By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Sony Discussion - The true PS3 super-computer

Loud_Hot_White_Box on 09 June 2008
ion-storm said:
I think it's a bit misleading. The cell wasn't just designed for the PS3.

It was designed first and foremost for PS3, at the request of Sony.  It would not have been created when it was if Sony didn't approach IBM for a processor for PS3.

If you mean that alll 3 companies (Sony, IBM, and Toshiba) knew from the beginning that it would have tons of other uses as well, then that is correct. 



Around the Network
Profcrab on 09 June 2008
alpha_dk said:
Profcrab said:
ssj12 said:
Profcrab said:
SpartanFX said:
crumas2 said:
The Cell wasn't designed for PS3, so the subject line is misleading. That's like saying a supercomputer was built from components designed for the previous generation of Macs because it used PowerPC processors.

Then again, there has been at least one super computer built from hundreds of rack-mount Macs, so let us know when the same is done with a giant array of PS3s.

it is done with 8 PS3s in a university ,,here is the PIC:

 

 

http://blogs.zdnet.com/storage/?p=220


I've always said that IBM came out much better on the Cell deal than Sony did. IBM got a great server processor out of it and Sony got a great server processor out of it . . . that can, in a far less effective role, run games. One of these companies makes servers and the other one makes game systems. When you compare relative cost benefit between two types of products, servers and game consoles, it's pretty funny that the PS3 is a much more cost effective server than a game system. As a game system it is just competitive with the 360. As a server it can, seemingly, be very effective.

I give this thread a 9.1.


You do realize that server processors are better then desktop processors right? They can be way more effective for gaming then a normal processor.


Not true, server applications are very different from most regular applications. Sometimes server and PC tech can overlap but some aspects are more benificial to servers than games. Multithreading is one of them. Gaming apps also don't usually require the bandwidth that server apps do. The Cell is an excellent server processor. I'm still waiting to see how the PS3's very expensive investment into the Cell translates to significantly better graphics than the 360. Right now I see a very small improvement. So, lets just go from the evidence we have. The PS3 Cell is performing about at the level of a more conventional processor (the Tri-core Xenon) for games but seems to do disproportionately better in server applications. So, who got what they wanted? Sony or IBM?

I give that post a 9.0.


I agree with what you are saying, but keep in mind that the way personal computing has trended in the past is that the server technology ('big iron' as some call it) tends to be a precursor for what is going to happen in personal computers.

Decades ago, there was the mainframe and dumb terminals. Then, the technology got small enough that everyone could have their own mini-mainframe. Thus, personal computers were born. Mainframes continued to get faster and more streamlined, making new advancements to speed up execution (hyperthreading, multiple processors, larger caches, branch prediction, etc). This eventually made its way into personal computers as the price went down.

Today we are already seeing the benefits of the move to multiple cores/processors. We are also at an interesting junction in the split of server technology. On the one hand, you have Intel moving with more and more general purpose cores. On the other hand, you have AMD buying ATI and including some of their tech into AMD's processors as a Specialized Processing Unit, IBM and Sony developing the Cell as another format of General CPU + multiple specialized processing Units.

It's an interesting time to be following processor tech, in that no-one is quite sure yet which will be the better solution, a small number of general cores with specialized cores that can be used as needed, or a larger number of general cores. They each have their own strengths and weaknesses, and each could easily be viewed as a 'waste' of die space by proponents of the other view (why have a separate core for XXX if task XXX will never be needed? Why have a whole 8 general cores when you will only ever need 4? etc)

To say that the cell is not useful for normal (read: games) programming is only true because it is still remarkably new tech. As compilers get better at SIMD-izing code, and new programming toolkits are made to make SIMD-izing easier (IBM has a really cool one under development (octopus if you are interested in looking it up), and I assume other companies do as well).

These kind of tools will only improve with time, whereas we are reaching the point of diminishing returns with standard CPU compilers; it has yet to be shown how far we can go with specialized processors. I agree that the Cell is not as good for use in games yet; Sony definitely jumped the gun on when it would be ready for mass acceptance. Next generation the compiler technology for the Cell and processors like it will likely be so much better that it could be used with very little extra work; now it is just too early in the Cell's lifecycle.


