The New Cell Processor

[marine73]

Found this article discussing the operation of the new Cell processor. The author pulled the patent , read and then "decoded" it for mere mortals such as myself. Pretty neat stuff and boy are Intel and AMD gonna have to pick up the pace soon. Or get left...

http://www.blachford.info/computer/Cells/Cell0.html

 

[MisterChief]

Blah, blah, blah. Hype, hype, hype, Disapointment, disapointment.

 

[CTho9305]

Direct copy of what I said in the slashdot story:

People seem to think this is leaps and bounds above everything else, but they're missing the details. In order to obtain that much performance, you'll need a task which parallelizes well so it can be broken up into chunks for the 8 SPEs. Graphics rendering falls into this set of tasks, but a lot of general applications just don't gain that much from parallel processors. Even when you have a task that does parallelize, writing parallel code is quite a bit harder than writing code for just a single thread of execution.

I've seen a lot of hype about having the Cell in your laptop talk to the Cells in your desktop, microwave, and TiVo, but you have to consider real-world limitations. When you set up a network like that (presumably wireless), you're going to be limited to around 100Mbps. In computer clusters and supercomputers, one of the main limitations of performance is the communcation bandwidth available between processors, and the latency of the network. To build a "home supercomputer", you not only need a task that parallelizes well, but one that doesn't require so much inter-node communication that it's held back by a slow network. You can't work around this problem with hardware magic - if the task you're working on requires lots of communication bandwidth, you're going to be held back.

So how much beyond a modern PC is 250GFLOPS anyway? Not much! A GeForce FX at 500MHz does 200 gigaflops. An AMD Athlon's peak performance is 2.4 GFLOPS at 600 MHz... if we scale this up to 2.2 GHz (high-end Athlon), that's 8.8GFLOPS (note: As we're talking about theoretical performance, nonlinear factors like bus speeds can be ignored). Basically, if the Cell dedicates most of its power to graphics rendering, you'll have computation power in the same range as a fast PC of today. Given that we're not going to see any products based on the Cell for a while, this isn't going to be the end of the world for Intel and nVidia (let alone the fact that Cell isn't x86).

Consoles using the Cell will have the advantage of only having to render for TV resolutions - at most 1080 lines, while PCs will be rendering at up to 1600x1200, but if you look at recent history, you can compare the xbox to a then-good PC with a GeForce3 (which came out at around the same time) - the xbox looked better, but PCs did catch up and surpass it's performance and it didn't take all that long. Consoles have to be very high-end when they're released, because the platform doesn't change for 2-3 years, and they still need to be "good enough" after a couple years, before the next generation is released.

 

[pm]

Great post, Chris.

 

[irwincur]

I've seen a lot of hype about having the Cell in your laptop talk to the Cells in your desktop, microwave, and TiVo, but you have to consider real-world limitations. When you set up a network like that (presumably wireless), you're going to be limited to around 100Mbps. In computer clusters and supercomputers, one of the main limitations of performance is the communcation bandwidth available between processors, and the latency of the network. To build a "home supercomputer", you not only need a task that parallelizes well, but one that doesn't require so much inter-node communication that it's held back by a slow network. You can't work around this problem with hardware magic - if the task you're working on requires lots of communication bandwidth, you're going to be held back.

Bingo. I have argued this point in other forums, some just don't get it. The average Joe does not have a fully networked home, and will not for the next decade or so. Furthermore, to meet its full potential there are a lot of things that will have to happen...

Like, Sony selling the PS3 as a base unit and screwing people into buying a 'cell enhanced' toaster to improve their games. I know that this is there plan, it is simple to see. Sony will put cell infrastructure in everything they make, and sell it as if it makes the PS3 better. Why, because they know that they will sell 70 million PS3 units (at least) and it is a good launching platform that will suck people in.

Why this will not work. People that game on consoles do it because they don't want the upgrade issues that PCs have - and they are simple. If Sony decides to push 'enhancements' they will piss the average consumer off. A PS3 has to work as well as it did on its first day as last, there should be nothing in between that increases the requirements.



