needlesmcgirk
Member
^ Agreed. Only takes a couple of bad apples to ruin the whole batch, ey?
-
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
Psychotext
Member
Read what he posted... carefully.sakuragi said:
On the actual subject, I feel that the PS3 may indeed be slightly more powerful than the 360... but the amount of processor power / memory locked off for the operating system coupled with the weaker GPU and difficulty getting this power out of the SPEs means that it's going to be extremely rare that we're able to see this advantage. We've been through all of this before, multiple times. Even going so far as to work out exact peak performance figures taking into account the things mentioned above as well as industry standard efficiency expectations for code running across this many cores.
Hardware differences or not, it's time the games started talking for themselves.
Beer Monkey
Member
sakuragi said:
I thought it was a thread about how games are being developed on the PS3 by different developers, but somebody wants to turn it into the millionth 360-versus-PS3 thread.
Sho_Nuff82
Member
Jesus christ at this thread. It's like the Emotion Engine all over again.
@ ex0du5
I agree, the more the SPEs can take workload off the PPE and RSX, the more the RSX and PPE have power to spare to do other stuff. It's up to the developers to decide which code should be best executed on the PPE, RSX or SPEs for their particular game.
As you can judge from my avatar like the people who developed Super Stardust HD (I am currently ranked 73 !) I come from an Amiga background. The Amiga was very powerful for its time, this was mainly achieved by its custom chips which took the workload off the main CPU. The system was very innovative with regard to dividing tasks amongst simultaneously operating custom chips.
Early Amiga games didn't really use the custom chips to their advantage, but were mostly straight ports from the cheaper Atari ST (some early ports were even inferior). However with later games the custom chips were put to some good use. Regarding the Housemarque guys, Terramarque pushed the A500 well with Elfmania and Bloodhouse did a great job with the orginal Stardust (great game!), the prequel to Super Stardust (even better!) for the A1200.
Elfmania for the A500 (a lowend Amiga system released in 1987 based on the Amiga 1000 from 1985):
http://youtube.com/watch?v=zCqLqj3QOEQ
I agree, the more the SPEs can take workload off the PPE and RSX, the more the RSX and PPE have power to spare to do other stuff. It's up to the developers to decide which code should be best executed on the PPE, RSX or SPEs for their particular game.
As you can judge from my avatar like the people who developed Super Stardust HD (I am currently ranked 73 !) I come from an Amiga background. The Amiga was very powerful for its time, this was mainly achieved by its custom chips which took the workload off the main CPU. The system was very innovative with regard to dividing tasks amongst simultaneously operating custom chips.
Early Amiga games didn't really use the custom chips to their advantage, but were mostly straight ports from the cheaper Atari ST (some early ports were even inferior). However with later games the custom chips were put to some good use. Regarding the Housemarque guys, Terramarque pushed the A500 well with Elfmania and Bloodhouse did a great job with the orginal Stardust (great game!), the prequel to Super Stardust (even better!) for the A1200.
Elfmania for the A500 (a lowend Amiga system released in 1987 based on the Amiga 1000 from 1985):
http://youtube.com/watch?v=zCqLqj3QOEQ
Crayon Shinchan
Aquafina Fanboy
Psychotext said:
By extremely rare, I suppose you're right if you discount all Sony first party games.
sakuragi said:
All I have to say is: PS3forums -------------> that way.
I'm sorry but I can't believe I'm reading this. Both consoles have a ton of potential. I'm so excited about what's around the corner, for both platforms.
A year ago, sure it was fun. Xbox this, PS3 that. Hyperbole here, FUD there. But the consoles are here now. We can work with them, see reality rather than promise. We know PS3 kicks ass in certain areas, and the 360 excels at others. Both have their limitations. Can't we just accept that they're great in different ways?
Zombie James
Banned
MikeB said:
Holy shit, that looks awesome for 1985 hardware.
@ soco
I know you didn't ask me, but IBM claims this to be the case. The Cell inside the PS3 is between 2.5-3 times more powerful than the Xenon in general (at some specific tasks it may be much more!).
From Forbes, compare IBM's given figures:
http://www.forbes.com/free_forbes/2006/0130/076.html
Parts from the article, complete with a honourable Amiga reference!