As it stands today, most server applications are built to take advantage of multi-processors/cores.  This is nothing new there.  Multi-threaded gaming in general is much newer.  IBM was in a position to put the Cell right to work.  They new this.  What Sony did with it wasn't their problem.  So far as gaming applications are concerned, it remains to be seen if the Cell can deliver.  As it stands today, the Cell is clearly not the Swiss Army Processor.  As I am fond of saying whenever people start to put it on too high of a pedistal, the proof is in the pudding.  When it lives up to the potential that many Sony fanboys think it has, then I'll believe it.  Until that point, they are just parroting Sony's marketing.

As far a technology goes, it is interesting to see Processor tech fragment in that way.  Since engineers have had to find another way to get performance than just increasing clock speed, processor tech has gone in different directions.  Some solutions, however, may not be great for all applications.  In all likelyhood, there will not be one better technology that serves everyone.

I give that post a 9.8. 



Thank god for the disable signatures option.

alpha_dk on 09 June 2008
Profcrab said:
alpha_dk said:
Profcrab said:
ssj12 said:
Profcrab said:
SpartanFX said:
crumas2 said:
*snip, getting too long*




Not true, server applications are very different from most regular applications. Sometimes server and PC tech can overlap but some aspects are more benificial to servers than games. Multithreading is one of them. Gaming apps also don't usually require the bandwidth that server apps do. The Cell is an excellent server processor. I'm still waiting to see how the PS3's very expensive investment into the Cell translates to significantly better graphics than the 360. Right now I see a very small improvement. So, lets just go from the evidence we have. The PS3 Cell is performing about at the level of a more conventional processor (the Tri-core Xenon) for games but seems to do disproportionately better in server applications. So, who got what they wanted? Sony or IBM?

I give that post a 9.0.


I agree with what you are saying, but keep in mind that the way personal computing has trended in the past is that the server technology ('big iron' as some call it) tends to be a precursor for what is going to happen in personal computers.

Decades ago, there was the mainframe and dumb terminals. Then, the technology got small enough that everyone could have their own mini-mainframe. Thus, personal computers were born. Mainframes continued to get faster and more streamlined, making new advancements to speed up execution (hyperthreading, multiple processors, larger caches, branch prediction, etc). This eventually made its way into personal computers as the price went down.

Today we are already seeing the benefits of the move to multiple cores/processors. We are also at an interesting junction in the split of server technology. On the one hand, you have Intel moving with more and more general purpose cores. On the other hand, you have AMD buying ATI and including some of their tech into AMD's processors as a Specialized Processing Unit, IBM and Sony developing the Cell as another format of General CPU + multiple specialized processing Units.

It's an interesting time to be following processor tech, in that no-one is quite sure yet which will be the better solution, a small number of general cores with specialized cores that can be used as needed, or a larger number of general cores. They each have their own strengths and weaknesses, and each could easily be viewed as a 'waste' of die space by proponents of the other view (why have a separate core for XXX if task XXX will never be needed? Why have a whole 8 general cores when you will only ever need 4? etc)

To say that the cell is not useful for normal (read: games) programming is only true because it is still remarkably new tech. As compilers get better at SIMD-izing code, and new programming toolkits are made to make SIMD-izing easier (IBM has a really cool one under development (octopus if you are interested in looking it up), and I assume other companies do as well).

These kind of tools will only improve with time, whereas we are reaching the point of diminishing returns with standard CPU compilers; it has yet to be shown how far we can go with specialized processors. I agree that the Cell is not as good for use in games yet; Sony definitely jumped the gun on when it would be ready for mass acceptance. Next generation the compiler technology for the Cell and processors like it will likely be so much better that it could be used with very little extra work; now it is just too early in the Cell's lifecycle.


As it stands today, most server applications are built to take advantage of multi-processors/cores.  This is nothing new there.  Multi-threaded gaming in general is much newer.  IBM was in a position to put the Cell right to work.  They new this.  What Sony did with it wasn't their problem.  So far as gaming applications are concerned, it remains to be seen if the Cell can deliver.  As it stands today, the Cell is clearly not the Swiss Army Processor.  As I am fond of saying whenever people start to put it on too high of a pedistal, the proof is in the pudding.  When it lives up to the potential that many Sony fanboys think it has, then I'll believe it.  Until that point, they are just parroting Sony's marketing.

As far a technology goes, it is interesting to see Processor tech fragment in that way.  Since engineers have had to find another way to get performance than just increasing clock speed, processor tech has gone in different directions.  Some solutions, however, may not be great for all applications.  In all likelyhood, there will not be one better technology that serves everyone.

