Here is a side by side video i found somewhere else: http://xbox360.qj.net/We-F-E-A-R-PS3-Xbox-360-side-by-sider-video/pg/49/aid/89725

@MikeB

Quote:

There's more than 100 GFLOPSs worth of performance difference between the PS3 Cell and Xenon for running their GameOS together with games, the PS3 Cell can do this in 9 threads (although if needed more are possible) and the Xenon does this in 6.

I ask again and I am not joking: Doing what calculations? Using which units? Acting on what data that is aligned in what way? Where is that data located?

Moreover, you are comparing apples with oranges, the Cell does *NOT* have 9 general purpose threads, it has 2, the others are the SPEs. The Xenon has 6 general purpose threads, each of them able to use the VMX128 unit. I'm sure you've heard people rave about that unit before, its predecessor is called the Altivec. You can't and I can't claim that one or the either is universally faster.

Moreover, "if needed more are possible"? How? The Cell has 2 PPE parallel hardware threads and 7 SPEs, how exactly do you plan to add more *hardware* threads?

Quote:

The performance gap is so obvious, I am really surprised this needs to be discussed.

And of course we have to take your word for it, since you cannot seem to provide any links about the specific details. Or do you have the illusion that computing speed is something you can count just like that?

Quote:

Procedural synthesis is nothing new, I find it funny that one of the PS3 strongpoints probably due to Microsoft marketing gets turned against it to demonstrate a higher capacity drive wouldn't be needed instead of demonstrating that the amount of system memory present is more than sufficient for the coming years.

I find it funny that you think that I turn it against the PS3. Does your brain only think in black and white? With you or against you?

Quote:

Procedural synthesis can find interesting uses for games, but will not completely negate other game design methods.

It is not just a design method anymore, it is middleware.

Quote:

I find it funny for Amigans who know so much about the demoscene to finally "rediscover" procedural synthetical programming methods mainly when Microsoft starts to talk about this.

My interest in procedural synthesis started ages ago and had nothing to do with Microsoft or the XBox, it had to do first with Planet Potion and then with Farbrausch.

And for the nth time, it's not just a programming technique anymore, read what people write to you fully instead of trying to find snippets that agree with your point of view for once. Your artists need not be coders to utilise it and in fact you don't even need to implement it yourself, there are middleware engines to do it for you.

Oh and for the XVIIXVIXXIVIXVIetc time, I do not have an XBox, I do not have a PS3 and I am not planning on buying either, my interest in them is *academic*, so stop painting me with that fanboy brush of yours just to discredit my word.

Last edited by AMiGR on 21-Apr-2007 at 02:49 PM.
Last edited by AMiGR on 21-Apr-2007 at 02:32 PM.
Last edited by AMiGR on 21-Apr-2007 at 02:26 PM.

_________________

@MikeB

Quote:

At the end of the day the PS3 is simply a different machine that handles data in a different way than 360. So we take advantage of the machine?s capabilities to run our games as fast and as efficiently as possible. I don?t know how many processors we used or didn?t use." IMO this sounds dishonest, surely they know how many SPEs they've used for their game, but simply don't want to point this out to the readers. My personaly guess would be they used one SPE for this game, nothing to be ashamed of considering their engine wasn't really build with the PS3 in mind.

Or perhaps they wrote multithreaded code and do not rely on loading it to a specific SPE before running it?

_________________

@AMiGR

Thanks for the 360 Elite link. I've heard Wallyworld has slipped up on launches like this and nice to see someone's wiling to void their warranty a week before actual release so we can get a peek inside.

@MikeB

Quote:

There's more than 100 GFLOPSs worth of performance difference between the PS3 Cell and Xenon for running their GameOS together with games, the PS3 Cell can do this in 9 threads (although if needed more are possible) and the Xenon does this in 6. The performance gap is so obvious, I am really surprised this needs to be discussed.

