By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Gaming Discussion - 360 CPU and Cell are fairly equal according to Dave Shippy

Badassbab said:
What I don't understand is why if the PS3 is so much more powerful, we haven't seen any evidence of it. Yeah it's got some great looking exclusives but nothing that really trounces say Gears 2 or even multi plat games like Resi 5. And the upcoming Lost Planet 2 looks gorgeous. If the PS3 is so much more powerful, than the 'technically inferior' 360 is really giving it a real good run for it's money. I don't ever remember a console war were the power of the consoles was up for so much (legit) debate. Used to be so clear in the past. SNES>Mega Drive. N64>PSOne. Xbox>PS2.

much more powerful is subjective..and in what context, and in what you have done with the hardware.

if you notice the in general Games that have been released for both HD system's have been arround the same quality in visual's..there is a reason for that.

1) the Xbox360 has been on the market for a year longer. and since the design is along the same line's as current system's are designed, there is more Developer's that have experience with processor design's and developing game's on that type of platform. if you look, even the xbox360 is multi core, and there was still a learning curve for development for it. but that's not the same  as with the PS3.

2) because the Cell like Hybrid Processor's are memory managment processor's there is more development function's that give more ability to program to the bare metal, but also requires more tool's to get the most out of the hardware. and of course, time for developer's to pick up on the new hardware, to gather more experience, and thus it's like" you have some great experience with previous" hardware development but since there are now more function's to relay, more option's, you have to start over from scratch to get experience in development in that enviroment.

Time and money expended now, greater result's later..and in later month's if not year's, but once it's done, you can just like they have been with current hardware , get better result's.

 

 



I AM BOLO

100% lover "nothing else matter's" after that...

ps:

Proud psOne/2/3/p owner.  I survived Aplcalyps3 and all I got was this lousy Signature.

Around the Network

@NJ5

Anyone who uses most animation middleware packages, or physics middleware, etc. can and will use all 6 available SPEs (the 7ths is not usable by games). There's no reason not to. You just are going to have trouble using them all the time, because not all game tasks are covered by middleware, and not everything is embarrassingly parallel.  Its not impossible to use them more, however -- as KZ2 demonstrates.  Why they don't use 6 is beyond me... its possible that the graph is outdated, or that the other 2 SPEs are off doing something uninteresting in the background, but in real-time (like audio, or streaming work), and that they didn't care to show it on the graph.

Like any parallel system, anything that can be done in parallel (like animation processing, a very commonplace example) can be shotgunned on the SPEs, via middleware at the very least.  Therein lies the Cell's advantage over the Xenon.  Which is better, double-barrel shotgun (or up to 5 slow-RoF barrels, not 6, because one of the Xenon processor threads needs to manage the operation), or 6-barrel, high-speed minigun (the PPU thread managing them is different, and I suppose there is one more PPU thread to spare as well)?

Beer says that these operations are frickin' fast, esp. 3D vector math stuff (hence some previous posters' "GPU" comparisons), when done on all 6 SPEs in parallel. Beer is good for you, and never wrong.



 

NJ5 said:

I don't deny that multi-core will become more and more prevalent, but which if Sony's games use all of the Cell's cores? Last I heard KZ2 uses 4 SPEs, and even then quite sparsely (a lot of idle time on those 4). It was on some presentation slide by Guerilla, unfortunately I can't find it right now.

EDIT - Found it:

 

that depend's on what you are wanting do do with Each core :

example Warhawk's developer stated this:

Incognito's  talking about using Cell to render geometry:

4. What did you do on this game that you couldn’t do on another platform?

It’s hard to answer this and not sound like a gratuitous SONY sales pitch
Although I would say it’s the sum-total of all of our natural phenomenon in the game. Our clouds, procedural water, atmospheric scattering, terrain, etc. All of this stuff runs in parallel on all 7 SPUs simultaneously every frame – I’m still not sure if the game community is giving enough credit to just how fast the SPUs really are.

-dylan-jobe

 

 



I AM BOLO

100% lover "nothing else matter's" after that...

ps:

Proud psOne/2/3/p owner.  I survived Aplcalyps3 and all I got was this lousy Signature.