I give that post a 9.8. 


I agree I was commenting more on the now-bolded part, as most 'regular' programs are built upon a layer that is in the process of becoming closer to server tech.  Servers are a very good guide to where General Computing is going, I just think Sony entered this next phase a bit too soon.  Not only were the programmers not ready for the new paradigm, but the other supporting technology that usually has a greater chance to mature before reaching them is also still unmatured because the tech is so different.  I still feel that in 10 years, the Cell *might* be a good choice for development.

 Off topic: What is the deal with rating posts all of a sudden?  I'm not on much for a week cause I had to move, and all of a sudden there's new memes that I don't get.  I give that meme a 0.5.



Please, PLEASE do NOT feed the trolls.
fksumot tag: "Sheik had to become a man to be useful. Or less useful. Might depend if you're bi."

--Predictions--
1) WiiFit will outsell the pokemans.
  Current Status: 2009.01.10 70k till PKMN Yellow (Passed: Emerald, Crystal, FR/LG)

Slimebeast on 09 June 2008

Is this where I should say "merge 1000 units of the Cell and it can actually accomplish something"?



MikeB on 09 June 2008

@ ProfCrab

I've always said that IBM came out much better on the Cell deal than Sony did. IBM got a great server processor out of it and Sony got a great server processor out of it . . . that can, in a far less effective role, run games.


To quote a friend (of PS3coderz and his many Cell articles fame):

"As for performance there's plenty of research papers out there now showing Cell is not only fast on the areas it's designed for but also in many areas it's not designed for. On raw performance anything less than 8x faster than an x86 core is *low*."

The Cell is a most excellent chip for gaming and multi-media. The only real issue is porting legacy game engines. Building an engine from scratch results in no issues to harvest its potential (of course some initial R&D may be required to get a good idea of how to best approach game engine design).

The SPU code should ideally stick to half- or single precision format, be designed to run parallel as possible (which requires some planning talent) and be cut down into small enough pieces if needed to fit the LS (for example fetch new data and process other data simultaneously streaming data around the super high bandwidth ring, a multi-buffer design).

Main issue: The Cell is very different.
Main advantage: The Cell is extremely powerful.

Comparing the 360 with PS3 CPU, the Cell's PPE is roughly the equivalent of 1 360 core, however the PS3 Cell has 7 SPUs which are considerably more powerful than the PPE can be for game, plain number crunching and multi-media code.



Naughty Dog: "At Naughty Dog, we're pretty sure we should be able to see leaps between games on the PS3 that are even bigger than they were on the PS2."

PS3 vs 360 sales

Around the Network
Million on 09 June 2008
Profcrab said:
alpha_dk said:
Profcrab said:
ssj12 said:
Profcrab said:
SpartanFX said:
crumas2 said:
The Cell wasn't designed for PS3, so the subject line is misleading. That's like saying a supercomputer was built from components designed for the previous generation of Macs because it used PowerPC processors.

Then again, there has been at least one super computer built from hundreds of rack-mount Macs, so let us know when the same is done with a giant array of PS3s.

it is done with 8 PS3s in a university ,,here is the PIC:

 

 

http://blogs.zdnet.com/storage/?p=220


I've always said that IBM came out much better on the Cell deal than Sony did. IBM got a great server processor out of it and Sony got a great server processor out of it . . . that can, in a far less effective role, run games. One of these companies makes servers and the other one makes game systems. When you compare relative cost benefit between two types of products, servers and game consoles, it's pretty funny that the PS3 is a much more cost effective server than a game system. As a game system it is just competitive with the 360. As a server it can, seemingly, be very effective.

I give this thread a 9.1.


You do realize that server processors are better then desktop processors right? They can be way more effective for gaming then a normal processor.


Not true, server applications are very different from most regular applications. Sometimes server and PC tech can overlap but some aspects are more benificial to servers than games. Multithreading is one of them. Gaming apps also don't usually require the bandwidth that server apps do. The Cell is an excellent server processor. I'm still waiting to see how the PS3's very expensive investment into the Cell translates to significantly better graphics than the 360. Right now I see a very small improvement. So, lets just go from the evidence we have. The PS3 Cell is performing about at the level of a more conventional processor (the Tri-core Xenon) for games but seems to do disproportionately better in server applications. So, who got what they wanted? Sony or IBM?

I give that post a 9.0.


