Support NeoGAF

[Reanimated]

People were posting all of this comparison information last week and had PS3 putting out somewhere around 400 GFLOPs.

Was it Blim Blim? I can't remember.

In any event, I'd like to see those comparisons and calculations again if anyone has them.

I just want to see how it stacks up to Microsoft's claim of Xenon putting out better than a TeraFLOP.

 

[Kleegamefan]

If PS3=400Gflop and XENON=>1 Teraflop, there is your comparison right there?

What are you trying to accomplish?

 

[peedi]

If PS3=400Gflop and XENON=>1 Teraflop, there is your comparison right there?

What are you trying to accomplish?

He's trolling, as always.

 

[Aizu_Itsuko]

Nothing beats a classic.

 

[Reanimated]

I'm asking because I need the numbers for another thread on another board.

 

[Kleegamefan]

OK

 

[gofreak]

a) this didn't need its own thread

b) i don't know where that 400Gflop number came from

c) we don't know where MS is pulling that teraflop number from - they say in the same release that games on xenon will be almost indiscernable from real life, and they said the original xbox was capable of a teraop, so you know.. ( :lol ) However they're coming up with that figure, it's for the whole system (not just the CPU), and it's very much a PR figure.

d) a 1-8 cell configuration at 4 Ghz yields almost 300Gflops. The difference between this number of the number MS is giving is that currently it is quite transparent how that figure is arrived at, but MS's figure isn't - until we can see where they got that from, it's best to take it with a grain of salt. Calculated in a similar way to Cell's peak performance, a tri-core Power chip at 3Ghz, as in Xenon, will probably come in at 90Gflops..so you can see the discrepancy between that and the "over a teraflop" figure MS are throwing out there. There's either another processor in Xenon (because I doubt the GPU accounts for over 900Gflops, despite the ease with which one can get fuzzy with GPU performance figures e.g. the notorious "NVidia flops" etc.), or MS are being "creative" in the hopes that people will think it's more powerful than the competition.

On a side note, I think it was a pretty silly thing for MS to do. They were going well before by focussing on software only and letting hardware play second fiddle, but now people's attention is on this. But when we cut to the chase, and Sony shows its hand, the numbers battle won't be one it wins, I don't think (unless there are some pretty major surprises in Xenon).

 

[aaaaa0]

Eleventy-zillion GFLOPS!

It doesn't matter, because whatever Sony says, they're forced to make it bigger than whatever Xenon has.

We don't know how 1 teraflop is derived for Xenon, need to wait for more info.

 

[Kleegamefan]

a) this didn't need its own thread

b) i don't know where that 400Gflop number came from

c) we don't know where MS is pulling that teraflop number from - they say in the same release that games on xenon will be almost indiscernable from real life, and they said the original xbox was capable of a teraop, so you know.. ( ) However they're coming up with that figure, it's for the whole system (not just the CPU), and it's very much a PR figure.

d) a 1-8 cell configuration at 4 Ghz yields almost 300Gflops. The difference between this number of the number MS is giving is that currently it is quite transparent how that figure is arrived at, but MS's figure isn't - until we can see where they got that from, it's best to take it with a grain of salt.

Good post mate....

 

[Reanimated]

1-8 cell configuration at 4 Ghz yields almost 300Gflops.


All I needed, thanks.

One other thing - When MS says teraflop it's BS, but when Sony says it it's true?






o...k...

 

[gofreak]

1-8 cell configuration at 4 Ghz yields almost 300Gflops.


All I needed, thanks.

One other thing - When MS says teraflop it's BS, but when Sony says it it's true?






o...k...

No. But sony never said 1Teraflop. When IBM said ~300gflops for that Cell configuration (which, btw, is not the PS3 chip), they showed where that figure was coming from. We're lacking that vital information with this Xenon 1 TFlop figure for now, so I'm saying, wait for more information, and take it with a grain of salt for now until things are more transparent.

GPUs are notoriously "flexible" in terms of how you can measure their performance, just remember that. If pushed, I'm sure sony and nvidia can get equally creative and come up with some equally ridiculous figure MS seems to be goading Sony right now, which could back fire as far as "having the bigger performance numbers" is concerned (and probably will, imo).

