zippyjuan

PlayStation 3 processor unveiled

The chip has been many years in development
The Cell processor, which will drive Sony's PlayStation 3, will run 10-times faster than current PC chips, its designers have said.
Sony, IBM and Toshiba, who have been working on the Cell processor for three years, unveiled the chip on Monday.

It is being designed for use in graphics workstations, the new PlayStation console, and has been described as a supercomputer on a chip.

The chip will run at speeds of greater than 4 GHz, the firms said.

By comparison, rival chip maker Intel's fastest processor runs at 3.8 GHz.

Details of the chip were released at the International Solid State Circuits Conference in San Francisco.

The new processor is set to ignite a fresh battle between Intel and the Cell consortium over which processor sits at the centre of digital products.

The PlayStation 3 is expected in 2006, while Toshiba plans to incorporate it into high-end televisions next year. CELL SPECS

Faster than 4 GHz
256 billion calculations per second
2.5MB of on-chip memory
Able to shuttle data to and from off-chip memory at speeds up to 100 gigabytes per second,
234 million transistors

IBM has said it will sell a workstation with the chip starting later this year.

Cell is comprised of several computing engines, or cores.

A core based on IBM's Power architecture controls eight "synergistic" processing centres.

In all, they can simultaneously carry out 10 instruction sequences, compared with two for current Intel chips.

Later this year, Intel and Advanced Micro Devices plan to release their own "multicore" chips, which also increase the number of instructions that can be executed at once.

The Cell's specifications suggest the PlayStation 3 will offer a significant boost in graphics capabilities but analysts cautioned that not all the features in a product announcement will find their way into systems.

"Any new technology like this has two components," said Steve Kleynhans, an analyst with Meta Group.

He said: "It has the vision of what it could be because you need the big vision to sell it.

"Then there's the reality of how it's really going to be used, which generally is several levels down the chain from there."

While the PlayStation 3 is likely to be the first mass-market product to use Cell, the chip's designers have said the flexible architecture means that it will be useful for a wide range of applications, from servers to mobile phones.

Initial devices are unlikely to be any smaller than a game console, however because the first version of the Cell will run hot enough to need a cooling fan.

And while marketing speak describes the chip as a supercomputer - it remains significantly slower than the slowest computer on the list of the world's top 500 supercomputers.

IBM said Cell was "OS neutral" and would support multiple operating systems simultaneously but designers would not confirm if Microsoft's Windows was among those tested with the chip.

If Cell is to challenge Intel's range of chips in the marketplace, it will need to find itself inside PCs, which predominantly run using Windows.

__________________
I add new pictures to my photo gallery pretty regularly. You can see them here if you are interested: http://www.pbase.com/jeffryz

hapoo

What in the world is a television going to do with a 4ghz integrated chip?

DarkFury

Not sure if this is gaming news or hardware news...

But the real question is... "HOW MUCH WILL IT COST!!!"

Will a PS3 cost more than a PC?

__________________


DarkFury's Pimptopia - Don't Hate the Playa, Hate the Game!
Home of the Original OG Pimp (accept NO imitations)

Jcranmer

Just think how fast the OSD will come up!

gear02

Let's take bets.

I'm betting the PS3 and Xbox 2 will come out at $399 (non-bundled). Nintendo will come out at $299.

SmokeyDP

Yeah I know. The graphics on these games are all ready pretty top notch. On the computer you can atleast bump up the resolution. I hope the new PS2 will be HDTV compadible.

LegendKiller

Even if they are HD compatable, it is *still* limited to by the TV.


LK

Cantacuzene

It would definitly be nice if the games were future proof to 1080p since those TV are right around the corner.

SmokeyDP

Even if the system could output to HDTV the games would have to be compadible to run at that resolution. The way the console games are designed right now I do not think the textures can support a high res such as that.

verve247

Does anyone know how many cores these chips have? I've heard anything from 5 to 9.

__________________
Vegetarian - Old indian word meaning poor hunter.

Daversinger

soon it will be... IBM vs. Intel vs. AMD lol!!!
just because it's 4ghz+ doesn't mean its performance is better than todays CPUs look at AMD's fx-55 vs. Intel's EE. Intel has a much higher clock

obtw. how hot is this sucker going to be?

shocky123

MKay... lol, it's always been IBM vs. Intel vs. AMD. the only thing is, IBM has dominated the server-side applications of chips (i.e. slower clock = higher efficiency )...
whereas Intel has predominately been:
"Give it a fast clock, the inefficiency will be overpowered by the clock"
and AMD has been somewhere inbetween, though more lately towards desktops than I that would like to see....

And just to give you and idea..... of the world's super-computing power...
~1.127 PFlop/s (Nov 04)
and to give you a better idea.... here are some stats from the top 500 computers in the world.

IBM remains the clear leader in the TOP500 list with 43.2% of systems and 49.3% of installed performance
HP is second with 34.6% of systems and 21% of performance.
No other manufacturer is able to capture more than 7% in any category.

Share of Computing Power(world-wide):
1xBlueGene/L (32 Racks ) = (360TFlop/s)/(1130TFlop/s) = ~32%
2xBlueGene/L (64 Racks*) = (720TFlop/s)/(1490TFlop/s)= ~49%

( the 64 rack system is currently being built for the DoE, when completed, will effectively double the world's computing power )

Just so you know... and I didnt know this prior to just now looking it up.... the floating point performance of the Intel chipsets is laughable..