Now we were discussing one aspect procedural analysis. Something the 360 was configured from the start having a 1 step dot product to complete. Something the PS3 takes multiple step dot productS to complete. On this 1 item the need to take multiple passes for the PS3 may well be likely enough to off set that 100GFLOP advantage. Once again this is one aspect not totality for example I doubt anyone will just write code that only does procedural analysis and see which system wins. Though it would be interesting if they would.

As for your 100GFlop don't fall into buying the marketing hype, once again this is a theoretical peak something that never occurs. The PS2 is supposedly twice as powerful as the Xbox and we saw huge improvements on the PS2 over the Xbox? Nope we didn't. Developers play a big role in making the best and most efficent use of these consoles. In a car race when there's turns just because a car can go 200MPH and the other only 150 miles per hour that driver is going to play a significant role in winning, the driver in this case are our loyal and trusted developers.

115GFlops for the 360 and 218 GFlops for the PS3. Now something you and many neglect is that 218 number comes from IBM's Cell usage which as 8 SPEs, you have to take away 1 because the PS3 only uses 7 SPEs. 2 SPE is a bit over 25GFlops so the PS3 is only 193Glops.. Still ahead in the theoretical but only about 75 not 100. Then add to this the OS needs, PS3 loses a SPE and part of the PPE and these losses are greater then the 360 OS because Microsoft amazing made a small tight OS. Sony's claim of the 100Gflop lead is simply a paper theoretical number and in the real world these 2 are much closer.

Quote:

IMO this sounds dishonest, surely they know how many SPEs they've used for their game, but simply don't want to point this out to the readers.

Of course there's only 1 processor the CELL which SPEs are components of. So truly the answer to how many processors did you use is 1.

Quote:

Thanks to Bethesda for doing a quality port of IMO most likely the XBox 360's best overall game to the PS3.

Sorry DOAX2 has big independant jiggly breasts. Also the Jet Skiing is really lame on Oblivion. Therefore DOAX2 must be the best game.

@AMiGR

Quote:

I ask again and I am not joking: Doing what calculations?

Heavy calculations, single precision floating point operations. Physics, Procedural Synthetical geometry, AI, decompression, etc, etc.

Quote:

The Xenon has 6 general purpose threads, each of them able to use the VMX128 unit

Your point? You want to imply the Cell (PPE + SPEs) can't?

@MikeB

Quote:

Heavy calculations, single precision floating point operations. Physics, Procedural Synthetical geometry, AI, decompression, etc, etc.

*Sigh*, you do not get it, do you? "Heavy calculations" and "single precision floating point operations" is an incredibly wide brush.

Quote:

Your point? You want to imply the Cell (PPE + SPEs) can't?

No, the SPEs cannot use the VMX unit, they are vector units of their own. My point is that you do not *know* that the SPEs are gonna be so much faster than the VMX128 units at something, unless if you know exactly where the data being worked on comes from, what instructions are operated on the data and where the data goes. You use that incredibly heavy brush of "heavy calculations" to mask the fact that you do not *know* what calculations we are talking about. Simple instructions? Complex instructions? What on earth do you mean by "heavy calculations"? "Physics, Procedural Synthetical geometry, AI, decompression, etc etc." is also generic and does not say anything about the algorithms used, what instructions act on the data, etc. Or do you want to imply that the Cell is faster at every single instruction there is when acting on all data on the planet streamed from anywhere in memory? If yes... Where on earth is your proof? I do not need proof to state that I do not *know* whether it's gonna be faster or slower, *you* need proof as you are claiming it's faster.

Numbers, please?

_________________

@AMiGR

Quote:

Numbers, please?

If I may let's fix that a bit. MikeB loves to quote the on paper theoretical max for his numbers, something that never can occur. Even the real world maxes IBM showed for the Cell, around 155 w/ 8 SPEs , are contrived to simply obtain and measure the actual maximum. Using that number gives one a warm fuzzy about performance but since a game isn't going to run that exact code for the majority of the time real world actual are obviously below real world max and definitely well below a theoretical max. The claimed PS3 wins by 100Gflops is simply comparing on paper advertised theoretical maxes not in actual use.

