proposition
Banned
Are you just adding every single quote that mentions Cell or SPUs rather than just the ones that have vaguely interesting information?
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
PS3 games list & SPE usages
- Thread starter MikeB
- Start date
proposition said:
I try to add the most interesting quotes to the original post(s). The fact that the 1080p game FlOw didn't use the SPUs doens't come as a surprise to me as I noted in the OP (for a semi-high profile Sony published game like Genji 2 it was), it's not a very demanding game. But IMO it's nice they note that they like other developers are now starting to dig deeper into the PS3's architecture.
I add comments I think may be of interest to some, you are free to do the same. I value information on smaller projects by small dev teams as well.
test_account
XP-39C²
I would guess that there arent exactly tons of "uninteresting" comments about the Cell and the SPU uses on PS3 games so i dont think its any "danger" that this thread will be flooded with "uninteresting" comments about the Cell and the SPU uses on PS3 games.
I like this thread, i think its cool to read about how the developers use and thinks about the SPUs. So for my sake i wish everyone would post info about this, even if its small og big info
I like this thread, i think its cool to read about how the developers use and thinks about the SPUs. So for my sake i wish everyone would post info about this, even if its small og big info
Strictly speaking it's not on topic, but could be interesting to keep an eye on things like this as to future plans.
IBM has officially announced the variant of Cell to be used in the Los Alamos Road Runner project, the double-precision enhanced PowerXCell 8i 8i.
http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=207602892
IBM has officially announced the variant of Cell to be used in the Los Alamos Road Runner project, the double-precision enhanced PowerXCell 8i 8i.
http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=207602892
belvedere said:
Some additional Ghostbusters comments from Terminal Reality:
http://www.gamasutra.com/php-bin/news_index.php?story=18671
I'll add some of these comments to post 32.
AgentOtaku
Member
cool info to read
BruceWayneIII
Member
New enhanced Cell-based IBM supercomputer announced(video):
http://www-03.ibm.com/press/us/en/presskit/24010.wss
http://www-03.ibm.com/press/us/en/presskit/24010.wss
Mr. Wonderful
Member
Instead of simply increasing the number of SPEs, wouldn't it be a better idea for IBM to make a dual core CELL or two CELLs on one die, so that the basic architecture might be more compatible with normal computer programming?
Killzone 2 tech interviews:
http://www.gametrailers.com/player/43389.html
http://www.gametrailers.com/player/43389.html
Behind the scenes:
http://www.gametrailers.com/player/43389.html
Interesting SPU comments:
""It's incredible to see huge levels and see the deferred rendering and note that on all the SPUs, even on the heaviest load were coming up to about 60%," Haynes said. "They weren't coming close to maxing out. .They had about 40% of space before they started tripping or saw slow down on some of the processes."
http://www.psu.com/Killzone-2-only-using-60--of-SPUs-overall-power-News--a0005629-p0.php
Furthermore, Haynes noted that the preview codes recently distributed to various media outlets were of an alpha build of the title, and have since been polished up . Visiting Guerilla Games in Amsterdam, Haynes was shown an updated version of the game, mid to late beta, noting that elements such as lighting, cut scenes, glitches, load times, have all been optimized."
My take: Looks like the game engine is maturing fast, but further game engine optimisations can bring down the 60% SPU load figure considerably. So potentially there should be even more headroom to do additional stuff.
Current processor loads can be easily determined through benchmark software, but future wizardly coding optimisation cannot be quantified.
http://www.gametrailers.com/player/43389.html
http://www.gametrailers.com/player/43389.html
Behind the scenes:
http://www.gametrailers.com/player/43389.html
Interesting SPU comments:
""It's incredible to see huge levels and see the deferred rendering and note that on all the SPUs, even on the heaviest load were coming up to about 60%," Haynes said. "They weren't coming close to maxing out. .They had about 40% of space before they started tripping or saw slow down on some of the processes."
http://www.psu.com/Killzone-2-only-using-60--of-SPUs-overall-power-News--a0005629-p0.php
Furthermore, Haynes noted that the preview codes recently distributed to various media outlets were of an alpha build of the title, and have since been polished up . Visiting Guerilla Games in Amsterdam, Haynes was shown an updated version of the game, mid to late beta, noting that elements such as lighting, cut scenes, glitches, load times, have all been optimized."