Ascended_Saiyan3 said:
CAL4M1TY said:
Funny thing is, with all this hate going on in these threads, I have yet to see a PS3 game that trounces all 360 games in every aspect of performance, ditto for a 360 game.

To some people "trounces" could mean 20% better.  To others people "trounces" is basically an unreachable sliding bar that constantly changes based on whatever the top performing game is.

The reality of it is that Killzone 2 "trounces" ALL games before it except Crysis.  KZ2 "trounces" Crysis in SOME areas (animations, character models, characters on screen, gun models, particle physics, 7.1 discrete audio, etc) but is close in other areas and gets beat in other areas.  Killzone 2 uses the SPEs to achieve up to a 40% increase in GPU performance (WAY beyond the X360)!

 

Also, Shippy is only speaking of the part HE designed (the PPE of the Cell and the XeCPU cores of Xenon).  That's actually a good benchmark to use.  We know that in most cases of optimized code, the SPE can be around 3x faster than the PPE.

One guy said said something that was incorrect about the BD drive speed in the PS3 vs. DVD drive speed in the X360. Even though the X360 drive is a 12x DVD drive, it maxes out at 8x for DL-DVDs (almost ALL X360 games).  The X360 drive has to build up to that speed over the span of the disc.  The BD drive has a constant read speed across the entire disc.  This ends in the BD drive being just over 1MB/s faster than the DVD drive in the X360.

Here are some links to Cell performance...

http://www.mc.com/uploadedFiles/Cell-Perf-Simple.pdf  (Against the fastest single cores [like Intel Xeon]...1 SPE can perform up to 7 times better than a single core 3.6GHz Intel Xeon)

http://www.research.ibm.com/cell/whitepapers/alias_cloth.pdf  (Cloth Physics on 2.4GHz Cell vs single core 3.6GHz Intel Xeon)

http://www.simbiosys.ca/science/white_papers/eHiTS_on_the_Cell.pdf  (Performance comparison between Intel/AMD dual-core processors versus Cell...also partially explains that realism in games is not limited by the GPU anymore)

http://www.cs.berkeley.edu/~samw/research/papers/ipdps08.pdf  (Cell beats Intel Quad-Core [Clovertown] at DP GFLOPS, which it's terrible at compared to it's SP (just divide the chart number in half for Cell due to Cell blade being used).

http://www.ibm.com/developerworks/power/library/pa-cellperf/  (IBM document on Cell and per SPE performance)

http://www.tomshardware.com/news/ibm-lead-architect-cell-cpu-ps3-gaming,1336.html  (SPEs capable of "running single core scalar programs in their entirety)

http://lzhan.wikispaces.com/Cell+Programming?f=print  (Individual programmer experiments on PS3...shows per SPE output at bottom)

http://arxiv.org/PS_cache/physics/pdf/0611/0611201v2.pdf  (Cell almost 20x (2 orders of magnitude) faster than Opteron processor at Molecule Dynamics)

http://www.sintef.no/upload/IKT/9011/SimOslo/eVITA/2008/hagen2.pdf  (Cell versus Intel Core2 Duo in power consumption and theoretical performance)

http://www.power.org/resources/devcorner/cellcorner/hpcspe.pdf  (Cell versus Intel Quad Core [Clovertown] theoretical GFLOPS...and real world Multigrid Finite Element solver running at an unprecedented 52GFLOPS sustained performance)

http://www.power.org/devcon/07/Session_Downloads/PADC07_Bergmann_Sourcery_VSIPL.pdf  (Cell almost 14x the sustained performance of single core 3.6GHz Intel Xeon at VSIPL++ fast convolution)

http://forum.beyond3d.com/showpost.php?p=1024663&postcount=39  (What DirectX means for your GPU[for the DirectX ? fans])

http://www.cis.udel.edu/~cavazos/cisc879/papers/cellFMwhitepaper.pdf  (Cell [1GB RAM] 11x faster than Intel Quad-Core [Clovertown] w/16GB RAM at SP for financial market applications)

 

That was just for the people interested in learning.  The others will do what they do best (try to discredit these unversities' real world implementations and me).

