PlayStation 3's Cell processor demoed

We're not so sure this was the kind of demo we'd have liked from Sony's, Toshiba's, and IBM's much hyped and lauded Cell processor, but Toshiba ran a live demo of the chip wherein it decoded 48 simultaneous standard-def MPEG-2 video streams, which were then projected onto a 1,920 x 1,080 (aka 1080i HD) display. That's great in sticking to their plan to also carve the microprocessor an embedded multimedia powerhouse niche, but we're gonna remain aporetic until we see just how many polygons this thing's really going to push.

[Via I4U]

[Terrence @ Dec 19th 2005 12:53AM]

This isn't a flame, but when I first glanced at the headline I thought it read, "PlayStation 3�s Cell processor doomed."

[Eran Hawke @ Dec 19th 2005 12:53AM]

That's it?

Wouldn't it be cheaper to buy 48 tiny mpeg decoders and snap them together instead of developing a whole new chip and programming structure?

[Frangible @ Dec 19th 2005 12:53AM]

They used a cell processor, but that's not necessarily the PS3's cell processor. They are making both high-end and low-end versions, and video game consoles always get the low-end to cut costs.

[Victor Perez @ Dec 19th 2005 12:53AM]

Who cares if it can encode a gazillion movies in .2 secs , Show me the damm games.

[FrankH @ Dec 19th 2005 12:53AM]

Aporetic is not a typo. When I read that word, I thought these idiots misspelled apathetic (and didn't even use it correctly), but then I went and looked aporetic up and learned a new word! I am sorry for my initial derogatory thought and thanks for enlarging my vocabulary, Engadget.

[Jon Acheson @ Dec 19th 2005 12:53AM]

Eran,

I think they were using the parallel MPEG decoding as an indicator of multimedia horsepower. Not as a standard task.

Although, I would bet that yes, it is cheaper to use one Cell chip than 48 individual decoder chips. For what it's worth.

[Chris @ Dec 19th 2005 12:53AM]

I'm not sure what "standard-def" equates to as a resolution, but 48 images into 1920x1080 works out as 8 columns x 6 rows of 240x180, which to me doesn't actually sound that impressive. Yes, in terms of parallelism it's good, but in terms of throughput it doesn't seem like much. I'm thinking that the display is the limiting factor in this demo, rather than the cell.

[Pari @ Dec 19th 2005 12:53AM]

Interesting news and would be even more interesting to see how Microsoft spins against CELL capability.

For #2, sticking 48 decoder would be inefficient because of bus bottleneck. Simple rule of thumb, more closer the units are, more efficient they are in shuttling data back and forth. The internal bus runs at Chip speed, but the external bus runs right now at about 800MHZ. This is the reason Intel tries to make bigger and bigger Cache built into the chip. Accessing data in the Hard disk is always slower than accessing it in the Cache. Also mutli core is the future, AMD & Intel have announced dual core chip. Wait for IBM & SUN to announce their multi core chip.

For #3 there is only one CELL, low end and high end difference would be in Frequency ie 3GHZ or 4GHZ. Right now CELL has the maximum theoretical compute power for multimedia data crunching.

I would like to the video of the demo.

[Pari @ Dec 19th 2005 12:53AM]

Interesting news and would be even more interesting to see how Microsoft spins against CELL capability.

For #2, sticking 48 decoder would be inefficient because of bus bottleneck. Simple rule of thumb, more closer the units are, more efficient they are in shuttling data back and forth. The internal bus runs at Chip speed, but the external bus runs right now at about 800MHZ. This is the reason Intel tries to make bigger and bigger Cache built into the chip. Accessing data in the Hard disk is always slower than accessing it in the Cache. Also mutli core is the future, AMD & Intel have announced dual core chip. Wait for IBM & SUN to announce their multi core chip.

For #3 there is only one CELL, low end and high end difference would be in Frequency ie 3GHZ or 4GHZ. Right now CELL has the maximum theoretical compute power for multimedia data crunching.

I would like to the video of the demo.

[Matthew Purdy @ Dec 19th 2005 12:53AM]

Does anybody else see the bottle of Absolute Vodka in the middle of the chip?

[Fredi @ Dec 19th 2005 12:53AM]

Each stream had an original resolution of 704x480 with a medium bitrate. 6 SPU's where doing the decoding, one did the downscale part and one was idle. The PE was doing the HDD load and SPU data management.

Fredi

[Eric @ Dec 19th 2005 12:53AM]

Thanks for the new word! Being a journalist, I'm programmed to hate words of obfuscation, but in this case, I'll willing to cit you some slack. ;-)

And for those wondering, from the Witgenstein dictionary:

aporia - Wonder and amazement before the confusing puzzles and paradoxes of our lives and of the universe. Socrates and the other ancient philosophers tried to evoke the philosophic spirit in young men by awakening their aporia, not by simply providing answers to these puzzles. The aporetic voice is the voice that expresses wonder and perplexity.

