*Game Development Issues*

[joker454]

The install is only 3+ Mb (3340Kb), not 3Gb.
http://forum.beyond3d.com/showpost.php?p=1153113&postcount=634

From this link:

http://kotaku.com/384422/which-version-of-gta-iv-should-you-buy-we-compare-the-ps3-and-360-versions

"The PlayStation 3 version has a mandatory 3.4 GB hard drive installation that will be the longest 7 minutes and 20 seconds of your life if you're chomping at the bit to play."

Maybe that 3mb is for save games.

 

[patsu]

Hmm... if that's the case, they should change the case description to 3Gb. The mandatory install would make it more than 3340Kb minimum.

 

[makattack]

Hmm... if that's the case, they should change the case description to 3Gb. The mandatory install would make it more than 3340Kb minimum.

Perhaps 3MB is only for the save game file -- of course, this doesn't make sense if there's a 3GB required install, but this is off-topic. 7 minutes seems awfully long to install 3MB of data, unless that data was spread all over a disc.

I do find it curious that there isn't even an optional HDD install on the 360. I don't know if that's a policy decision/requirement (on MS' part) or if they just didn't want to bother writing a component for installing or caching to HDD specific to the 360.

 

[betan]

Hmm... if that's the case, they should change the case description to 3Gb. The mandatory install would make it more than 3340Kb minimum.

3MB mimimum label doesn't mean 3MB for saves 3GB for installation.
It means the game requires at least 3MB free space to run properly.
If for some reason your HDD has less than 1GB left, the game will still run.
What's the confusion?

 

[betan]

There is a reality going on today that people won't like, but here goes anyways. It's better to leave 360 performance unused, and have both 360/PS3 versions run at parity.

Fortunately for MS, they have more than enough exclusives that can be used to prove fanboys wrong.

 

[patsu]

I'm curious. Is "no install" a good way to indicate no HDD streaming ? Insomniac implemented streaming without multi-gigabyte install right ?
http://www.insomniacgames.com/tech/articles/1107/files/texture_streaming.pdf

Can't the developer cache the data to HDD no-demand/temporarily (i.e., without install) on the 360 as long as space is available ? Is there any way to measure if HDD space usage changes during GTA4 gameplay ?


EDIT:

3MB mimimum label doesn't mean 3MB for saves 3GB for installation.
It means the game requires at least 3MB free space to run properly.
If for some reason your HDD has less than 1GB left, the game will still run.
What's the confusion?

Possible. May be. I am wondering whether streaming requires a 3Gb install myself. There is probably a baseline which is covered in the 3Mb install., if true.


EDIT 2:
I read Kotaku's note about superfluous SIXAXIS use in GTA4. I think we are missing an opportunity here. In DMC4 (auto mode), the player can mash the same button to generate a myriad of effective combos.

I feel that SIXAXIS, like Wiimote, is a high level control device. It generates very little returns for long time players who prefer the traditional controllers anyway. Given the GTA4 sensation, they could have introduced SIXAXIS as a simplified/high level control for casuals. e.g., Hold SIXAXIS like a rock and point an on-screen cursor to indicate whether they want the character to go (immersion be damned), hit anyone with the SIXAXIS during combat to generate the right combos (like DMC4), or carjack by hitting the vehicles with SIXAXIS.

This avoids the complains of hardcore gamers, and if implemented correctly, can allow casuals to get a taste of the game world. It may be awkward to hold the controller for long hours, but it may be a reasonably good start for some follow ups. Alternatively, they can do it with PS Eye + any IR remote (just like the the Desktop VR demo -- using a cheap negatives cap to cover PS Eye's lens). They will get free marketing when MS does heavy advertising on GTA4.

 

[FirewalkR]

The sad part is that nobody want to go though the expanse of creating something custom on the 360 / rare who seems iddle...
I don't even speak about trying to reach the dedication some sony team go through to put out technical marvels...