For those that didn't get a chance to see the post above.

Plus, from one of those links are Dr. Hofstee's words...

"Dr. Hofstee warned, we tended toward a trap into which others have fallen, in which the role of the SPEs appears to be reduced in importance. More than just co-processors, Dr. Hofstee said, the SPEs are fully-capable processing units that are capable not only of running threads spawned off from a main program, but also running 'single-core,' scalar programs in their entirety - not only multithreading, but multitasking."

"Dr. Hofstee wanted to make certain we recognized the Cell as a powerful general-purpose processor. '[Cell] is already fairly general-purpose, even today,' he said, 'but of course, over time, we expect it to go even further.'"

For those that haven't realized that the BD drive in the PS3 is faster than the DVD drive in the X360:

2x Blu-ray Drive (72Mbps(9MB/s))
Single Layer (2x CLV) - Constant Linear Velocity (Same speed across entire disk)
Double Layer - Couldn't find any data but no games have been released on a double layer yet.

Entire Blu-ray Disk is read at 9MB/s.

12x DVD-Rom Drive SL (9.25MB/S-15.85MB/s(AVG ~8x(10.57MB/s) DL (4.36MB/s-10.57MB/s(AVG ~6x(7.93MB/s)
SL(DVD-5) 12x Max (5-12x Full CAV) - Constant Angular Velocity (Speed Varies from edge to edge)
DL(DVD-9) 8x Max (3.3-8x Full CAV) - Constant Angular Velocity (Speed Varies from edge to edge)

SL DVD is 1.57MB/s > SL Blu-ray
DL DVD is 1.07MB/s < SL Blu-ray

Sources:
Hitachi 12x DVD-Rom Faq (Page 2)
http://www.hitachi.us/supportingdocs...ead%20speed%22
What is DVD?
http://www.videohelp.com/dvd
Blu-ray.com Blu-ray FAQ
http://www.blu-ray.com/faq/
Wikipedia - Constant Linear Velocity
http://en.wikipedia.org/wiki/Constant_linear_velocity
Wikipedia - Constant Angular Velocity
http://en.wikipedia.org/wiki/Constant_angular_velocity

Almost all of X360 games are on DL DVDs.  That SHOULD dispell a lot of these arguments with RATIONAL people.  The others will continue their delusions no matter what.



Joeorc,

I dunno man. I think only time will tell which is what you say the PS3 needs to really flourish. I mean I've seen the LP2 trailer running on 360 and that really is a stunning looking game so the 360 will continue to improve it's graphics too. One thing the PS3 definately has over the 360 is physics. That's obvious to me anyway. (LPB + KZ2 are great examples- amazing animations).



Around the Network
Procrastinato said:

@NJ5

Anyone who uses most animation middleware packages, or physics middleware, etc. can and will use all 6 available SPEs (the 7ths is not usable by games). There's no reason not to. You just are going to have trouble using them all the time, because not all game tasks are covered by middleware, and not everything is embarrassingly parallel.  Its not impossible to use them more, however -- as KZ2 demonstrates.  Why they don't use 6 is beyond me... its possible that the graph is outdated, or that the other 2 SPEs are off doing something uninteresting in the background, but in real-time (like audio, or streaming work), and that they didn't care to show it on the graph.

Like any parallel system, anything that can be done in parallel (like animation processing, a very commonplace example) can be shotgunned on the SPEs, via middleware at the very least.  Therein lies the Cell's advantage over the Xenon.  Which is better, double-barrel shotgun (or up to 5 slow-RoF barrels, not 6, because one of the Xenon processor threads needs to manage the operation), or 6-barrel, high-speed minigun (the PPU thread managing them is different, and I suppose there is one more PPU thread to spare as well)?

Beer says that these operations are frickin' fast, esp. 3D vector math stuff (hence some previous posters' "GPU" comparisons), when done on all 6 SPEs in parallel. Beer is good for you, and never wrong.

