Question for developers... PS3 and framerate

[joker454]

Crowds have a lot of opaque overdraw, so early z-reject and front-to-back drawing seems like an obvious and powerful optimization that should largely (almost completely) negate the bandwidth advantage of eDRAM. Or am I missing something?

The sorting part is actually simple since generally the seat in front of you will also be lower, hence you can just sort seats by increasing 'Y' to get a front to back draw order for field area cameras.

The problem is that crowd is instanced, so even though you see 50,000 in a stadium, there may only actually be 30 or so individual crowd members that are repeated. When you draw them, you'd want to setup crowd dude#1 and draw all instances of them first since you'd be using the same geometry, texture, etc for every draw call. Plus there is other trickery you can do as well when you are drawing many of the same guy over and over. But, this 'instanced' crowd member will be scattered all over the place, which doesn't work well with front to back sorting alas.

You can do a z-prepass, but crowd can be vertex heavy so it's not worthwhile.

 

[Nesh]

Something I was wondering about. I noticed in the Heavenly Sword Demo I played that there are bars above and below screen. Is that what Capcom did with Resi4 to maintain high detail and steadier framerates?

 

[Betanumerical]

This kinda fits in here I guess sincei t would effect framerate. I'm having a arugment with someone at the moment who says that developers only get access to 5 of the SPU's and that two are permantly taken for the OS/they cant rely on using them, is this true or does the developer have access to all 6? I know this has been asked a lot but searching would be impossible since the words are to short.

 

[Dot50Cal]

Something I was wondering about. I noticed in the Heavenly Sword Demo I played that there are bars above and below screen. Is that what Capcom did with Resi4 to maintain high detail and steadier framerates?

That might be your TV, I have nothing of the sort when I use a capture card on HS Demo
http://dot50cal.the-horror.com/b3d/hsdemo/

 

[jobboy]

Another sports game developer, Konami, experiencing a frame rate challenge with the PS3 version of their game title.



http://ps3.ign.com/articles/827/827315p1.html

even worse they dont know how framerate drops

 

[Nesh]

That might be your TV, I have nothing of the sort when I use a capture card on HS Demo
http://dot50cal.the-horror.com/b3d/hsdemo/

Weird. Thats the only game that has black boards. Everything else is full screen on my TV

 

[Arwin]

even worse they dont know how framerate drops

Maybe they have the 24 frame option enabled for BluRay?

 

[ShootMyMonkey]

even worse they dont know how framerate drops

I don't follow how that's shocking. If they knew, there wouldn't really be any problem... assuming that they had some countermeasure that was reasonable. Of course, if they didn't have a countermeasure that was reasonable (which is entirely possible), then that would be the issue, wouldn't it?

 

[Arwin]

I think they managed to fix the issue using a patch, but now the PS3 version is plaged with online issues apparently. Both issues seem to be recognised and still being worked on.

http://www.pro-g.co.uk/news/26-10-2007-6760.html

disclaimer: this site has a very bad reputation for being very much anti-PS3, but to be honest, in this case I expect they're probably right.

EDIT: at the end of the 360 review they post that the online issues are actually across all platforms. Nice of them to edit this into the http://www.pro-g.co.uk/news/26-10-2007-6760.html article (not)

 

[Graham]

One of our dear leaders would prefer his subjects not link to the 'jbooth' blog. Threads have already been closed over that particular article....
Don't shoot the messenger please

 

[Betanumerical]

This is semi related, I think we have all heard the stuff about the Cell having very poor integer performance, but I couldn't find any results or articles on this. I noticed on IBM's site that the latency for integer when compared to float is only 1 cycle higher, is this a lot or little?.

 

[Acert93]

This is semi related, I think we have all heard the stuff about the Cell having very poor integer performance, but I couldn't find any results or articles on this. I noticed on IBM's site that the latency for integer when compared to float is only 1 cycle higher, is this a lot or little?.

Cell has good Integer performance.

 

[Betanumerical]

Cell has good Integer performance.

Good being how good? in comparison to its FP performance and the XCPU's integer performance, its hard trying to cut through the poo poo.

 

[Shifty Geezer]

In terms of processing integer maths, Cell's integer performance is comparable to its FP performance. What people call 'integer performance' though is something else, basically meaning an ill-defined term. It might include working on heavily branching code, large scatter/gather problems, or any other sundry activities. However, if you design your algorithms for Cell it has shown it's at no disadvantage there either, with benchmarks showing even in tasks people considered Cell ill-suited for, it can out-perform conventional processors. The same has been shown in GPGPU code too. The idea that maths-strong SIMD architectures are only good at maths-crunching tasks seems to be naive and from a perspective of people hadn't really tried to adapt tasks to the new architectures, instead mapping their existing solutions to them and seeing they wouldn't work well.

 

[SPM]

Good being how good? in comparison to its FP performance and the XCPU's integer performance, its hard trying to cut through the poo poo.

Not sure what you are talking about here, integer arithmetic, or integer performance meaning code that isn't limited by floating point computations. Either way, if you have code optimized to fit in local store and doesn't do context switching or waiting for another process, an SPE can run rings around an XCPU core, and you have seven of them on Cell in addition to the PPE. There are some things that the SPE isn't good at doing like quick context changes, long branches, and handling certain types of large data structures, but you have the PPE for that.

 

[jobboy]

i read in different forums people reporting different framerate experiences while playing the same game. IIRC Cell is 8 spu's with one inactive for redundancy, is it possible that similar differences come from another faulty or broken spu, so games are running under 6 spu's instead of 7??

 

[rekator]

Games run sure on 5 SPU … Sony OS keep 1 SPU and can request one other…

 

[Betanumerical]

i read in different forums people reporting different framerate experiences while playing the same game. IIRC Cell is 8 spu's with one inactive for redundancy, is it possible that similar differences come from another faulty or broken spu, so games are running under 6 spu's instead of 7??

The number of SPU's that can be used by a programmer at any given time in all the PS3's in the world is 6, 1 is redundancy as you said and one is OS tasks. It could be possible that the people who are experiencing higher framerates are playing at a lower resolution(480/576P) this would definitely increase the framerate, other then that I can't see much more increasing the framerate.

 

[_phil_]

Games run sure on 5 SPU … Sony OS keep 1 SPU and can request one other…

you did read that on what part of the sdk doc ?
I can't find that line.

 

[rekator]

you did read that on what part of the sdk doc ?
I can't find that line.

If I remember well it's in a old frimware discussion, may be with actual frimware it's only 1 SPU.
But this (a possible other SPU request by OS) can explain some framerate drop on some PS3… if they are on "old" frimware…
And it's not for some "console war", I really intrigue with this framerate drop on some PS3… don't seem logical…may be there are some tasks in the behind?