My take: Looks like the game engine is maturing fast, but further game engine optimisations can bring down the 60% SPU load figure considerably. So potentially there should be even more headroom to do additional stuff.
Current processor loads can be easily determined through benchmark software, but future wizardly coding optimisation cannot be quantified.
Doc Evils said:
Exactly. You only need one brain, as long as you have many hands doing the work. Its not much use to have two brains with one hand each...
That said, a 2/16 or 4/32 system as is currently experimented with would see incredible amounts of capability. Both for traditional developing that only likes to use brains and also for smart developing that uses the hands.
In addition to those PS3 Blu-Ray specs vs the 360 DVD drive, some more interesting data versus competing Blu-Ray movie players:
The bigger the above bar the better.
The shorter the bar the better.
More data:
http://forum.blu-ray.com/showthread.php?t=74197
With regard to gaming 12x 360 DVD drive vs 2x PS3 Blu-Ray drive:
A 360 dual layer DVD can hold 6.8 GB of game data, a single layer Blu-Ray can hold 25 GB of data, dual layer Blu-Ray disc can hold 50 GBs of data, it's claimed the PS3 is capable of supporting 1 TB Blu-Ray discs in the future (not that I think we will see PS3 games that big, but having technical headroom is great).
http://www.t3.com/news/pioneer's-1tb-blu-ray-discs-could-work-in-ps3?=37463
Some additional notes to keep everything in one post:
- 360 single layer DVD games are small enough to easily fit on any PS3's harddrive, which loads much faster. (this refers to identical 360 to PS3 ports, sporting identical assets)
- In combination with a default harddrive this provides the PS3 with a lot better streaming options (data caching) than it would have been without.
- Many 360 games ported to the PS3 are optimised for variable CD/DVD speeds which have been the standard for over a decade. To easily get around this, the part of the disc where a 360 Dual layer DVD is being read faster, the easy workaround (overkill in terms of what's needed) is to install this data onto the much faster harddrive. But starting from scratch it's much more easier to optimise for the far more easily predictable sustained reading speeds of Blu-Ray disc.
- Due to a much higher density of data per track of Blu-Ray disc, PS3 Blu-Ray discs have to spin much slower to read data quickly as compared to the 360. This results into less noise production and potentially fewer wearing issues over time.
- Blu-Ray discs are scratch resistant, you won't have to deal with scratched discs like many are experiencing with using the 360 DVD drive over time.
- For the usual game spreading data onto several discs will not exactly equal the amount of data on a bigger capacity disc.
What I mean if you have a game cramming 25 GB of data on a single layer Blu-Ray disc and trying to span this data onto dual layer DVDs in equal quality, this will normally not result into 25/6.8 = 3.67 (of course meaning 4) DVDs. This is due to even for linear games developers are usually re-using graphics and audio data from earlier sections in later sections as well. To prevent constant disc swapping developers will then duplicate re-used data for each disc. So this could mean spanning the content could actually require 5 or even 6 discs depending on the game.
With regard to disc swapping, it is sometimes believed by some gamers that by spanning data onto several discs this will not impact game design. In addition to what is mentioned above, developers will try to keep disc swaps and production costs (box size, disc production) to a minimum. An extremely linear game (like Blue Dragon) will usually lead to far fewer sacrifices, but by having a far more open world like GTA IV or other games where you will have to re-visit previously visited areas this poses far more a problem. Game designers in such cases will then usually make any kind of technical sacrifice to keep the game on no more than 1 disc.
Another comment from a multi-platform developing company, this time Team Ninja's boss, the company that brought us Ninja Gaiden Sigma on the PS3 and Ninja Gaiden II for the 360.
"For any developer that's been working on all of the platforms that are available today,I think they would agree that the PlayStation 3 is the most powerful system out there"
http://www.gamasutra.com/view/feature/3836/team_ninja_ready_for_more.php?
I hope we all can agree on that too. :lol
"For any developer that's been working on all of the platforms that are available today,I think they would agree that the PlayStation 3 is the most powerful system out there"
http://www.gamasutra.com/view/feature/3836/team_ninja_ready_for_more.php?