I believe if there's a team capable of truly showcasing the 360 it is indeed Rare. Kameo had very impressive tech at launch. And they're not idle, they're showing Banjo in 2 weeks, May 13th.

However, Sony does seem to have a lot more teams with the same level of technical talent.

 

[betan]

Possible. May be. I am wondering whether streaming requires a 3Gb install myself.

I'm not sure I understand what you mean.
The engine is clearly capable of streaming from optical disc, and even if it's slower, the texture popin implies the streaming is not hard realtime anyway.

 

[MazingerDUDE]

There is a reality going on today that people won't like, but here goes anyways. It's better to leave 360 performance unused, and have both 360/PS3 versions run at parity. It just makes good business sense. Why? There's many reasons, but in the end having a deficient PS3 version gives bad press, and reflects badly on the studio. On the other hand, putting out two versions that are identical reflects very nicely on the studio and yields lots of good press. Whether or not technical boundaries of a given platform are maxed out turns out to be irrelevant. Making a good looking game is important, but making it run the same on both machines is becoming even more important. If that means leaving performance/memory unused, then so be it. Is this happening today? Yes. Does it suck? Perhaps. Does it really matter? I don't think so. In the end, the consumer doesn't really know whats going on, they are just happy that their version is running nice. 360 owners still get a good looking game and are happy. PS3 owners get a version that runs the same as the 360 version, and they are happy. Everyones happy, which is good for business.

Seems like your take on multi platform game development has changed drastically in last few month :smile:


http://forum.beyond3d.com/showpost.php?p=1108753&postcount=51

 

[patsu]

I'm not sure I understand what you mean.
The engine is clearly capable of streaming from optical disc, and even if it's slower, the texture popin implies the streaming is not hard realtime anyway.

I was refering to HDD streaming betan.

 

[betan]

I was refering to HDD streaming betan.

I guess I'm little slower than usual today, but HDD streaming requires HDD install.
Or are you talking about hdd caching ala Uncharted or RCF?
Or hdd caching like Oblivion?

 

[patsu]

I guess I'm little slower than usual today, but HDD streaming requires HDD install.
Or are you talking about hdd caching ala Uncharted or RCF?
Or hdd caching like Oblivion?

Yes, like the R&C texture streaming from HDD (without gigabytes install).

 

[betan]

Yes, like the R&C texture streaming from HDD (without gigabytes install).

OK, but RCF streams from disc and cache (or prefetch) some of the data to HDD, I'm not sure I would call this HDD streaming.

In anycase GTA4 reportedly installs 3GB in advance, and that data do not look like sys cache data of RCF to me, if the statement "HDD streaming requires 3GB" implies that.

 

[nAo]

You cannot index into sampler registers on RSX.

You can index into sampler registers on NV4x/G7x if your index is a loop index.

It's better to leave 360 performance unused, and have both 360/PS3 versions run at parity.

It often happens both ways.

 

[Mefisutoferesu]

You can index into sampler registers on NV4x/G7x if your index is a loop index.

It often happens both ways.

Any chance we could get some examples, nAo?

EDIT:: Many thanks, nAo. Always happy to learn something new.

 

[nAo]

Any chance we could get some examples, nAo?

Referring to the example Joker made here (not being able to index texture samplers) I had the same issue while trying to code a fast implementation of parallel-split shadow maps.
Eventually I found a solution (based on RSX) that not only doesn't require to index texture samplers but it's also faster than any implementation based on indexing samplers while allowing me to take a variable number of samples from the shadow maps (without employing dynamic branching)
Now if I had to re-work the same technique on a multiplatform game the 360 version would probably be quite slower.
The same kind of problem can happen while working with SPUs as you can got carried away to later realize that you really can't afford all those computations on a 360..

Once again nothing new under the sun: we have two advanced platforms, each has its strengths and you when you work on multiplatform games you make compromises every day.
360 engineers don't like when they have to scale down stuff that doesn't run well on PS3 and PS3 engineers don't like finding themselves in the same situation

 

[joker454]

Seems like your take on multi platform game development has changed drastically in last few month

Yeah, it has. I don't have to like it though Some parts of that post still apply, but fiscal realities have change my line of thinking there. So it goes.