"This chip will give you performance that is not achievable with any other architecture." Adds H. Peter Hofstee, an IBM scientist and the chief architect of a key part of the Cell chip:"
" In the early 1980s the chip in the Amiga home computer far outraced those in the Intel line, but Intel conquered the market anyway. "
I personally attended various IBM presentation, including this one at an Amiga party:
Mikael Haglund, technical specialist of IBM Sweden educating a few hundred fellow Amigans about the Cell's strongpoints (2005):
I know you didn't ask me, but IBM claims this to be the case. The Cell inside the PS3 is between 2.5-3 times more powerful than the Xenon in general (at some specific tasks it may be much more!).
From Forbes, compare IBM's given figures:
http://www.forbes.com/free_forbes/2006/0130/076.html
Parts from the article, complete with a honourable Amiga reference!
"This chip will give you performance that is not achievable with any other architecture." Adds H. Peter Hofstee, an IBM scientist and the chief architect of a key part of the Cell chip:"
" In the early 1980s the chip in the Amiga home computer far outraced those in the Intel line, but Intel conquered the market anyway. "
I personally attended various IBM presentation, including this one at an Amiga party:
Mikael Haglund, technical specialist of IBM Sweden educating a few hundred fellow Amigans about the Cell's strongpoints (2005):
Crayon Shinchan
Aquafina Fanboy
@MikeB
I hope Xbots don't try to discredit IBM on that point... the guys that made both chips.
I hope Xbots don't try to discredit IBM on that point... the guys that made both chips.
soco said:yeah you probably don't wanna throw that number out much anymore. if you wanna do it, you should at least consider that each of the 360 cores can handle two simultaneous threads (there are restrictions on how that works, but in essence, it's possible) and adjust the math accordingly, but still, it's a useless calculation in a technical discussion
Hmm... is this right? I thought that it could issue multiple instructions to several different "task oriented" pipelines (similar to the Pentium's U and V Pipes).
Edit: Looks like the technique Xenon is basically Hyper-Threading. And while that can yield good performance in certain circumstances, it's hardly akin to implying the 360 can run 6 threads at full-tilt.
artredis1980
Banned
May 2005 - September 2007
People have not realised, PS3 can output 50 sparks and 360 can output 30 sparks
but 360 can output better looking sparks because of that GPU.
and that is what matters in the end. quality over quantity
People have not realised, PS3 can output 50 sparks and 360 can output 30 sparks
but 360 can output better looking sparks because of that GPU.
and that is what matters in the end. quality over quantity
Psychotext
Member
We're not talking about using the SPEs (which you have to do to match the 360's power anyway), when I say rare we're talking about getting above 75% efficiency out of them - which may sound easy but in the real world is a pretty major challenge.Crayon Shinchan said:
Psychotext
Member
Your link is interesting... because it claims there that the CELL can do 230 gigaflops and the 360 processor does 77. Except we know (IBM documents) that the CELL is theoretically capable of 25.6 gflops * 8 (1 PPE, 7 SPE) = 204.8 for single precision instructions at 3.2ghz (The PS3 CELL has one of its SPEs disabled for yield).MikeB said:
http://www-128.ibm.com/developerworks/power/library/pa-cellperf/
We also know from IBM documents that the xenon is capable of 115.2 gflops.
(Note, that obviously the above numbers don't take into account anything like amount of processor time etc reserved for the operating system and are pure theoretical peak numbers)
@ Psychotext
According to IBM its peak perfomance is really around 70 gflops range. Also if programmed well the Cell should allow for a much higher level of efficiency.
Yes, currently game developers or hosted 3rd party operating systems can use 6 SPEs. But first of all that's not a big deal right now, as developers are currently only scratching the surface of the PS3's performance and are overwhelmed by the headroom this architecture gives them. IMO reserving 1 SPE for the OS may well be an excellent approach from a design perspective (from a marketing point of view, it's less than optimal, as it's being used against them by people who don't understand the reasons) as Sony is still building features on top of the CellOS/GameOS, so they can rest assured any SPE perfomance the OS needs will not interfere with games.
Obviously any OS functions running on the SPEs won't take perfomance away from the PPE or other SPEs used by games, so I don't agree with some people who completely neglect the existance of the 7th reserved SPE regarding performance.
According to IBM its peak perfomance is really around 70 gflops range. Also if programmed well the Cell should allow for a much higher level of efficiency.
Yes, currently game developers or hosted 3rd party operating systems can use 6 SPEs. But first of all that's not a big deal right now, as developers are currently only scratching the surface of the PS3's performance and are overwhelmed by the headroom this architecture gives them. IMO reserving 1 SPE for the OS may well be an excellent approach from a design perspective (from a marketing point of view, it's less than optimal, as it's being used against them by people who don't understand the reasons) as Sony is still building features on top of the CellOS/GameOS, so they can rest assured any SPE perfomance the OS needs will not interfere with games.