I hope we all can agree on that too. :lol
AranhaHunter
Banned
MikeB said:
I hope they can bring NGII Sigma to the PS3
DCharlie
Banned
woah, hold on a second there - how many games do we have on the PS3 that use 25gb of unique data?
surely reusing assets on a BR in the manner you suggest would incur horrible loadtimes, hence more likely the same data will be duplicated throughout the disk to improve load times?
Also - i'd suggest that games that truly use the BR are more likely to be first party efforts so it's probably all a bit moot anyways.
DCharlie said:
Although data duplication makes great sense to reduce seektimes if you have tons of free space left on the disc (which in worst case scenarios would still be many many times faster than actually swapping DVD discs), PS3 developers can also opt to cache to be re-used data onto the harddrive.
AranhaHunter
Banned
MikeB said:
This makes a lot of sense, but it still doesn't fully explain why some games like LBP and HSG OOB require a huge install.
AranhaHunter
Banned
MikeB said:
I thought LBP was 4GB, I think HSG is 5GB and GT5P is something like 2GB, Warhawk is around 1GB, MGS4 is what? 4GBs or something, Motorstorm is like 250MBs if I'm not mistaken....does Resistance and Ratchet have installs?
industrian
will gently cradle you as time slowly ticks away.
AranhaHunter said:
Originally it's around 800MB, but as I have all the expansion packs it's 1270MB on my XMB.
810MB according to the back of the box.
AranhaHunter said:
R1 - 269MB
R&CF - 417MB
R2 - 378MB
WickedLaharl
Member
Orlics said:
it's MikeB. what did you expect?
AranhaHunter said:
LBP requires 4 MB, with the latest update it's 480 MB.
- Uncharted: Drake's Fortune 256KB minimum, 13 MB for the latest update.
- Resistance 2 with latest update is 333 MB (300 MB minimum)
- Resistance 1 with latest update is 311 MB (130 MB minimum).
- Ratchet & Clank: Tools of Destruction with latest update 417 MB (137 KB minimum).
- Motorstorm: Pacific Rift currently uses 44 KB.
Even a 20 GB PS3 can easily manage dozens of such amazing games and of course they can always opt for a cheap 320 GB or more hard drive upgrade any time.
there's something incredibly important that i didn't see mentioned in that beyond3d thread and people always leave out.
i'd have to dig back into my old filesystem code and such, but the filesystem and the OS handling of the filesystem can make huge differences. i don't want to start mistating things and spread FUD, but i do remember that when i had to implement the UDF filesystem reader for some CE devices several years ago, we found that it significantly slower than say ISO9660, just as an example. my vague memory tells me that the issue was a significant increase in the number of seeks that were made just to get to the actual file data.
i can't compare it against FATX because i don't fully know the structure of FATX, but from what little i've seen it seems comparable to ISO9660, especially if it's even remotely similar to FAT16/FAT32.
it is possible to work around this, but most people never stop to think that this is a potential bottleneck or have a decent enough understanding of how the filesystem works in order to work around the issue. the simplest approach is to greatly reduce the number of files on the disc and rely instead on a few very large files. you can build your own embedded filesystems in these larger files and cache the game data as you need, which is almost always going to bring huge improvements over trying to rely on the OS file reading. you could potentially compress this on the PS3 and dedicate an SPU to decompressing and searching and such. (i've not dealt with the PS3 architecture to know if there are communications limitations that could prevent this from being useful or negate potential speed gains).
i hope that it isn't the case, but i wouldn't be surprised that most of the caching done for the filesystem is done in the form of simple block caching, which is really easy to blow, at which point you're going to begin running without a cache, and if the implementation is really poor, it could cause be worse than if there was no caching (such as attempted read-ahead block caching).
i remember trying to expand the freedb database on QNX (not neutrino) for a product i worked on once. this is essentially > 100K small files. about 0.5-2K a piece or so. the initial few files were super fast, but then something would happen and it seemed to blow the caches and then it got to a point where every additional file would take about a second to write.
i've seen several games which just have tons and tons of files thrown all over their discs and are no doubt the reason for their long loading times. i remember an old PS1 disc that i was curious what was going on so i popped it in and browsed the FS to see how bad it was. they probably had like every texture in a separate file. it was horrible.
i'd have to dig back into my old filesystem code and such, but the filesystem and the OS handling of the filesystem can make huge differences. i don't want to start mistating things and spread FUD, but i do remember that when i had to implement the UDF filesystem reader for some CE devices several years ago, we found that it significantly slower than say ISO9660, just as an example. my vague memory tells me that the issue was a significant increase in the number of seeks that were made just to get to the actual file data.