 

[Reanimated]

Of course I take all this PR crap from BOTH companies with a grain of salt. I never said I thought any of it was solid fact. I just wanted the numbers because someone else asked about them and I too wanted to see how the GFLOPs calculations were made.

You Sony fans are just too goddamn sensitive.

I mean I can't believe that someone like Klee has the brass to call ME a troll when he's got a post, the thread title of which, refers to Xenon as a "steaming pile".

 

[Lord Error]

One other thing - When MS says teraflop it's BS, but when Sony says it it's true?

Well, Sony didn't made a press release saying that PS3 will have 1TFlop yet, but you also have to understand that there is no chance either of them will have 1T of an actual programmable Flops. Cell in PS3 will have 200-300GFlops at most, CPU in Xbox 2 will probably have 90-100GFlops. There's no way GPU in either of them will have over 900 of true GFlops.

However, they will of course both claim over 1T getting there by counting fixed functions in the GPU.

 

[border]

You Sony fans are just too goddamn sensitive.

Maybe because there is this joke character on another forum pretending to be you, and he's a huge troll.

 

[gofreak]

Of course I take all this PR crap from BOTH companies with a grain of salt. I never said I thought any of it was solid fact. I just wanted the numbers because someone else asked about them and I too wanted to see how the GFLOPs calculations were made.

You Sony fans are just too goddamn sensitive.

I mean I can't believe that someone like Klee has the brass to call ME a troll when he's got a post, the thread title of which, refers to Xenon as a "steaming pile of shit".

I'm a fan of Sony as much as I'm a fan of MS or Nintendo. I may have greater expectations of PS3 hardware, but that's not because I particularly like Sony or "hate" MS or Nintendo, it's just how I expect things to pan out right now.

If you want to know where the Gflops come from:

For the ISSCC cell chip:

8 flops per cycle for the Power core's VMX unit + 2 flops per cycle for the Power Core's FPU + 8*8 flops per cycle (for each of the 8 SPEs) = 74 flops per cycle for the whole chip, multiply by 4Ghz = 296Gflops.

For the Xenon chip:

(speculative)

3 * 8 flops per cycle for the 3 VMX units + 3*2 flops per cycle for 3 FPUs (one per core) = 30 flops per cycle, multiply by 3Ghz = 90Gflops + ? = >1000Gflops.

That ? is why people are taking the figure with a grain of salt. I'd guess they're counting all the hardwired functionality in the GPU, calculating their "equivalents" in floating point performance etc. But even doing that they'd be stretching it to close that gap. Maybe there's another chip in Xenon, who knows, but for now I'm guessing it's a carefully crafted PR figure that, while technically it can be derived, is of very little relevance.

 

[Amir0x]

Haha, this thread was worth the effort for laughs.

 

[Blimblim]

Was it Blim Blim? I can't remember.

That wasn't me. I sure would not dare do such comparisons util we do not know the official final specs for both systems.

 

[TheQueen'sOwn]

You Sony fans are just too goddamn sensitive.

 

[sonycowboy]

I mean I can't believe that someone like Klee has the brass to call ME a troll when he's got a post, the thread title of which, refers to Xenon as a "steaming pile".

Look. It's a troll in sheep's clothing.

Maybe because there is this joke character on another forum pretending to be you, and he's a huge troll.

It's been well established that they're one and the same. (Of course, I know that you knew that

 

[human5892]

...when he's got a post, the thread title of which, refers to Xenon as a "steaming pile".

:lol

fission mailed

 

[Kleegamefan]

mean I can't believe that someone like Klee has the brass to call ME a troll when he's got a post, the thread title of which, refers to Xenon as a "steaming pile".

I never called you a troll in this or any other thread, and you can quote me on THAT

And since misquotes seem to be the theme of today, here is the title of my thread:

J. Allard Xenon (aka, the "Get yer steaming pile of Teraflops right here") Thread

Yes it is tongue in cheek...no I do not refer to XENON as a steaming pile....


Again, what are you trying to accomplish here?

 

[Reanimated]

I'm a fan of Sony as much as I'm a fan of MS or Nintendo. I may have greater expectations of PS3 hardware, but that's not because I particularly like Sony or "hate" MS or Nintendo, it's just how I expect things to pan out right now.

