Playstation turns Supercomputer - BBC
This is the homepage for the group orgainising the project.
It's always nice to see thinking "outside the box"
Graham
Playstation Supercomputer
- Thread starter Graham
- Start date
-
- Tags
- playstation supercomputer
Playstation turns Supercomputer - BBC
This is the homepage for the group orgainising the project.
It's always nice to see thinking "outside the box"
Graham
PixyMisa
Persnickety Insect
Easy. It's not out yet.Frostbite said:
Also, the hardware is specialised. The "Cell" processor being developed for the PS3 will do about one teraflop (a thousand gigaflops) compared with 12 (I think) gigaflops for a top-of-the-line P4, but it won't be able to run a word processor or a database 100 times faster. In fact, a word processor might actually run slower on the Cell than on a P4. (Databases are likely to be faster, because the Cell actually includes 4 CPUs on one chip.)
I can give you more details if you're interested
EdipisReks
Custom Title
this is assuming that the Cell chip is actually going to be in the PS3. as of right now, it is supposedly a year behind it's development roadmap.
Skeptical Greg
Agave Wine Connoisseur
Dymanic said:
From the article:
PixyMisa said:Easy. It's not out yet.
Also, the hardware is specialised. The "Cell" processor being developed for the PS3 will do about one teraflop (a thousand gigaflops) compared with 12 (I think) gigaflops for a top-of-the-line P4, but it won't be able to run a word processor or a database 100 times faster. In fact, a word processor might actually run slower on the Cell than on a P4. (Databases are likely to be faster, because the Cell actually includes 4 CPUs on one chip.)
I can give you more details if you're interested
I'd be interested . . .
This may be a dumb question but why does four processors mean it runs a database faster but not a wordprocessor?
Graham
PixyMisa
Persnickety Insect
Well, it depends on the database, I guess. Databases are often used by multiple people at the same time, so there are multiple jobs that can be spread across the different processors. Also, advanced databases like Oracle have features that allow a single operation to be split up into several simpler operations, each of which runs on a different processor.
It's not really practical to do this with word processors.
As for the Cell:
Not a lot of detail has been announced, but between certain press releases and the patent found here we can work certain things out.
It would appear that it will have 4 processors on a single chip, most likely MIPS or PowerPC architecture, and probably 64 bits. Each of these general-purpose processors will be accompanied by 8 128-bit vector units, which can each perform 4 32-bit floating point multiply-and-add operations every clock cycle.
Clock speed is expected to be 4GHz.
This is how the teraflop performance works out: 4 multiply-adds per cycle per vector unit = 8 floating point ops (4 multiplies and 4 adds). Times 8 vector units per processor is 64. Times 4 processors per chip is 256. Times 4GHz is 1.024 TFLOPS. Whooosh!
A 3GHz P4 can, I think, do 12 GFLOPS. (1 TFLOP = 1000 GFLOPS)
The chip will also have a significant amount of built in memory. I seem to recall the figure of 64MB being floated around, but I don't remember the basis for that. (There's a lot of information that can be gleaned from the drawings attached to the patent, but they're tiff files and my browser doesn't want to play right now.)
Another key feature, and the reason why the chip is called the Cell, is that it will have high speed optical links built in to connect it to other Cell chips (or, presumably, I/O devices). The idea is that to build a faster system you just link up more Cells.
Plans are to build the Cell initially on a 65nm process (0.065 micron), which is not likely to be available until at least 2005 and more likely 2006. (Current processors are built on a 130nm process, with Intel leading the move to 90nm starting later this year.)
It's not really practical to do this with word processors.
As for the Cell:
Not a lot of detail has been announced, but between certain press releases and the patent found here we can work certain things out.
It would appear that it will have 4 processors on a single chip, most likely MIPS or PowerPC architecture, and probably 64 bits. Each of these general-purpose processors will be accompanied by 8 128-bit vector units, which can each perform 4 32-bit floating point multiply-and-add operations every clock cycle.
Clock speed is expected to be 4GHz.
This is how the teraflop performance works out: 4 multiply-adds per cycle per vector unit = 8 floating point ops (4 multiplies and 4 adds). Times 8 vector units per processor is 64. Times 4 processors per chip is 256. Times 4GHz is 1.024 TFLOPS. Whooosh!
A 3GHz P4 can, I think, do 12 GFLOPS. (1 TFLOP = 1000 GFLOPS)
The chip will also have a significant amount of built in memory. I seem to recall the figure of 64MB being floated around, but I don't remember the basis for that. (There's a lot of information that can be gleaned from the drawings attached to the patent, but they're tiff files and my browser doesn't want to play right now.)