MikeB we want actuals not theoretical maxes.

Last edited by BrianK on 21-Apr-2007 at 05:31 PM.

@BrianK

Quote:

MikeB we want actuals not theoretical maxes.

The problem is that performance measures such as "Gflops" and "MIPS" do not show much about performance, they usually show that the processor is fast or slow at doing a very very specific thing. They are not real world calculations.

_________________

@MikeB

360 is a bomb...
Here's an article you'll like. http://biz.yahoo.com/seekingalpha/070418/32642_id.html?.v=1

Of course the author only looks at Microsoft not at Sony who is in a similar boat on the PS3, except worse as their losses are a larger % of profits then Microsoft's ratio. If Microsoft heeded this warning all they would sell is their OS and Office. Certainly they'd be more profitable but I'd venture overall weaker a company needs to have diverse products. Something similar to think about Saturn has only lost money for GM yet it's kept around, why?

@Tomas

Like I stated some time ago here at AmigaWorld I expected to see more differences in favour of this XBox 360 version, gladly this appears not to be the case and both versions are about on par.

If you want to judge the PS3 performance Motorstorm is a good preview as it's designed exclusively for the PS3. BTW, IMO Resistance is by far the better game compared to FEAR, this in terms of gameplay, performance as well as visuals.

@ BrianK

Quote:

MikeB we want actuals not theoretical maxes

According to IBM, the maker of both chips, the Cell is a much more powerful chip compared to the Xenon, don't buy into the Microsoft hype.

Xbox 360 processor Performance = 77 gigaflops.

Some interesting quotes with an Amiga mention from Forbes:

"Cell boasts a staggering fiftyfold advantage in handling graphics-intensive applications that will define the next generation of visual entertainment--blindingly fast and seductively immersive games, virtual-reality romps, wireless downloads, real-time video chat, interactive TV shows with multiple endings and a panoply of new services yet to be dreamed up."

"BM reckons Cell, potent and versatile, can do a lot more than just play games. It sees a role for it in mobile phones, handheld video players, high-definition televisions, car design and more."

"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. "

"He says it will take years to fully exploit Cell's capabilities. "

If you don't believe me at least believe IBM over Microsoft's claims.
http://www.forbes.com/free_forbes/2006/0130/076.html

Last edited by MikeB on 21-Apr-2007 at 07:57 PM.
Last edited by MikeB on 21-Apr-2007 at 07:48 PM.

@MikeB

Numbers in real world applications please, GigaFlops measured in the lab or incredibly tight algorithms like Protein folding and stuff like Seti@Home do not cut it, please find and post some examples where one CPU is by far far far faster than the other. While you're doing that, you may want to consider that *fact* that a chip that is very fast at doing one thing isn't necessarily as fast at doing another and a chip that is slow at one thing isn't necessarily slow at all other things.

Quotes are not scientific measurements, dear Mike.

I do not know which chip is gonna be faster at doing one thing until I see it or someone tries it. You do not know that either, no matter how many quotes you dig out to claim that you do.

Last edited by AMiGR on 21-Apr-2007 at 07:55 PM.

_________________

@MikeB

But all this quotes about the Cell is from people who are in some ways involved with it. It is all propaganda in my eyes to hype up their own product. I dont believe propaganda from microsoft regarding the xbox 360 or any of their other products either.

@Tomas

Quote:

But all this quotes about the Cell is from people who are in some ways involved with it

That's normal, because the people who have the most in-depth knowledge either designed the technology or are working extensively with the chip. IMO it's like the early Amiga vs Atari ST years, although I think this time the clear winner will become evident much quicker.

@MikeB

Quote:

" In the early 1980s the chip in the Amiga home computer far outraced those in the Intel line, but Intel conquered the market anyway. "

Carefull with "Intel line" e.g. factor in Intel's 860 RISC (with MMX style 64bit SIMD) +3D GPU SOC vs MC68040+AGA... The technology from Intel’s 860 RISC developments ultimately arrived in the form of Pentium Pro and MMX.