If you want to know where the Gflops come from:

For the ISSCC cell chip:

8 flops per cycle for the Power core's VMX unit + 2 flops per cycle for the Power Core's FPU + 8*8 flops per cycle (for each of the 8 SPEs) = 74 flops per cycle for the whole chip, multiply by 4Ghz = 296Gflops.

For the Xenon chip:

(speculative)

3 * 8 flops per cycle for the 3 VMX units + 3*2 flops per cycle for 3 FPUs (one per core) = 30 flops per cycle, multiply by 3Ghz = 90Gflops + ? = >1000Gflops.

That ? is why people are taking the figure with a grain of salt. I'd guess they're counting all the hardwired functionality in the GPU, calculating their "equivalents" in floating point performance etc. But even doing that they'd be stretching it to close that gap. Maybe there's another chip in Xenon, who knows, but for now I'm guessing it's a carefully crafted PR figure that, while technically it can be derived, is of very little relevance.

AH, a civil reply. THANK YOU.

At the end of the day I think it's probably smart to realize that everything is carfully crafted PR with little real world relevance. I need not harken back to the EE.

 

[Izzy]

From our favourite 'Other' forum:

Reanimated Today, 12:07 PM Post #31


Veteran


Group: Members
Posts: 1101
Joined: 26-October 04
Member No.: 3203


PS3 = 300GFLOP
Xbox 2 = 1000GFLOP


Xenon has the performance of a fucking multi-Cell server.

Kuturagi Ken truly is suiciding.

Reanimated Today, 07:48 AM Post #5


Veteran


Group: Members
Posts: 1101
Joined: 26-October 04
Member No.: 3203



Let me say that my statement about the EU has a caveat - New Nealand is f*g free. Kiwis rawk.

:lol

 

[TheQueen'sOwn]

Klee... this research was used to produce such enlightening posts as this:

Basically.

Xenon has the performance of a fucking multi-Cell server.

Kuturagi Ken truly is suiciding.

EDIT: Damn you Izzy.

 

[gofreak]

AH, a civil reply. THANK YOU.

At the end of the day I think it's probably smart to realize that everything is carfully crafted PR with little real world relevance. I need not harken back to the EE.

You need not, but the raw figures were in place

Oh, and I feel dirty for even helping you at all with those posts you've made on "that" forum

*adds Reanimated to ignore list*

 

[Kleegamefan]

words escape me....

 

[Doom_Bringer]

*adds Reanimated to ignore list*

should have done that long time ago mate!

 

[God's Hand]

Hopefully a mod does something about this idiot.

 

[Izzy]

Reanimated, you just got lit up on another forum! (OA in-joke) :lol

 

[human5892]

Basically.

Xenon has the performance of a fucking multi-Cell server.

Kuturagi Ken truly is suiciding.

Oh, dear.

 

[Reanimated]

I only know Sonycowboy from IGN. I needed the numbers for a big thread I have there. All this "joke character" stuff is just a poor defense/trolling mechanism.

 

[Lazy8s]

Total system FLOPs ratings would be inaccurate without including, as Microsoft apparently is, fixed-function work rates. Total programmable FLOPs ratings are a different measurement, and they're as incomplete a standard for processing comparisons as any other isolated spec.

 

[gofreak]

Total system FLOPs ratings would be inaccurate without including, as Microsoft apparently is, fixed-function work rates. Total programmable FLOPs ratings are a different measurement, and they're as incomplete a standard for processing comparisons as any other isolated spec.

True, but the methods used to count that kind of functionality can be very vague and "fudgy". And usually not at all transparent to people outside the companies, which is why it annoys some people

 

[Lord Error]

Total system FLOPs ratings would be inaccurate without including, as Microsoft apparently is, fixed-function work rates. Total programmable FLOPs ratings are a different measurement, and they're as incomplete a standard for processing comparisons as any other isolated spec.

Yeah, but without some kind of breakdown, it's pretty useless. You just know that Sony is now going to put out some even crazier number, which again, would mean little, without knowing how much of the total goes to programmable flops.

 

[human5892]

A good comparison to this is the millions of polgyons per second game that both Sony and Microsoft were playing around this time last gen (IIRC, Nintendo was the only one that provided game environment numbers...and conservative ones, at that). No "real world" numbers, just theoretical figures that didn't mean much in terms of actual gameplay.