[Fredi @ Dec 19th 2005 12:53AM]

To make things a bit clearer, that meens that each SPU (of those six that do the decoding) decodes 8 streams. The one SPU that does the scaling receives nearly 500million pixels per second from those six decoding SPU's and scales them down for the screen.

Fredi

[Oliver @ Dec 19th 2005 12:53AM]

Oh! Dictionary.com summarises it as 'tending to doubt' which is a bit different to what you've got!
Aporetic

[Randall @ Dec 19th 2005 12:53AM]

From a video editing geek:

48 simultanious mpeg-2 streams is very VERY impressive. Typically, you can't even edit one stream of mpeg-2 in realtime, let alone play 48 back.

Just to keep you up to date with video nerdry, SD (Standard definition) is 525 lines. Typically its 720 x 486 (DVSD) or sometimes people say its 720 x 480, but just depending on who you talk to. (I edit in 1080i for our local TV NEWS, so I dont know.)

[OddManOut @ Dec 19th 2005 12:53AM]

Interesting...

Main Entry: aporetic
Part of Speech: adjective
Definition: tending to doubt
Example: The aporetic voice is that which expresses wonder and perplexity.
Etymology: from a Greek word meaning `to be at a loss'

That's what I got from dictionary.com, so from the defenition, I think the mighty 'E' is saying they don't want a representational test, they want to see the PS3's cell in action proper before they form an opinion.

[Justin @ Dec 19th 2005 12:53AM]

Just a quick comment: 1920x1080 is not necessarily 1080i (interlaced); it may be 1080p (progressive) and it I think it would be a big difference as 1080p pushes twice as many pixels at any given time as 1080i does, and would be a bit more impressive. Not a big deal for this story, but it would be a big deal if they announced PS3 would roll with 1080p since 1080p, not 1080i, is the holy grail of HDTV. Therefore I would bet *read HOPE* that it was projected in 1080p not 1080i.

It may be helpful to note that Blueray DVDs will likely use 1080p even if the PS3 does not. The ability to play next generation DVDs is a big difference between the PS3 and XBOX 360, which uses traditional DVDs not BlueRay or HDDVD. This is how I will justify buying both!

[GeGi @ Dec 19th 2005 12:53AM]

Actually, this is incredibly impressive.

Using 6 SPEs to decode forty-eight 704x480 MPEG2 streams at once, while 1 SPE downsizes it 1,920 x 1,080, and another SPE just idling is mighty impressive.

...Especially for a chip that isn't specifically designed for MPEG2 decoding.

And practical. Says in the article that it would be used browsing video content. Sounds useful for a DVR or Blu-ray applications.

Microsoft may be touting the XBox 360 as a revolutionary machine that is leading us into the HD era, but in reality it looks like the PS3 is really the true HD harbinger. Especially since the PS3 actually supports HD format media (Blu-ray) while XB360 doesn't.

[Fredi @ Dec 19th 2005 12:53AM]

I think the most impressive thing is that this demo was done with Toshiba's software platform where threads are automatically assigned to SPEs so programmers can write programs without doing thread scheduling by themselves. So if someone would take the time and write a decoder and scheduler in assembler or actualy only the scheduler alone in assembler would most likely almost double that number if the bandwidth this would generate could be handled, not internaly, there's more then enough bandwidth, but from the HDD.

Fredi

[Anthony @ Dec 19th 2005 12:53AM]

Haha! Good eye, #9.

[p3ngwin @ Dec 19th 2005 12:53AM]

http://news.teamxbox.com/xbox/8119/Toshiba-Gives-First-Public-Demo-of-the-Cell/
here you can read that the streams were all read from a HDD,decoded and then projected onto a 1080i
rez screen divided into 8*6 cells, with each stream resized to the cell on the screen. so it WAS decoding the full rez of each stream and one of the SPE's was doing the resizing for the cell of the screen it was going to. so no 240*180 stuff going on in the chip, just that as you guessed the display was the limiting factor. it would actually have been EASIER for the chip if it DIDN'T have to re-scale the display for final output! the chip is dealing with the full rez of the streams the whole way.

[-neon- @ Dec 19th 2005 12:53AM]

#9 - Actually looks like a bottle of skyy ;) right colour and all.

[Brad @ Dec 19th 2005 12:53AM]

Uh you people make it sound like it's really really impressive which in reality it really isnt'. Just do the math and it'll tell you that 48 DVD resolution streams is only about 48MB/s LOL. A hyperthreading dual core P4 running at 3.6GHz can probably decode 40 simultaneous streams.

[Jahooba @ Dec 19th 2005 12:53AM]

Well, 21, think of it this way. This Cell processor is crunching a lot more numbers than the PS2 can crunch. The graphics on the PS2 (while never as nice as my computer) are beautiful (FFX!). If this thing can pump out graphics >4x prettier, we're in for some amazing gaming experiences.

It's funny how us gamers grew up, now we all know what these technical numbers mean...yet we still can't predict how fun these games will be. Too many gamers are becoming cynical. Anyone agree?