Actually, Killzone 2 developers (GG) decided to use all 6 SPEs at CERTAIN TIMES.  Killzone 2 MOSTLY uses 2 to 3 SPEs. GG determined that it was better, for stable framerates, to use a MAX of 60% of 6 SPEs than a MAX of 100% of 4 SPEs.  The only times that 60% of the 6 SPEs are used are in 32 player online with lots of explosions and gunfire and near the end of the SP campaign.

 



Badassbab said:
Joeorc,

I dunno man. I think only time will tell which is what you say the PS3 needs to really flourish. I mean I've seen the LP2 trailer running on 360 and that really is a stunning looking game so the 360 will continue to improve it's graphics too. One thing the PS3 definately has over the 360 is physics. That's obvious to me anyway. (LPB + KZ2 are great examples- amazing animations).

it's not just time will tell with just the PS3 it's also with the xbox360 also, the problem in thinking like this "relates" to the previous gen system cycle, it's not. for the firm fact DEVELOPER'S now have to relearn thing's that some may not want to but they will have to sooner or later or, they may be without a Job.. the Big indication is look at the system's spec's of PC's and how the developer's make game's for PC. still most game's only need a Single core processor to run the game.

example:

fallout 3

Minimum System Requirements:

  • Windows XP/Vista
  • 1GB System RAM (XP) / 2GB System RAM (Vista)
  • 2.4 Ghz Intel Pentium 4 or equivalent processor
  • Direct X 9.0c compliant videocard with 256MB RAM (NVIDIA 6800 or better / ATI X850 or better)

Recommended System Requirements:

  • Intel Core 2 Duo processor
  • 2 GB System RAM
  • Direct X 9.0c compliant video card with 512MB RAM (NVIDIA 8800 series, ATI 3800 series)
  • Supported Video Card Chipsets:
  • NVIDIA GeForce 200 series

now Look at Fall out 3 Min and recommended requirement's

the point is was fall out 3 made with

Parallel processing in Mind?

as you can see it support's, it but does it take full advantage of it. See that's the real Issue, another great example would be can fallout take full advantage of the  SLI function .

my guess No. it suppoert the hardware but does not get the best result's it could if the game was designed that way from the start.

not to say it's A Must requirement far from that but for the developer to do so, but it's what the developer may not be able to do because of cost, and not everyone has a SLI configure PC.:D.

but take for instance a PS3 or an Xbox360 the hardware is there it does not change, and if the DEVELOPER want's to take advantage of Parallel processingin the design of their game, the developer can, and most likely will due to the fact that's the way the trend's are going.

PC,PS3,XBOX360 all multi core..hell even the PSP is multi-core machine



I AM BOLO

100% lover "nothing else matter's" after that...

ps:

Proud psOne/2/3/p owner.  I survived Aplcalyps3 and all I got was this lousy Signature.

Procrastinato said:

Yes, even the Cell, the first such processor, kicks the Xenon's ass... in the hands of a badass programmer.  The Xenon rules all in the hands of a newb <-- which describes a LOT of game programmers.

 

Funny you should say that...  The following was stated by John Carmack in a interview he gave yesterday:

"And there is a little bit of least common denominator-ing, where usually the PS3 becomes the sticky point in terms of [being] a little bit slower on the graphics, you’ve got a little bit of less memory to deal with but a little bit more processing power. So you kind of have to balance the things between all of that."

http://multiplayerblog.mtv.com/2009/02/24/john-carmack-talks-possible-wii-development-iphone-doom-4-more/



I would wager that the "60%" number is actually near the max that any game will ever honestly use, with regards to the SPEs, unless it is designed around massively parallel concepts.

Its just not possible to keep that many cores busy at times in a game application. However, in regards to the "Xenon vs Cell" debate, the Cell still has the advantage over the Xenon, when it *does* come to those shotgun-prone parallel tasks, and otherwise, it's pretty much the same. Heck the PPU is practically the same processor as a single core of the Xenon -- designed by the same guy, even, according to this article... although I think the article actually states that he led the PPU team... not the Cell team as a whole. In other words, he wasn't the guy actually responsible for the parallelism concepts of the Cell -- he's is part of the supporting cast. Still, I'm sure he knows his stuff.