That's what the TERAFLOP!!!1 number is to me, and Sony's inevitable answer to that as well.

 

[Aizu_Itsuko]

Not even Ackbar could save this thread.


A few examples for future reference, all created by reanimated:


Sony finally admits that Cell is a piece of shit, PS3 version to be EVEN WORSE

New Evidence That Ps3 Will Be Another Sony Shitbox, using PROTOTYPE Cell processor

Anyone Else Rooting For Psp And Sony's Bankruptcy?, PSP selling at a loss helps the cause

Blu Ray Fucked Before It Even Gets Out Of The Gate, HVD to store TERABYTE

Ivan From Ign-ps2 Speaks Out On Ps3, Says it's a piece of shit. Xenon better

Prime Outperforms Echoes. Nintendo A Sinking Ship, Damage control teams to duty stations

 

[Lazy8s]

Even a full disclosure of FLOPs origins aren't useful because performance is a function of the balance between speed and manageability. Chip area spent on computational logic is space not available for memory and control logic.

 

[gofreak]

Even a full disclosure of FLOPs origins aren't useful because performance is a function of the balance between speed and manageability. Chip area spent on computational logic is space not available for memory and control logic.

Of course, no one's saying otherwise. It's just a lot of people have an academic lust for certain metrics like floating point performance, no matter how theoretical it might be. It just annoys when a figure is thrown out there with no context and no elaboration. MS know exactly what they're doing, though, it's a very calculated PR swipe. They had just better be prepared for the inevitable reaction.

 

[MegaByte]

A few years ago, Kutaragi said that PS3 could run at 1TFlop. Later (2003) they announced each core (1 CPU + 8 APU) would run at about 256GFlop, prompting several news sites to speculate that there would be four cores. This was all long before NVidia was ever mentioned, but if what Kutaragi said was true (hah) then the 300Gflop number being thrown around right now is far too low.

 

[Lazy8s]

Those wanting a floating-point speed rating regardless of meaningfulness to actual performance got exactly that with Microsoft's claim.

 

[xexex]

6.2 TFLOPs. that's over 6000 GFLOPs, and therefore, a thousand times more power-performance than PS2. :lol

(quoting Sony's PR machine circa 1999-2002)

ha ha ha....







ok really now...

minimim: 160 GFLOPs
128 GFLOPs ( 4 SPEs) plus 32 GFLOPs (1 VMX in the PPE)

maximum: 288 GFLOPs
256 GFLOPs ( 8 SPEs) plus the 32 GFLOPs from the VMX

these are both assuming 4 GHz clockspeed, unless i am mistaken.

also, not counting the GPU since there are NO publicly known, reported or even rumored GFLOP estimations for the Nvidia GPU. other than that Nvidia said it would be around 50x more powerful than PS2 graphics chip which isnt measured in GFLOPs.

 

[xexex]

A few years ago, Kutaragi said that PS3 could run at 1TFlop. Later (2003) they announced each core (1 CPU + 8 APU) would run at about 256GFlop, prompting several news sites to speculate that there would be four cores. This was all long before NVidia was ever mentioned, but if what Kutaragi said was true (hah) then the 300Gflop number being thrown around right now is far too low.

I don't remember any comments from Kutaragi that actually said PS3 would provide 1TFLOP.

but as far back as autumn 1999, Kutaragi and SCEI said that PS3 would be 1000 times more powerful than PS2. if measured in GFLOPs, that would require significantly more than 1TFLOP, it would need 6.2 TFLOPs since PS2 CPU was rated at 6.2 GFLOP.


ok, im officially tired of FLOP talk now
:lol

 

[bigfurb]

People, please don't feed the troll. Reanimated, or leonk1 as he's known on the IGN boards, is merely taking the info you give him, and twisting it around so that the people even more clueless about this stuff than he is can believe that xenon>>>>>>>>>>>>>ps3. He is not worth responding to.

Some choice reanimated/leonk1 quotes from the thread he's doing "research" for

loooool, and? All the Sony fanboys were saying that PS3 - at 400 GFLOPS would be blowing Xenon away. Now when you find out Xenon is doing TFLOPs the story changes? Stop trolling here, plz.