Obviously any OS functions running on the SPEs won't take perfomance away from the PPE or other SPEs used by games, so I don't agree with some people who completely neglect the existance of the 7th reserved SPE regarding performance.
artredis1980
Banned
MikeB said:
is that peak performance in actual game code or what for processing data optimized for the cell. and is that programmed efficiency level in terms of game code?
Game code involves using both GPU and CPU, cell was optimized to use its own structure only, and most of that peak performance is theoretical. Cell will only be as powerful as the way you design what software to work for it. As you have seen Cell is great in Image manipulation and processing data which requires multiple sources together. Its the same case with 64 bit processors and Core 2 Duo. You can optimize it all you want but just becuase you have dual cores doesnt mean the machine runs twice as fast nomatter how much you design for it.
the GPU is passed the data from the CPU to output the video,
Cell =
= =
= -> GPU = -> (TV)
= =
=
=
=
the GPU can only output so much.
if Cell is processing 100 billion Pixels but the GPU can only output 20 billion at a time, nomatter how much optimization you do for the Cell, the GPU will ALWAYS stand in its way of delivering what Cell is capable of
Its what happened with PS2 with its weak GPU over its powerful CPU and same is happening with PS3
Crayon Shinchan
Aquafina Fanboy
Psychotext said:
Isn't that true of any real world applications? You can't get 100% peak efficiency out of the chip the whole time; some scenes will naturally be less taxing then others...
e.g been first place with no other cars in sight around a tight corner (i.e. minimal draw distance) vs last place on a field of 16 cars down a long decorated stretch (e.g. New York).
I mean, I understand your argument; even extremely talented developers are going to find it difficult to extract 95% of the systems power under peak load; because the skill and effort to exploit that last 5% can be great.
But I don't know where you're getting your numbers and your comparisons from; "which you have to do to match the 360's power anyway"? Saying it like there's an on/off switch, and turning it on will only just match the X360.
As laymen... the only valid comparisons we really have to go off are games... and we can only get a feel for it; we can't seperate developer talent with system power. Hell, we even find it difficult to seperate artistic merit from technical merit, so been able to seperate the technical merit of the hardware and software is only something that can be loosely guessed at by us.
That said... seeing Sony's top tier first party stuff... KZ2, GT5, leaves me with little doubt that at optimal performance, even with the GFX chip disadvantage and the RAM disadvantage, the abilities of the PS3 outstrip those of the X360 by what could be considered a visually noticeable margin.
That said, it's Sony's failing for failing to anticipate a market place like the one now; where it becomes less profitable for developers to exploit the power of the PS3 then of the X360. But that's a situation independent of the potential of both consoles.
Psychotext
Member
Got any links for that other than the article above? Because it's the first time I've seen anything but 115.2. Also, there's no way you're going to get better efficiency over 8 cores compared to three... it's just not going to happen.MikeB said:
Even the document I posted above states 74% as ideal real world max efficiency for the SPEs.
Psychotext
Member
It's a simplification. We know what the 360 can output from three cores and we know what the PS3 can do with just the PPE. Without the SPEs the PS3 is no-where near as powerful as the 360.Crayon Shinchan said:
RubberJohnny
Banned
SPEs are essential for 60FPS and Sparks.
Doom Bringer school of design
ape2man
Member
Come on guys dont make this into a 360 vs ps3 war.
Anyway what i understand about the cell is, That its a beast of mathimatical
calculations and its way more powerfull then anything on the market.
It was designed to be the new work horse of super computing and graphical applications.
Like HD calculations and anything else that needs heavy calculations that a normal x86
cpu was not designed to do.
Anyway what i understand about the cell is, That its a beast of mathimatical
calculations and its way more powerfull then anything on the market.
It was designed to be the new work horse of super computing and graphical applications.
Like HD calculations and anything else that needs heavy calculations that a normal x86
cpu was not designed to do.
Psychotext said:
I'm not sure if IBM discussed it ever, but it's the opinion of many that the flops rating of both the Xenon PPEs and the Cell PPE are inflated, and that the 'real figure' should be closer to 70Gflops for Xenon, and the Cell PPE's contribution to its flops rating should also be reduced.
edit - actually, i can't remember the exact figures..just that they're lower than the officially provided peaks.