Sony needs to be careful. IBM and Toshiba can ride this storm. IBM is already busy enough with everyone else and Toshiba is about to take over the TV world with the release of the long awaited SED. Sony makes over 50% of their revenue on the PS3 - if it fails, they fail, plain and simple. Some of their recent qaurters have shown less than $20 million in profit - unforgivable in the electronics world where the average is 100x more. Trying to tie the PS3 to the rest of your products is STUPID.

We may very well be witnessing the end of Sony.

 

[bfonnes]

Originally posted by: irwincur

We may very well be witnessing the end of Sony.

LOL, the end of SONY??? I don't think you know SONY or its customers very well. There are some people that only buy SONY stuff... sony computer, tv, sound system, speakers, CD player, satellite, mp3 player, car stereo system... Hell, if SONY made a car, they'd buy it. No matter how bad things get for SONY, they will still have their hardcore customers that will keep them in business. And we are talking about the most powerful sub-$500 processor ever made. You must really not like SONY.

BFonnes

 

[bfonnes]

Originally posted by: CTho9305
Direct copy of what I said in the slashdot story:

Basically, if the Cell dedicates most of its power to graphics rendering, you'll have computation power in the same range as a fast PC of today.

where are you getting your information??? I read that the Playstation 3 will still have a graphics processor... One that is manufactured and engineered by NVidia and Sony together that will have performance in comparison to say a 1.5version of the current generation of NVidia video cards. Plus, the main processor does not ever "render" the graphics. The CPU typically does vertex calculations and everything else is done by the video card. The GPU "colors" the polygons, shades, and calculates lighting and applies various effects.

see here

quote
"After checking the general features of modern processors, let?s see the analysis of 3D video performance; you can wonder how does it deal with processors and the answer is that video and CPU are strictly connected. Let?s start with an introduction: in the generation process of a 3D scene (let?s think to a game) many factors are involved, both hardware and software; many think that most work is carried by the video card, and it is true, but the processor too plays an important role as it processes many calculations necessary to define the image. For instance, if we want to display a cube, the calculations for the corner position are carried by the processor and not by the video card; more complex is the scene, more calculations are to be carried and more will be the operations executed by the processor. These are floating point calculations, so the FPU (Floating Poin Unit) has to work hard. Considering the fact that 3D scenes aim at achieving real-life scene and showing many details leads to the creation of video processors more complex and with high performance, there will be an even increasing demand for floating point performance of the CPU to create scene which are even more complex (so with many polygons) and realistic."

So, your main argument is not valid as the CPU is not a "renderer," and would not be slowed down by advanced graphics. Now considering that it has 8 FPUs, it shouldn't have any trouble with its task, and also should not be compared to current processors as there is no comparison.

BFonnes

 

[91TTZ]

Originally posted by: CTho9305
Basically, if the Cell dedicates most of its power to graphics rendering, you'll have computation power in the same range as a fast PC of today. Given that we're not going to see any products based on the Cell for a while, this isn't going to be the end of the world for Intel and nVidia (let alone the fact that Cell isn't x86).

[/quote]

But the thing is that the Cell isn't going to be rendering graphics, it's not a graphics coprocessor it's a general purpose CPU.

In the PS3, an Nvidia GPU will be doing the graphics processing and it will be the bottleneck even if the Cell is infinitely fast.

 

[TomKazansky]

Originally posted by: 91TTZ

Originally posted by: CTho9305
Basically, if the Cell dedicates most of its power to graphics rendering, you'll have computation power in the same range as a fast PC of today. Given that we're not going to see any products based on the Cell for a while, this isn't going to be the end of the world for Intel and nVidia (let alone the fact that Cell isn't x86).

But the thing is that the Cell isn't going to be rendering graphics, it's not a graphics coprocessor it's a general purpose CPU.