Efficiency in cycles/flop: (from SETI distributed computing) (peak)

Athlon64 (1mb cache) 3.5cycles/flop
P4 EE (2mb cache) 7.9cycles/flop
AthlonXP(Barton) 4.4cycles/flop
IBM PowerPC III (4mb) 5.1cycles/flop

and on an interesting trend,
note:

Pentium Pro 4.5cycles/flop
Pentium 2 4.9c/f
Pentium 3 5.8c/f
Pentium 4 6.4c/f
Pentium 4 EE 7.9c/f

and......
AMD K5 8.9c/f
AMD K6 8.1c/f
AMD K7-ThndBird 5.9c/f
AMD Athlon 5.3c/f
AMD AthlonXP 4.9c/f
AMD Athlon64 3.5c/f



But let us not forget the $1500+/-(infinity) Itanium2's..... they boast
a 1.3 cycle time for a flop.... WOW! ...thats um, badass.

and when you do the math here.... your 3.8Ghz P4EE that costs like $900
will give ya a peak of 0.481 Gflop/s
and your Athlon64 3200 @2.0Ghz = 0.571Gflop/s (cost me $200)

__________________
"me awaits onslaught of estrogen fury." -CornMonkey

"5. When your friend dates a woman that is absolutely terrible for him, and she's wrecking his life, it's your duty to openly hate her, and point out how evil she is." -ialsohaveadream

gear02

wow...did you do all those calculations in your head? jk

absolutely great post!

I don't understand this part:


I guess I don't understand a cycle. Does this mean AMD is increasingly inefficient while Intel is getting better at it?

shocky123

Cell has 8 Cores.
Here are the official stats:

Clock speed 4GHz+
100 GBytes per second aggregate Memory & I/O speed:
- Dual XDR controller gives 25.6 GBytes per second.
- Dual configurable interfaces give 76.8 GBytes per second.
8 X "SPEs", 128 bit vector engines, 128 registers each.
2 instructions issued per cycle per SPE.
Peak = 256 GigaFlops
Double precision maths operations supported.
256KBytes "Local Store" per SPE.
Internal communication is via 4 X 128 bit rings, up to 96 Bytes per cycle.
PPE can handle 2 threads (update: NOT POWER5 based).
PPE includes VMX.
PPE includes 512 KBytes Cache.
"Dynamic Power Management" technology.
Ten heat sensors
221 square mm in 90nm.
234 million transistors
90nm SOI, Low K, 8 layers of metal & Copper interconnect.


ta ta

~Kyle


LOL no, the exact opposite in fact.
A cycle is one trip around the clock on a cpu. I.E. it corresponds/and is.. the frequency of the processor.
Cycles/sec = frequency
1Ghz= 1billion cycles/sec

cycles per flop (c/f) refers to the length of time the processor needs to be able to calculate a floating point operation. (which is no easy task..)


i'll clarify more later if anybody needs..
and no, I didnt make these numbers up

~Kyle

verve247

Thanks for the spec list schocky123. 10 heat sensors? Is that normal? I wonder how hot this thing runs?

shocky123

No, 10 heat sensors is actually quite uncommon.. The most I've seen (except in weiiird setups..) is two heat sensors. (..on the cpu..)
I would actually assume that it runs fairly cool, or cool in respect to the intels and AMD's... No word officially on exact temperatures, but I would suspect the chips would run 50-60C normally, any higher and they wouldnt be able to cool it sufficiently with air. Also, the fact that the size of the cpu is significantly smaller than normal cpu's implies that there(must) be utilization of new(er) thermal control properties. I have read in many articles about Intel's claim at having a "dual-core Itanium-2" which I believe is largely bull****, as they have included their "90-nm model, boasting 1.72-billion transistors, and reducing total power dicipation(sp?) by 30%"
However, upon brief, but further inspection of the principles behind Intel's claim they did make one _possible_ claim... (I'll explain it quick here, cuz i have a program due that i have barely started... )

Basically, people have done thermal image-processing to try to model the thermal characteristics of a cpu. Many problems have alluded the researches.. like configuring their instruments to run at similar frequencies to their experiments, which proves to _fuzz-up_ your signals and communication with noise. Anyways, they(IBM's researchers...) found something that I'm sure you can understand if you've ever seen a picture of a silicon die. If you havnt, google it... it's a pretty impressive show of todays technology.
Anyways, heat dissipiation happens as a result of current leaking in transistors, which is completely normal in a transistor, it just happens.. it's part of the physics involved....
Anyways, the density of transistors inside the processor is different depending on what the functions it has to perform are. They(IBM Researchers) found that by far the Data Cache(i.e. L1 Cache... etc) area of the die is very dense in terms of transistors and due to the functions of the Data Cache, has an EXTREMELY high power output. To give you an idea... a 2.0ghz P4's output (avg. power density) over about 16cm^2 is about 20W/cm^2, however, the Data Cache area has a power output of ~300W/cm^2.

Okay, so we've established there are HotSpots on the cpu, the heating is not necessarily ( or even remotely ) uniform.

Now, simple physics here... if you put two high temp sources next to eachother the temp in that spot will be much higher in aggregate and in general this is not the best way to deal with heat, and creates higher temp Hot spots than if you seperate the hot spots.. say put them on opposite corners, then your highest temp is reduced.
This is how Intel supposedly lowered their new chip's power consumption. However, moving things around doesnt make something use less juice. They switched from 130nm to 90nm, which reduces the power consumption..though I wont go into that here.


***********************************
Anywho, in the Cell chip here.. the cache will be distributed over 8 processors there, so instead of one big hotspot, there are 8 smaller hotspots spread out across the cpu. This will hopefully help the cpu lower it's temperature.

Anyways, my program still hasnt been touched. So I'll have to hit this up later.

Enjoy peoples,

~Kyle

Dave_7

Yeah, but don't you remember saying that about Madden '94 on the Sega, too?

"It's SO REALISTIC!!!"

Can't wait for the next gen!




Dave.

gear02

Oh i get it now. Intel takes more clock cycles to perform one flop of processing than AMD...sweet