X86 line was aided by clones which uses a single hardware reference platform. Compared to the disjoined 68K based PC market, X86 PC market was superior in terms of the distribution model.

Also in the PC market, AMD's R600 is not just a GpGPU i.e. it also handles sound(1) and physics processing(2) via it's 320 scalar(3) processors.

1. For HDMI.
2. Ruby 4 demo,
3. Recently, AMD claims to have 320 shaders processors for R600. It's possible to obtain 320 shader processors via MIMD model from 64 Vect5D processors.

Last edited by Hammer on 22-Apr-2007 at 01:26 AM.
Last edited by Hammer on 22-Apr-2007 at 01:24 AM.
Last edited by Hammer on 22-Apr-2007 at 12:47 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@MikeB

Quote:

"IBM reckons Cell, potent and versatile, can do a lot more than just play games. It sees a role for it in mobile phones, handheld video players, high-definition televisions, car design and more."

CELL lite will be competing against AMD's Xenos IP + ARM (e.g. from ST-Micro's Nomadik) or MIPS combos.

Last edited by Hammer on 22-Apr-2007 at 01:03 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@MikeB

Quote:

If you don't believe me at least believe IBM over Microsoft's claims.

The article is clearly wrong. It makes a 230GFlop claim for the Cell. This again is theoretical max on paper. If they talked a bit about actuals it might be more impressive.

Here's [url= http://www-128.ibm.com/developerworks/power/library/pa-cellperf/]IBM tested Cell performance[/url] own page the best I can find is Eight SPUs running Linpack at 3.2GHz achieves 155.5GFLOPS on a 4Kx4K matrix. You'll see the same article has a single SPE at 25GFLOPS. PS3 having only 7 SPEs would therefore perform that same calulation at a rate of 120GFLOPS. Now take out another SPE lost for the OS - 95GFLOPS and then there's obviously a bit more loss because the OS also needs the PPE and you will clearly see that your claimed 100GFLOP number is right on paper but wrong in the real world. Also a game doesn't continually run only 4kx4k matrices.

It's too bad IBM doesn't have a comparable sheet for the Xenon. I think it to be mainly Microsoft's fault. But, again as the Cell is going to be used, according to STI, in more then the PS3 it's obviously more important to sell that solution to others. The Xenon was optimized for gaming, the Cell was too but the Cell was optimized for video streaming and other considerations also. I believe Sony claimed the Cell can stream 24 HD broadcasts at once very impressive! However, in the realm of gaming mostly useless in predicting how the Cell is really 'Playing Beyond (tm)'

@MikeB

Quote:

Procedural Synthetical geometry

Recall PS3’s SpeedTree yields vs X360...

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@BrianK

From the article:

Quote:

Number of SPUs 1 SPEsim (GFLOPS) 25.12 Hardware (GFLOPS) 25.01 Accuracy (%) 99.6% Since operations in each data block are independent from those in other blocks, the matrix multiplication algorithm is easily parallelized to all eight SPUs. Figure 5 shows that the matrix multiplication performance increases almost linearly with the number of SPUs, especially with large matrix sizes. Using eight SPUs, the parallel version of matrix multiplication achieves 201GFLOPS, very close to the theoretical maximum of 204.8GFLOPS.

Actually the early test from 2005 is pretty impressive and doesn't take into account the PPE and VMX unit performance.

Also note the article states this was an initial 2005 implementation of LinPack evaluating just the SPEs, the early tests cannot be used to diss the Cell's potential.

On a game system, each SPE would normally be used for different tasks, so having all the SPEs crunching on the same task may be a poor indicating of real world performance gains.

Last edited by MikeB on 22-Apr-2007 at 07:44 AM.

@Hammer

Quote:

Recall PS3’s SpeedTree yields vs X360...

Please enlighten me again, what are you referring to exactly?