Another key feature, and the reason why the chip is called the Cell, is that it will have high speed optical links built in to connect it to other Cell chips (or, presumably, I/O devices). The idea is that to build a faster system you just link up more Cells.
Plans are to build the Cell initially on a 65nm process (0.065 micron), which is not likely to be available until at least 2005 and more likely 2006. (Current processors are built on a 130nm process, with Intel leading the move to 90nm starting later this year.)
From what I heard the cell chip won't actually be used in the PS3.
Also the idea of having multiple PS3's linked together sounds good in theory but practically it is not doable. You can't squeeze enough data at a fast enough speed between the 2 processors to run real time graphics...
Another thing to remember the "specs" are still just hype (correct me if i'm wrong) so they need to be treated with a lot of caution.
Don't get me wrong, the cell chip may be the next best thing since sliced bread, but I doubt we'll see it used on mass levels for a long time.
Also the idea of having multiple PS3's linked together sounds good in theory but practically it is not doable. You can't squeeze enough data at a fast enough speed between the 2 processors to run real time graphics...
Another thing to remember the "specs" are still just hype (correct me if i'm wrong) so they need to be treated with a lot of caution.
Don't get me wrong, the cell chip may be the next best thing since sliced bread, but I doubt we'll see it used on mass levels for a long time.
PixyMisa said:
PixyMisa
Persnickety Insect
The PS2, the Xbox and the Gamecube each have a general purpose processor (a MIPS Risc, P3 Celeron and PowerPC respectively) and a graphics processor (Emotion Engine, an Nvidia chip similar to the GeForce 4, and something I don't recall). The general purpose processors tend to be slower than found in general purpose computers (300MHz, 733Mhz, 455MHz respectively) because games consoles don't spend a lot of time doing general purpose stuff. The graphics chips tend to be close to the leading edge at the time they are released, though.Whomp said:
PixyMisa
Persnickety Insect
Yep. Current rumours have the Cell delayed to 2007 and an interim machine, basically an accelerated PS2, filling the gap.ImpyTimpy said:
The idea isn't to have multiple PS3 systems linked together, but to have multiple Cell chips inside a system. Sony demonstrated a graphics workstation that contained 16 Playstation 2 CPUs and modified PS2 graphics chips (they increased the built-in memory from 4MB to 32MB) all working together. The Cell is designed to make this sort of thing more practical.
As for linking individual PS3s together - well, that could be done too, but you'd need a very high speed link, and those are currently expensive. Large SGI systems are designed that way - and they are expensive. I don't know whether the Cell optical links are designed to connect to external systems or not, but it would be cool.
Well, you can hardly call it hype when they've hardly made any announcements and most of the information that is available has come from a patent filing.
2006 is a long way away. By that time desktop computers will be well past 4GHz and the Cell will look a lot less remarkable.
PixyMisa said:
That's the way I see it happening as well, however I think Cell chip won't be used in console gaming, perhaps in other things.
That's why I don't think it'd be feasable.
External systems yes, but I don't think other Cell chips. I just don't see getting enough bandwidth through one or two optical links to justify it.
That's exactly why I'm calling it hype. The specs for PS3 are not anywhere yet people are starting to come up with values. The patent does not cover that much in order to be able to ascertain what the PS3's specs will be. Then again, when I read through the patent I might've missed the exact figures given so feel free to correct me.
Maybe, don't know. Future will tell
Maybe Sony's filed the Cell patent to throw everyone off the real chip they're intending to use?
PixyMisa
Persnickety Insect
Well, Sony are definitely planning on it; the Cell is specifically designed for a future generation of games consoles.ImpyTimpy said:
Well, optical links are already commonly running at speeds of 40Gbit/s, and a single fibre can support multiple links. I don't know what bandwidth is proposed for the Cell links, but it's certainly possible to get useful bandwidth over a single optical link for this sort of application.
The patent covers the architecture, but not the speed. I'm not sure where the 4GHz number came from, but it does line up perfectly with the 1TFLOP figure that's also floating around. It's also a very reasonable figure for a chip due in 2005-06 on a 65nm process.
Call it speculation. Hype is when someone is promoting their own product with inflated claims. Speculation is (hopefully educated) guesswork based on the data available.
Very unlikely, since the Cell is clearly the direction in which things are headed anyway.