In the PS3, an Nvidia GPU will be doing the graphics processing and it will be the bottleneck even if the Cell is infinitely fast.[/quote]

everything just has to catch up =P
they can really use the bandwidth of the upcoming xdr for this cell structure.

 

[CTho9305]

Originally posted by: bfonnes

Originally posted by: CTho9305
Direct copy of what I said in the slashdot story:

Basically, if the Cell dedicates most of its power to graphics rendering, you'll have computation power in the same range as a fast PC of today.

where are you getting your information??? I read that the Playstation 3 will still have a graphics processor... One that is manufactured and engineered by NVidia and Sony together that will have performance in comparison to say a 1.5version of the current generation of NVidia video cards.

I have no information on the PS3 - I made an assumption based on what I'd read about the Cell. I still don't expect Cell-based systems to be leaps and bounds ahead of high-end PCs, and I don't have enough time to read up on the details of the Cell.

FYI, your article is from mid-1999, before the GeForce256 was released. T&L takes more load away from the CPU and puts it on the video card.


Some other random issues I have with the Cell:

I took a quick look at this article, and it looks like you pay a HUGE performance penalty for doubles (pretty much required since floats don't offer enough precision for many things) and you don't get IEEE754-complaint rounding (ok for games, requires analysis to see if it's ok for other tasks - IEEE754 floating point rounding is designed in a way that, IIRC, doesn't cause the amount of rounding error to grow too much as you do more and more operations).

Due to the 18 cycle branch misprediction penalty and the lack of a branch predictor, tremendous effort will have to be devoted to avoiding branches.

:Q For anything other than DSP-like tasks and scientific computations, that's going to be painful.

 

[ribbon13]

Maybe highly optimized software could use it for rendering?

 

[Sunner]

Seems like Sony is going the opposite way compared to EPIC, putting all the work in the hands of the programmer rather than the compiler.

 

[unipidity]

Which might make the PS3 less successful if the release date is pushed back to accomodate slower programmers in the software developers. Likewise, fewer games than previously seems likely.

 

[nyarrgh]

looking at all the hype,

PS3 Cell will be as ahead of the field as the emotion engine was when it came out.

read into that what you will...

 

[irwincur]

LOL, the end of SONY??? I don't think you know SONY or its customers very well. There are some people that only buy SONY stuff... sony computer, tv, sound system, speakers, CD player, satellite, mp3 player, car stereo system... Hell, if SONY made a car, they'd buy it. No matter how bad things get for SONY, they will still have their hardcore customers that will keep them in business. And we are talking about the most powerful sub-$500 processor ever made. You must really not like SONY.

Sony is barely scraping by as is. Three quarters ago their profit was less than $20 million. This is in a world where their competitor (Samsung ) is making $2.5 billion per quarter. MS is making even more. LG is now right behind Samsung. Sony is in trouble.

To make matters worse, they are almost entirely dependent on the whims of the PS and its offspring. The PS and PS2 account for over 50% of their current revenue. A tiny drop in games sales pushes them into the red. So it really does not matter what else they are selling because they are obviously not good at it. Their margins are horrible compared to the competition and their very future is at risk.

As a name Sony is good. As a managed company - they are staring down the barrel of a gun. So, even though there are plenty of idiots that blindly buy Sony goods, it is not doing them a damn bit of good.

 

[ribbon13]

Sony should sell a PS2 SBC that goes into a PCIe x4 slot. That would make them some cash.

Historically, the game systems that had the strongest copy protection faired the worst.

 

[Velk]

Originally posted by: CTho9305

Consoles using the Cell will have the advantage of only having to render for TV resolutions - at most 1080 lines, while PCs will be rendering at up to 1600x1200

[/quote]

1920x1080 (1080i) > 1600 x 1200

That doesn't stop the original article being silly though - reading a patent and then giving real world performance predictions for a product built off it is more or less akin to reading performance specs in tea leaves.