I agree with what you are saying, but keep in mind that the way personal computing has trended in the past is that the server technology ('big iron' as some call it) tends to be a precursor for what is going to happen in personal computers.

Decades ago, there was the mainframe and dumb terminals. Then, the technology got small enough that everyone could have their own mini-mainframe. Thus, personal computers were born. Mainframes continued to get faster and more streamlined, making new advancements to speed up execution (hyperthreading, multiple processors, larger caches, branch prediction, etc). This eventually made its way into personal computers as the price went down.

Today we are already seeing the benefits of the move to multiple cores/processors. We are also at an interesting junction in the split of server technology. On the one hand, you have Intel moving with more and more general purpose cores. On the other hand, you have AMD buying ATI and including some of their tech into AMD's processors as a Specialized Processing Unit, IBM and Sony developing the Cell as another format of General CPU + multiple specialized processing Units.

It's an interesting time to be following processor tech, in that no-one is quite sure yet which will be the better solution, a small number of general cores with specialized cores that can be used as needed, or a larger number of general cores. They each have their own strengths and weaknesses, and each could easily be viewed as a 'waste' of die space by proponents of the other view (why have a separate core for XXX if task XXX will never be needed? Why have a whole 8 general cores when you will only ever need 4? etc)

To say that the cell is not useful for normal (read: games) programming is only true because it is still remarkably new tech. As compilers get better at SIMD-izing code, and new programming toolkits are made to make SIMD-izing easier (IBM has a really cool one under development (octopus if you are interested in looking it up), and I assume other companies do as well).

These kind of tools will only improve with time, whereas we are reaching the point of diminishing returns with standard CPU compilers; it has yet to be shown how far we can go with specialized processors. I agree that the Cell is not as good for use in games yet; Sony definitely jumped the gun on when it would be ready for mass acceptance. Next generation the compiler technology for the Cell and processors like it will likely be so much better that it could be used with very little extra work; now it is just too early in the Cell's lifecycle.


As it stands today, most server applications are built to take advantage of multi-processors/cores. This is nothing new there. Multi-threaded gaming in general is much newer. IBM was in a position to put the Cell right to work. They new this. What Sony did with it wasn't their problem. So far as gaming applications are concerned, it remains to be seen if the Cell can deliver. As it stands today, the Cell is clearly not the Swiss Army Processor. As I am fond of saying whenever people start to put it on too high of a pedistal, the proof is in the pudding. When it lives up to the potential that many Sony fanboys think it has, then I'll believe it. Until that point, they are just parroting Sony's marketing.

As far a technology goes, it is interesting to see Processor tech fragment in that way. Since engineers have had to find another way to get performance than just increasing clock speed, processor tech has gone in different directions. Some solutions, however, may not be great for all applications. In all likelyhood, there will not be one better technology that serves everyone.

I give that post a 9.8.

 What's with this post/thread rating thing , i don't get it. 

 




Profcrab on 09 June 2008
MikeB said:
@ ProfCrab

I've always said that IBM came out much better on the Cell deal than Sony did. IBM got a great server processor out of it and Sony got a great server processor out of it . . . that can, in a far less effective role, run games.


To quote a friend (of PS3coderz and his many Cell articles fame):

"As for performance there's plenty of research papers out there now showing Cell is not only fast on the areas it's designed for but also in many areas it's not designed for. On raw performance anything less than 8x faster than an x86 core is *low*."

The Cell is a most excellent chip for gaming and multi-media. The only real issue is porting legacy game engines. Building an engine from scratch results in no issues to harvest its potential (of course some initial R&D may be required to get a good idea of how to best approach game engine design).

The SPU code should ideally stick to half- or single precision format, be designed to run parallel as possible (which requires some planning talent) and be cut down into small enough pieces if needed to fit the LS (for example fetch new data and process other data simultaneously streaming data around the super high bandwidth ring, a multi-buffer design).

Main issue: The Cell is very different.
Main advantage: The Cell is extremely powerful.

Comparing the 360 with PS3 CPU, the Cell's PPE is roughly the equivalent of 1 360 core, however the PS3 Cell has 7 SPUs which are considerably more powerful than the PPE can be for game, plain number crunching and multi-media code.

Ahem, in the words of Cuba Gooding Jr. in Jerry Maguire, "Show me the money!"

Where is it?  Where are the games that are so incredibly better looking than 360 games?  Again, you are parroting the marketing.  Get back to me when the games come.

 I give that post a 9.0.