Psychotext said:
The achieved flops will depend entirely on the task and its implementation. There is no one figure to encapsulate "real world max efficiency for the SPE", only efficiency of code and implementations. For example, code has been run that's achieved 90+% of Cell's theoretical capability, but no one's going to hold that up as "the SPE's efficiency". They're only as efficient as the code that runs on them, and that will differ in every application.
IBM does claim, however, that it is more possible to get closer to Cell's theoretical performance than on many other architectures because of its memory architecture, which is designed to keep the computational units fueled with data. This is also the reason Cell has been demonstrated to exceed the performance of other chips in various tasks in multiples well beyond that which its floating point figures would suggest.
There's also nothing to say that there's no way you'll get better efficiency with more cores vs fewer. Again, it depends completely on your code. (And it's worth mentioning that less efficient code on more cores could be doing more work that more efficient code on fewer).
And this thread has so many off-base posts..unsurprisingly.
test_account
XP-39C²
Nice thread, cool to read about things like this
Psychotext
Member
I'm interested in your opinion, as all the analysis I've seen says otherwise.Onix said:
On the subject in general. It's probably not worth discussing much further here because as I've already said it's been done to death (at places like arstechnica and beyond3d) and in much greater detail than we're likely to get here.
Plus I'm likely to get labelled as a fanboy despite not owning a PS3 or 360. :lol
Psychotext
Member
Ahh, I see what you're saying. This is of course great for hardcore number crunching for usage in HPC server clusters.gofreak said:
Psychotext said:
That figure is the Theoretical performance of the Xenon. And that is different from the actual peak performance..
But it has been shown for the Celll that it can run very close to Theoretical performance for sustained amounts of time because of its architecture.. I'll find the link when I get home..
@ Psychotext
The Cell with 8 SPEs managed to get pretty close to 100% efficiency, over 200 GFlops (excluding the PPE that wasn't measured).
From a more technical expert Amigan on the Xenon vs Cell (if you are interested in the Cell you probably read his articles, he also runs the new PS3coderz.com):
"The 115.2 figure is the theoretical peak if you include non-arithmetic instructions such as permute. These are not normally included in *any* measure of FLOPs."
"If you want to count non-arithemitic peak figures, the usable Cell components in the PS3 will get 268 Gflops (6 SPEs + PPE) - over twice that of the 360."
The Cell with 8 SPEs managed to get pretty close to 100% efficiency, over 200 GFlops (excluding the PPE that wasn't measured).
From a more technical expert Amigan on the Xenon vs Cell (if you are interested in the Cell you probably read his articles, he also runs the new PS3coderz.com):
"The 115.2 figure is the theoretical peak if you include non-arithmetic instructions such as permute. These are not normally included in *any* measure of FLOPs."
"If you want to count non-arithemitic peak figures, the usable Cell components in the PS3 will get 268 Gflops (6 SPEs + PPE) - over twice that of the 360."
Psychotext
Member
But in most instances outside of HPC you'd have trouble splitting up the processing units in such a way that you'd be keeping these pipelines full. Hence why I said earlier that all this talk of theoretical maximums is largely academic, and the games should speak for themselves.gofreak said:
Anyway, as interesting as this all is, I'm going to go play some games rather than talk about them. :lol
This article has links to a lot more technical information and analysis for those interested in the PS3 and 360 CPUS: http://arstechnica.com/news.ars/post/20050629-5054.html?95741
Plus obviously I can recommend the arstechnica and beyond3d forums.
MikeB said:
yup, i'm not trying to imply that the CELL isn't more powerful than the Xenon, just that the terms of measuring things are just as pointless as the whole PPC vs. X86 discussions from the late 90s. there are many measurements that must come into place when trying to measure a system, and although you can simplify it to single characteristics, it becomes pointless in the end. on paper, the CELL is significantly more powerful, but what is available to the PS3 developers is a bit less than the theoretical limit for reasons specified elsewhere. there are even going to be certain areas where the Xenon will likely run significantly faster than the CELL for certain sections of code, but neither of these cases are ever going to occur in applications.
the best measurement, IMO, is pretty much what you've done already, is to take the end results and to look at whether or not they're even attempting to make use of all the features of the chips. this is made a bit more complex than direct CPU comparisons largely because you have rather complex GPUs that can do a fair amount of the processing from games in them.
Psychotext said:
Within HPC, and within the types of task that have been written up about on Cell etc. there is a wide computational variety. You talk about it like all HPC code is the same, and that anything outside of that is wildly different. Well, I've news for you..
Second, leveraging the memory heirarchy hasn't much to do with task division.