Alot of the time, the PPU is basically identical to the Xenon... and then when its parallel-time, the Cell blows the Xenon away, because the Xenon has 2 buddies@3.2 GHz, and the Cell has 6.



 

joeorc said:
Procrastinato said:

^^ The Cell's SPEs can perform many functions similar to GPUs, but they aren't GPUs. Anyone who has ever written a software rendering engine would concur that the SPEs downright phenominal, and supremely flexible, for such an endeavor -- but they aren't going to replace a dedicated GPU when it comes right down it it. They can assist the RSX by relieving it of some work, realistically. "Preprocessing" some graphics tasks, if you will.

That being said, in terms of real-world performance, the Cell is a little more powerful than the Xenon, but the 360's GPU usually typically makes up for the difference, unless the engine is a dedicated PS3 one, because not many games are CPU-bound, they are GPU-bound.

I think the issue most people fail to understand, regarding the Cell, is that the Cell itself is not some magical phenominon that is going to change the landscape of computation -- the idea* behind the Cell is, however. Future iterations of the same concept will outperform "full" multicore architectures, from a heat perspective, from a transitor count/cost perspective, and every other perspective that matters, except a software development perspective.

And that last bit is where architecture's like the Xenon get their praise from.  In the hurried modern world, where time and budget is always in a concern, the Xenon has a fair sized advantage, and that's not to be underestimated.

 

* The idea is that "convenience" features of a processor, like a huge cache, out-of-order execution, and a good branch predictor, are much more expensive, and much less worthwhile, than the extra basic cores you could put on a chip are, with the same transitor count.  More power, less heat, less raw materials.  The added performance comes at a cost to the software developer, who can no longer rely on the aforementioned crutches to allow his/her code to pass for "fast".

while i pretty much agree with what you stated..the Cell processor is indeed a CPU/GPU combo chip.

and yea it is not meant to replace a dedicated GPU for the most part. except now the smaller nm scale CPU/GPU combo chip's do provide the didicated GPU finction in Mobile devices, that would require such low power processor's with the graphic's use ability to replace dedicated GPU's that would otherwise offset the advantages of the CPU/GPU chip's provide like you stated in power /watt and transister count.

but here is DR. Hofstee has stated this here is the PDF

lanl.gov/orgs/hpc/roadrunner/rrinfo/RR%20webPDFs/Cell_Hofstee_Non_Conf.pdf


and how AMD and Intel and Ati are also heading in that same direction.

 

The shift of AMD and Intel is driven by profits, and not necessarily capability.  Don’t get me wrong, as I’m not saying it’s totally impossible to deliver on this capability, but in a lot of respects, this reminds me of M$ attempts to organically put more function into windows…it extremely complicates the solution, and leaves more room for loss of efficiency.   AMD is doing so, because they own ATI, and are looking to deliver a cheaper solution to compete and stay alive, but Intel is shifting in this direction to rebuff the comments from Nvidia about the importance of the CPU in the mix.  Overall, I don’t think its going to work, and I think at the end of the day, CPU/GPU will remain separate…another example is the Sound Card…you notice the difference when you have a dedicated soundcard, vice using the board sound. 

BTW, That pdf had an interesting comment that somewhat outlines the expectation some had for Cell and the PS3...

Innovative Chip is best high-performance embedded processor of 2005

We chose the Cell BE as the best high-performance embedded processor of 2005 because of its innovative design and future potential....Even if the Cell BE accumulates no more design wins, the PlayStation 3 could drive sales to nearly 100 million units over the likely five-year lifespan of the console. That would make the Cell BE one of the most successful microprocessors in history.

 

As far as I know, no other major system uses the Cell/Broadband Engine processor, although some are “planning” to use it, which means the ps3 had better start selling A LOT of console in the next year.  My gut tells me that Intel is going to make a very strong to M$ to get its chips in the next console, which would present some interesting dynamics on cost.

 



"...You can't kill ideas with a sword, and you can't sink belief structures with a broadside. You defeat them by making them change..."

- From By Schism Rent Asunder