Behold the power of Xenon's TFLOP!

PS3 won't match the TFLOP!

Good thing Xenon has 2 extra cores to help produce that TFLOP.

Personally, I don't put any stock in these new Vector Units (now rebranded as SPEs).

Don't sweat it man. I'd be upset too if I had spent 2 years worshipping Cell and then found out that it's rebranded vector units got trounced by Xenon's 3Ghz tri-core TFLOP power

 

[Mooreberg]

This thread has been a terrible FLOP.

 

[fse]

So who has more FLOPs? Ps3 or MS?

 

[Vince]

The only reference of 1TFlop comes from an interview with Ken Kutaragi from the 30th Anniversary Interview he did for Nikkei in 2001. He talked in abstract terms of the future (perhaps why he's a visionary?!?) of digital content and media and his vision of the topology to support it, it's here that it was said:

"One CELL has a capacity to have 1TFLOPS performance"

Later, in 2003 the nascent SCE Suzuoki Patents on Cell, filed in 2001, were published. In them they had an architecture capable of 1TFlop using 4 PE's and 32APUs fabricated in 65nm. Subsequent to this initial patent from 2001 that's a "landgrab," STI altered the design of Cell - as was reflected in the STI patents on Cell from 2003 - likely based off internal preformance and effeciency simulation. They introduced the concept of the SPU, which differs from an APU in the complex built-up around it which is much more complex. Namely they distributed the work of the formerly centralized DMAC to each SPU, doubled LS size, the interconnections grew, as did buffer and control logic. Then, likely due to the competition (or lack there of), they scaled back their design and lithography goals and I believe their timetable for entry was skewed.

So, what's left is still a single processor which has more amorphous computational power than the competition's entire system and, while lesser in sheer computational power, is vastly more effecient and capable than the initial design. It's also a processor which will likely be much more profitable and profitable sooner as when they hit mass production with 65nm out of Nagasaki F2, they'll have an IC that's just slightly larger than then the 180/(relaxed)150nm Emotion Engine from 2002.

The IC from the Suzuoki patent that hits 1TFlop would be highly feasible based on what we see in the ISSCC Cell; all preformance goals on the macroblocks were hit at 90nm, area is inline with projection, but it's cost (both production and as measured in externalized costs of developing for), timeframe and lack of necessity seems to have doomed it.

 

[xexex]

The only reference of 1TFlop comes from an interview with Ken Kutaragi from the 30th Anniversary Interview he did for Nikkei in 2001. He talked in abstract terms of the future (perhaps why he's a visionary?!?) of digital content and media and his vision of the topology to support it, it's here that it was said:

"One CELL has a capacity to have 1TFLOPS performance"

Later, in 2003 the nascent SCE Suzuoki Patents on Cell, filed in 2001, were published. In them they had an architecture capable of 1TFlop using 4 PE's and 32APUs fabricated in 65nm. Subsequent to this initial patent from 2001 that's a "landgrab," STI altered the design of Cell - as was reflected in the STI patents on Cell from 2003 - likely based off internal preformance and effeciency simulation. They introduced the concept of the SPU, which differs from an APU in the complex built-up around it which is much more complex. Namely they distributed the work of the formerly centralized DMAC to each SPU, doubled LS size, the interconnections grew, as did buffer and control logic. Then, likely due to the competition (or lack there of), they scaled back their design and lithography goals and I believe their timetable for entry was skewed.

So, what's left is still a single processor which has more amorphous computational power than the competition's entire system and, while lesser in sheer computational power, is vastly more effecient and capable than the initial design. It's also a processor which will likely be much more profitable and profitable sooner as when they hit mass production with 65nm out of Nagasaki F2, they'll have an IC that's just slightly larger than then the 180/(relaxed)150nm Emotion Engine from 2002.

The IC from the Suzuoki patent that hits 1TFlop would be highly feasible based on what we see in the ISSCC Cell; all preformance goals on the macroblocks were hit at 90nm, area is inline with projection, but it's cost (both production and as measured in externalized costs of developing for), timeframe and lack of necessity seems to have doomed it.

absolutely fascinating. never read this much detail on the differences between the older version of Cell and the current one.