 

[TuxDave]

Originally posted by: Velk

Originally posted by: CTho9305

Consoles using the Cell will have the advantage of only having to render for TV resolutions - at most 1080 lines, while PCs will be rendering at up to 1600x1200

1920x1080 (1080i) > 1600 x 1200

That doesn't stop the original article being silly though - reading a patent and then giving real world performance predictions for a product built off it is more or less akin to reading performance specs in tea leaves.

Just to be a little more accurate.

1920x1080 > 1600x1200 > 1080i

That 'i' is a kicker.

 

[Pulsar]

The average Joe does not have a fully networked home, and will not for the next decade or so.

Re-evaluate your statement. Half the country doesn't even have broadband yet. Most people replace their appliances one every 15 years or so. My fridge lasted 30. It's going to be 30-50 years before the "average joe" has a fully networked home.

 

[MetalStorm]

How do they plan to make the PS3 affordable?
If each Ps3 is going to be using this Cell processer - which by the way uses exotic ram (Rambus XDR) and has a larger die than Opteron, along with what sounds like the newest nVidia GPU, all of this is going to add up.

Not to mention the fact that you will NEED an HD TV for it to be even worth getting a PS3 as the resolution on consoles at the moment is just terrible (my number 1 reason for not having consoles) and HDTVs aren't cheap either...

It looks like this console and the TV's probably going to be setting you back about 3k! Enjoy your gaming guys!

 

[fishmonger12]

Originally posted by: MetalStorm
How do they plan to make the PS3 affordable?
If each Ps3 is going to be using this Cell processer - which by the way uses exotic ram (Rambus XDR) and has a larger die than Opteron, along with what sounds like the newest nVidia GPU, all of this is going to add up.

Not to mention the fact that you will NEED an HD TV for it to be even worth getting a PS3 as the resolution on consoles at the moment is just terrible (my number 1 reason for not having consoles) and HDTVs aren't cheap either...

It looks like this console and the TV's probably going to be setting you back about 3k! Enjoy your gaming guys!

exactly my thoughts.

 

[imported_mitch2891]

It is widely knowen that games companies see their consoles for less than it costs them to build them and make the money back in the games.

 

[ghackmann]

Originally posted by: mitch2891
It is widely knowen that games companies see their consoles for less than it costs them to build them and make the money back in the games.

It's also untrue, except for a few special cases (like the Dreamcast and the Xbox). Nintendo in particular has repeatedly refuted that their consoles are sold at a loss.

Keep in mind that Sony owns the equipment necessary to fabricate most -- if not all -- of their consoles' components in-house. Plus, the price of manufacturing the console is a drop in the bucket compared to the R&D costs.

 

[imported_mitch2891]

Yeah nintendo is a special case but the PS2 was made at a loss to begin (albeit a very small one) as is the PSP and from the sounds of things so will the PS3 be.

If CELL is all it is cracked up to be I imagine that the trio involved look to recoup R&D costs over a longer period.

I also thought IBM were going to be the ones running the FAB's for these chips.

 

[Velk]

Originally posted by: TuxDave

Originally posted by: Velk

Originally posted by: CTho9305

Consoles using the Cell will have the advantage of only having to render for TV resolutions - at most 1080 lines, while PCs will be rendering at up to 1600x1200

1920x1080 (1080i) > 1600 x 1200

That doesn't stop the original article being silly though - reading a patent and then giving real world performance predictions for a product built off it is more or less akin to reading performance specs in tea leaves.

Just to be a little more accurate.

1920x1080 > 1600x1200 > 1080i

That 'i' is a kicker.

Not at all - to be able to display alternating interlaced frames you still need to be able to work in the full resolution in the first place. You'd have a point if it was, say, television broadcasting, but for a games console you still need to do all the work to calculate a 1920x1080 resolution image to be able to render the interlaced half frames for 1080i.

Of course, if you are going solely on the basis of how many pixels are displayed in a given image, then yes, 1080i is a lower resolution than 1600x1200, but we were talking about the work required to produce the image in the first place. 1080p would probably have been a better example, given it works on both levels.