[Karan @ Dec 19th 2005 12:53AM]

I too read this as "doomed" at first then did a double-take.

Brad, you fail to appreciate that the MPEG data is compressed, which means there's a whole bunch of processing going on to decode 48 streams at once, then transform them into a single screen and still have a whole pile of processing power hanging around.

Of course we will really know the power once there's a game to look at - or are you all going to cry foul again and wait for a live playable version claiming it's prerendered? - but there's no way to avoid the fact that that's a lot of processing being thrown around. A Pentium couldn't do that.

[nemi @ Dec 19th 2005 12:53AM]

Is it still correct that the PS3 is supposed to have Qty. *4* of these cell chips in it?

(and this demo was with Qty. *1* cell chip?)

[p3ngwin @ Dec 19th 2005 12:53AM]

hey Jahooba you refering to 21 me ? i was backing the PS3 by showing that it is processing huge amounts and even resizing the streams for the final display cells and STILL have idle power!. i was talking mainly to #7 Chris whop thought maybe that it was "only" processing small res streams of 240*180. i love the PS2 architecture even even more the PS3 design philosophy. when we start to get Field Programmable Gate Array multi core chips then things could start getting VERY interesting! heh heh

[jack @ Dec 19th 2005 12:53AM]

nemi, I read a couple of months back that Sony was having problems with their demo model (probably for E3) while trying to push *6* cells into it. I did read, also, though, that their original plan was 4. But they still have a bunch of time to determine the 'optimal' cost/performace combo. Back when I read it (and hardly knew anything about this chip), I thought the cell could have up to 8 SPE cores and they were trying to put 6 in there. I thought, "Well, that's nice of them," but now I realize that they are actually trying to put 6 cells each with 8 cores.
Of course, this could all be speculation on the part of the reporting site. But even so, just one or two cells would be great, 4 would be amazing, and by putting 6 in there would give the PS3 the ability to play Kasparov in some chess.

[nasim @ Dec 19th 2005 12:53AM]

Cell would run at 4.6 gig but thts not all its power could be raised after some time.if SONY adopts the mass marketing strategy then INTEL,AMD would surely be dead....IT is about the brand reputation that counts,...no one is willing to give up sophisticated japanese products for poorly assembled american counterparts....XBOX 2 is surely on the way for a disaster....The hype created by the cell is simply insurmountable.PS 2 has outsold XBOX by a ratio of 4.5:1....and PS3 would fare the same against XBOX2 but the margin would be 15:1

[Kuyu @ Dec 19th 2005 12:53AM]

That's a lot of video. I'd expect to see this sort of power used to push PS3 game menus to the next level. I doubt the gfx will be much better than X360, but the other factors will own M$ (assuming big-N doesn't spank sony). Draw distance, physics, and total interactibility is the name of Sony's game.

[jason @ Dec 19th 2005 12:53AM]

I can't figure out what this chip is supposed to be for. Is it an embedded processor, designed to do a few simple tasks? Or is it a high performance cpu, designed for high speed floating point computation?

I think that IBM's marketing department is doing a good job of confusing the public, getting them to believe its somehow three simultaneous things: a supercomputing-level floating point number cruncher, a tiny cheap embedded processor, and a multimedia processor.
I think Cell is just one of those things: a multimedia processor.

On ibm's website i read about how the cell processor is about making a single general purpose architecture for embedded processors. But this makes no sense. Embedded processors are supposed to be unique function specific processors. It doesn't make sense to make a 'general purpose' 'function specific' processor.

The use of the terms 'embedded' and 'supercomputer on a chip' are marketing things to build of public excitement. I think its just a multimedia processor, which is powerful like a more general purpose cpu (intel P4's), but with extra vector processing kick for extra multimedia punch.

I'm also annoyed by the demonstration they provide. Why would anyone need one single cpu to produce 48 simultaneous video streams on one screen?! And this is what they show us. I've learned that the only way to tell the value of a processor (in other words, provide a meaningful benchmark) is to run REAL applications on it. I want to see Mac OS X, Halo 2, or some video editing tool running on it. IBM is selectivly chooseing demonstrations to make the processor look good, regardless of whether or not the use is of any real use.

My guess is that the Cell unit is going to be extremly inefficient for realistic use. The 8 SIMD processing units are probably going to largly unused because many programs do not have 8 things to do simultaneously, all of which need vector processing (which performs the same operation on multiple data). Probably, only a few of the SIMD units will ever need to be active at once.

As currently specified, I think the Cell processor is going to be a niche thing; a few wierd areas of the market will be able to get good use out of it. If it can be tweaked and adjusted (perhaps reducing the number of SIMD units), then it could have a broader apeal.

[Jerett Biggs @ Dec 19th 2005 12:53AM]

Hi sorry this really isn't related ,however I was looking up my half brothers name Ryan Block on google search and came across this.My half brother may not know about me...my biological fathers name is Gordon Block from Selkirk Manitoba....I would really like to get in contact with my half brother,and apparently have 2 half sisters.Email me if this is [e-mail]