Thank god for the disable signatures option.

Million on 09 June 2008
Profcrab said:
MikeB said:
@ ProfCrab

I've always said that IBM came out much better on the Cell deal than Sony did. IBM got a great server processor out of it and Sony got a great server processor out of it . . . that can, in a far less effective role, run games.


To quote a friend (of PS3coderz and his many Cell articles fame):

"As for performance there's plenty of research papers out there now showing Cell is not only fast on the areas it's designed for but also in many areas it's not designed for. On raw performance anything less than 8x faster than an x86 core is *low*."

The Cell is a most excellent chip for gaming and multi-media. The only real issue is porting legacy game engines. Building an engine from scratch results in no issues to harvest its potential (of course some initial R&D may be required to get a good idea of how to best approach game engine design).

The SPU code should ideally stick to half- or single precision format, be designed to run parallel as possible (which requires some planning talent) and be cut down into small enough pieces if needed to fit the LS (for example fetch new data and process other data simultaneously streaming data around the super high bandwidth ring, a multi-buffer design).

Main issue: The Cell is very different.
Main advantage: The Cell is extremely powerful.

Comparing the 360 with PS3 CPU, the Cell's PPE is roughly the equivalent of 1 360 core, however the PS3 Cell has 7 SPUs which are considerably more powerful than the PPE can be for game, plain number crunching and multi-media code.

Ahem, in the words of Cuba Gooding Jr. in Jerry Maguire, "Show me the money!"

Where is it? Where are the games that are so incredibly better looking than 360 games? Again, you are parroting the marketing. Get back to me when the games come.

I give that post a 9.0.


Your reading comprehension obviously isn't on the same level as your post rating skills , he blaitantly states in his post that the thread cell is "very different" but also "extremley powerfull".

I'm no technical expert but i'm guessing that as meaning that the Cell proccesor technology is still quite new , it has the potential to exceed the XBOX 360 but the neccasery tools/knowledge hasn't reached the level neccasery to do so. This isn't marketing spin believe it or not.




Profcrab on 09 June 2008
alpha_dk said:
 

I agree I was commenting more on the now-bolded part, as most 'regular' programs are built upon a layer that is in the process of becoming closer to server tech. Servers are a very good guide to where General Computing is going, I just think Sony entered this next phase a bit too soon. Not only were the programmers not ready for the new paradigm, but the other supporting technology that usually has a greater chance to mature before reaching them is also still unmatured because the tech is so different. I still feel that in 10 years, the Cell *might* be a good choice for development.

Off topic: What is the deal with rating posts all of a sudden? I'm not on much for a week cause I had to move, and all of a sudden there's new memes that I don't get. I give that meme a 0.5.


I can see that with everone being dragged kicking and screaming into the multi-core world.  Server processors will still in some ways differentiate from PC processors to be tailored more to running server applications primarily.  As it stands with the Cell, it looks like we agree that the server side was definitely ironed out first (this is where I say that IBM.  We will have to see if it is as competitive in the future with the other technologies.  Sony certainly has an incentive to get games to run better on it, but so far it immature as you said.   If Sony wants BC with the PS4 though, the are going to have to stick with it.  I don't imagine it would be an easy processor to emmulate.  By that point, it will be more mature.

I give that post a 9.8. 



Thank god for the disable signatures option.

alephnull on 09 June 2008

Yay, once again nobody knows what they are talking about.

 @ All those who think cell was designed as a "server" chip.

The thing a SPE's are not good at are running what they used to call "ring 0" instructions back in the DOS days. These are things that happen alot when you do things like oh say opening a socket. But then again what kind of server would need to do that alot, so I guess you're right.

Anyhow everything is going to be in userspace as we are all going to be running mircokernels anyday now, right?

The cell is hard to program for because there is no cache-coherency between SPEs. This used to be the norm back in the Super-Mini days but single chips kept increasing in clock speed and the age of super computers died.

Well things have come full circle. They can't keep increasing clockspeed w/o assigning a nuclear generator to each house and they can't keep increasing the number of cores and maintain cache-coherency.

@Profcrab

There is a reason our research group bought a cluster of PS3s for our simulations. The reason the games aren't out is because 1) is today's average industry programmer is not what he/she used to be and 2) Sony released way before they were ready, there was such a shortage of cell chips none of the library developers could get ahold of them. The linux on cell group manager at IBM told me even they had trouble getting their hands on systems and this was last summer.