Third, there seems to be this insinuation or impression that some are trying to make that Cell has to be not suitable for everything that it hasn't been publically demoed for - that yeah, it's great for all these many things that have been published about, but for everything else, including games, it's not. I don't think it works like that. Certain other CPUs in certain other games machines have NO transparency on their performance in any tasks, so are we to believe they're not suitable for anything? Necessarily we don't have as much insight into the progress of game developer's with Cell in the same detail as we do from the research world, because game development is commercial and NDA-ed. But hey, at least we have some transparency on Cell, and from what we've seen for ourselves, it's good.
AFAIK games can run on CELL's PPE and the RSX GPU. the SPEs are not absolutely needed to run games. PPE iand RSX are both needed for just about any decent game. well I suppose very simple games could use the PPE alone, but to have decent graphics you need the RSX.
but to really tap into the huge floating point power the PS3 has, for complex animations, geometry setup, lighting, A.I. and all the other stuff you'd then need to use the SPEs.
but to really tap into the huge floating point power the PS3 has, for complex animations, geometry setup, lighting, A.I. and all the other stuff you'd then need to use the SPEs.
Sho_Nuff82
Member
Onix said:
I was speaking in the sense of "wait until devs use the vector units" kind of talk. It's all really getting rather old.
Games are much more convincing than any argument you guys will have about specs in this thread - Heavenly Sword, GT5, Unreal Tournament 3, COD4, Resident Evil 5, and KZ2 are a pretty good indication of where the PS3 is headed graphically.
Kleegamefan
K. LEE GAIDEN
Here is how I always understood how IBM got the performance figures of the PS360 CPUs:
XeCPU:
3 Cores * (4 FPU Flops + 8 VMX Flops) * 3.2 Ghz = 115.2 GFlops
CELL
(1 PPE * (4 FPU Flops + 8 VMX Flops) * 3.2 Ghz) + (7 SPE * 8 Flops * 3.2 Ghz) = 217.6 GFlops
If you don't want to count the 7th OS SPE in CELL then take away ~25Gflops from the final number....
XeCPU:
3 Cores * (4 FPU Flops + 8 VMX Flops) * 3.2 Ghz = 115.2 GFlops
CELL
(1 PPE * (4 FPU Flops + 8 VMX Flops) * 3.2 Ghz) + (7 SPE * 8 Flops * 3.2 Ghz) = 217.6 GFlops
If you don't want to count the 7th OS SPE in CELL then take away ~25Gflops from the final number....
RSTEIN said:
Yeah, they have tons of potential but you're going to reach the end on the 360 a lot sooner than the PS3 for the fact that the PS3 allows you to do many things in many different ways and only experience will teach developers how something should be done in the most proper fashion. It's why you see games like God of War 2 come out on the PS2 and wonder how they manage to pull off visuals on that level.
Kleegamefan
K. LEE GAIDEN
methane47 said:
It is common knowledge CELL architecture was designed to run close to its theoretical max performance.....its one of the reasons Sony payed IBM so much money for it
gofreak also touched on this a little as well...
artredis1980
Banned
Kleegamefan said:It is common knowledge CELL architecture was designed to run close to its theoretical max performance.....its one of the reasons Sony payed IBM so much money for it
gofreak also touched on this a little as well...
run close to theoritical max performance in terms of gamecode of IBM image and data test environments which are 2 things on opposite sides of technology
artredis1980 said:
The performance numbers were brought up because someone showed that Xenon = 115 GFLOPS..
when in reality it is much lower..
While on the other hand Theoretically Cell = 200+ and actual performance is around the same as well..
Theoretical performance and Peak performance were brought up so that we can all compare apples to apples ...
Compare theoretical to theoretical, because without access to hardware there is no way anyone can discuss achieved efficiencies here (and even then, AGAIN, its the mostly the efficiency of your code that you're talking about). You almost could via PS3 Linux and XNA, but managed code on 360 will be at a disadvantage probably in that regard.
painful fart
Member
Don´t forget about the Wii, it also excels in certain areas and has a lot of potential. Let´s say all platforms are equally powerful and we are all happy campers high on cocaine.RSTEIN said:
Core407 said:
I agree with you 100%. Keep in mind that both consoles represent a new shift for console devs (i.e. multi core/multi SPE architecture) and there are a lot of corners yet to be explored within the two consoles. Unfortunately MS has been a bit more progressive than Sony with helping devs with this transition. Sony is starting to step it up in this area (I guess complacency happens when you're the undisputed console king). Additionally, for devs, Sony's current userbase doesn't justify the expenditure required to explore the potential of the Cell.