Referring to the example Joker made here (not being able to index texture samplers) I had the same issue while trying to code a fast implementation of parallel-split shadow maps.
Eventually I found a solution (based on RSX) that not only doesn't require to index texture samplers but it's also faster than any implementation based on indexing samplers while allowing me to take a variable number of samples from the shadow maps (without employing dynamic branching)

Hmm, I know you can use the hardware 4 tap pcf on rsx to get multiple samples from a given shadow map and hence a cheap "percent coverage". That part I know is rsx specific. I'm puzzled though as to what other technique would be rsx specific with regards to eliminating the need to index samplers. In the shadows case you can just combine all the shadow splits into one uber wide texture, and hence no indexing needed. But that idea isn't rsx specific. Unless you've cleverly determined some way to use the hardware 4 tap pcf in and of itself to eliminate the need for multiple shadow maps and indexing?

Still, there are cases outside of shadows where indexing samplers is very handy!

 

[nAo]

Hmm, I know you can use the hardware 4 tap pcf on rsx to get multiple samples from a given shadow map and hence a cheap "percent coverage". That part I know is rsx specific. I'm puzzled though as to what other technique would be rsx specific with regards to eliminating the need to index samplers. In the shadows case you can just combine all the shadow splits into one uber wide texture, and hence no indexing needed. But that idea isn't rsx specific. Unless you've cleverly determined some way to use the hardware 4 tap pcf in and of itself to eliminate the need for multiple shadow maps and indexing?

I used PCF as well but that's completely orthogonal to the samplers indexing issue.
You can indeed remove indexing having all your shadow maps stored in a big texture, but such an implementation would map well to 360 as well (though it probably wouldn't be optimal).
In my case I simply replaced a shader that index samplers (or samples different shadow maps within the same texture) with a shader that always sample only one shadow map (no need to compute multiple sampling co-ordinates or to sample multiple shadow maps) via multi pass and depth bounds test.
This allowed me to switch to a different shader per shadow map split so that I could take a variable number of samples per split with no dynamic branching involved.
I wrote about it a few months ago on my blog

 

[Mintmaster]

In my case I simply replaced a shader that index samplers (or samples different shadow maps within the same texture) with a shader that always sample only one shadow map (no need to compute multiple sampling co-ordinates or to sample multiple shadow maps) via multi pass and depth bounds test.
This allowed me to switch to a different shader per shadow map split so that I could take a variable number of samples per split with no dynamic branching involved.
I wrote about it a few months ago on my blog

One think I've always wanted to ask you is how you combined multipassed, deferred shadows with MSAA.

Obviously you have to store the shadow term in either a texture (that you sample from later in your pixel shader) or the alpha channel of the rendertarget. I would think that the former would give you some artifacts, and the latter would be tough to combine with NAO32 or almost any non-trivial lighting equation.

A lot of what joker454 is talking about stems from him not wanting/able to multipass due to these sorts of complications, which are probably even worse for cases not related to shadowing.

 

[liolio]

In regard to recent events I think we should open another thread about this but from a busyness perspective.
the stakes are really hight, press have a lot pression.

It's not about derailing te whole forum with fan wars but come on the Quaz51 effect is becoming huge!

Editors seem to put a lot of pression on the press (on top of development effort) to not split user base for the multi platform games.

I'm sure one would think that I want to start a little war, but it's up to the mods to decide if it would worse it (ie I won't open this thread).
But major sites holding on screenshots/prvoding maybe wrong informations etc. sorry it 's huge.
As the market but bigger this could have been predicted, but what if we can't discuss this here, where?

And it won't be possible to make everybody about it I 'm plainly conscious, and it's likely that it could give the mods extra work... so their decision but this busyness fact is important IMHO we can't pass.