.com

 

[nAo]

No comment on my second question then?

I'd like to answer to your question but really..it's not worth the hassle. Fanboys (mis)quote me all the time on forums and websites, don't wanna run the risk of damaging the title I'm working on. Fool me once shame on you, fool me twice shame on me.

 

[nAo]

"I work on a multiplatform title, we max out Xenos and we match it in visual quality and framerate with the RSX without using any SPUs for rendering."

I work on a multiplatform title, when I joined the company Xenos was outperforming RSX on vertex shading by a vast margin, especially in simple color passes (zprepass, shadow maps).
Now (in those rendering passes) RSX is slightly faster than Xenos on average, with no SPU intervention, and I know it could be even faster (better performance were simulated through experiments on recorded push buffers) using memory in a different way.
Both platform shares EXACTLY the same high level shader code, Xenos wins when artists forget to check in some LODs and we render distant meshes with some insane triangle count, lots geometry that hardly generates any pixel and RSX gets vertex shader limited.

 

[MazingerDUDE]

I'd like to answer to your question but really..it's not worth the hassle. Fanboys (mis)quote me all the time on forums and websites, don't wanna run the risk of damaging the title I'm working on. Fool me once shame on you, fool me twice shame on me.

'working' ?

Does this mean you're now working on Heavenly Sword 2? Or maybe some PSN downloadable contents for HS1?

 

[djskribbles]

correct me if i'm wrong, but i think nAo is no longer with NT (hence "Location: San Francisco, CA").
also, from the looks of it, nAo is now working on a MP title.

 

[Terarrim]

Seems to me from your reply nAo (thanks for that) and your most recent comment, that its not just the Cell that many multi devs are having a hard time with its the RSX to. I would have thought that this would not be the case as its not as if Nvidia GPU's were invented for the PS3. And its reassuring that your getting such good results.

I wonder if a solution to this would be that Sony offers some type of RSX training course or something for those who buy the PS3 dev kit. Surely for vetran graphics programers the things that you are doing nAo (your knowledge set), would not be to difficult for other GPU programmers to understand given that the most Nvidia GPU coding would be common knowledge.

 

[patsu]

I work on a multiplatform title, when I joined the company Xenos was outperforming RSX on vertex shading by a vast margin, especially in simple color passes (zprepass, shadow maps).
Now (in those rendering passes) RSX is slightly faster than Xenos on average, with no SPU intervention, and I know it could be even faster (better performance were simulated through experiments on recorded push buffers) using memory in a different way.
Both platform shares EXACTLY the same high level shader code, Xenos wins when artists forget to check in some LODs and we render distant meshes with some insane triangle count, lots geometry that hardly generates any pixel and RSX gets vertex shader limited.

I had thought that having someone from an exclusive PS3 development team join a cross platform one would be the best way to spread Cell know-how. But I didn't think you would make such a move that quickly.

Potentially one more title for me to look out for.