i can't compare it against FATX because i don't fully know the structure of FATX, but from what little i've seen it seems comparable to ISO9660, especially if it's even remotely similar to FAT16/FAT32.
it is possible to work around this, but most people never stop to think that this is a potential bottleneck or have a decent enough understanding of how the filesystem works in order to work around the issue. the simplest approach is to greatly reduce the number of files on the disc and rely instead on a few very large files. you can build your own embedded filesystems in these larger files and cache the game data as you need, which is almost always going to bring huge improvements over trying to rely on the OS file reading. you could potentially compress this on the PS3 and dedicate an SPU to decompressing and searching and such. (i've not dealt with the PS3 architecture to know if there are communications limitations that could prevent this from being useful or negate potential speed gains).
i hope that it isn't the case, but i wouldn't be surprised that most of the caching done for the filesystem is done in the form of simple block caching, which is really easy to blow, at which point you're going to begin running without a cache, and if the implementation is really poor, it could cause be worse than if there was no caching (such as attempted read-ahead block caching).
i remember trying to expand the freedb database on QNX (not neutrino) for a product i worked on once. this is essentially > 100K small files. about 0.5-2K a piece or so. the initial few files were super fast, but then something would happen and it seemed to blow the caches and then it got to a point where every additional file would take about a second to write.
i've seen several games which just have tons and tons of files thrown all over their discs and are no doubt the reason for their long loading times. i remember an old PS1 disc that i was curious what was going on so i popped it in and browsed the FS to see how bad it was. they probably had like every texture in a separate file. it was horrible.
MikeB said:
BTW, Motorstorm 1 received a lot of updates, now up to version 3.1 with rumble, graphics enhancements, additional time mode, load speed ups for online, etc.
It originally used 725 KB and is now up to 233 MB.
An issue with launch games like Motorstorm and Resistance: Fall of Man was that each level had to be loaded into memory in one go, they only streamed their high quality 7.1 lossless audio.
Resistance 2 now uses streaming for graphics data as well and it seems like Motorstorm: Pacific Rift does this as well, at least loading times for the huge tracks are now very good.
industrian
will gently cradle you as time slowly ticks away.
RavenFox said:
Fun.
DCharlie said:
With regard to GTA IV Rockstar guys already stated they had to cut out stuff and they won't have enough space for future games on DVD.
Sacrifices can be lower quality assets like audio and textures, you can opt to make the game smaller in size or length or for example you can cut out extras like in GTA IV's case fewer radio broadcast/songs, TV shows, etc.
MikeB said:
it's a nice OS with some unique features, and most of that was improved greatly with neutrino.
i'm 100% positive it's still an issue on Blu-Ray discs. the basic point is that the organization matters a lot and using UDF can incur a lot more seeks just searching for files than other filesystems. i've done the benchmarking before and found the optimizations to overcome it.
even if the core drive is faster, using a filesystem like UDF without a decent caching mechanism code can kill any advantage it has unless the developer has just a handful of huge files and a lack of directories.
even with mirroring of data across the disc, if the caching isn't handled properly, you've still got a bottleneck there because of the number of seeks required to get to the filesystem data and back to the actual data. if you're talking even 100ms average seek time and you have to multiply that by 2-3x because the filesystem is inefficient, it nullifies any potential speed increase that have in reading contiguous data.
i've got a birthday dinner tonight, but i can check up the data and do some charts and graphs for you and you can add it into your calculations with the seek time and such
AranhaHunter
Banned
MikeB said:
Tntnnbltn said:
That explains a lot, so 1st and 2nd parties have acceptable levels of installation except for HSG5, what's up with Clap Hanz?
jonabbey said:
It appears to be QNX Neutrino (RTP) with on top Sony embedded and NetBSD technologies. QNX is Linux-like, but far superior in nearly all benchmarks (and like already stated provides many unique features). Each SPE is also running a mini-kernel, they are functioning like tiny seperate systems with their own tiny cache, super fast work memory (as fast as cache), DMA controller, etc.
Kazuma Kiryu
Member
MikeB said:
he's actually right, my pal box also says 4mb minimum, but it installs like 600mb