backtop


Print 38 comment(s) - last by Goty.. on Apr 25 at 4:53 PM


Cell Broadband Engine in its natural habitat
First ever course on the Cell Broadband Engine completed

The Cell Broadband Engine is widely known as a powerful but rather complex chip to work with. IBM and the Massachusetts Institute of Technology (MIT), however, are helping to educate students on the unorthodox processor with the first program in the U.S. structured around the capabilities of the Cell Broadband Engine. IBM, Sony, and Toshiba collaborated on helping to fund the course and Sony group provided the PS3 hardware to be used by students.

During the four-week course in January, students not only learned about the new microprocessor, but designed and implemented projects to run directly on PS3 system using open standards software. The student team with the best project – a 3D version of the classic pong game – later presented their work and discussed their experience at the Game Developer Conference in March 7 2007.

The course, which focused around introducing parallel programming to students, was taught by Saman Amarasinghe, a professor in MIT's Department of Electrical Engineering and Computer Science, and Dr. Rodric Rabbah of IBM.

“The fact that students – with no background in parallel programming or the Cell/B.E. – were able to get their projects done from scratch in just about one month largely goes to show the capability and determination of our students, coupled with the availability of a robust toolchain for Cell/B.E. development,” said Amarasinghe.

“Cell/B.E. is going to be an underlying architecture that has the potential to be included in a wide range of industry applications and solutions in the future,” said Dr. Rabbah. “This course was able to break down the details of a highly complex microprocessor and challenge students to see where the performance, power and versatility could be applied outside of gaming. Based on the feedback we received from the students, it was a tremendous success.”

A Web site hosted by the Computer Architecture Group at MIT has more information on the course, including demo source code and video. IBM is also currently hosting the Cell University Challenge programming contest for students in 25 different countries, offering cash prizes and awards for the most innovative applications on the Cell/B.E.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Would still be nice if...
By exdeath on 4/24/2007 12:51:14 PM , Rating: 2
Would still be nice if it had universal homogeneous cores.

Get rid of the PPE and add 2-4 more SPEs with a few more general purpose instructions.

Biggest pain in the ass with the Cell is segregation of PPE vs. SPE code and data formats.




RE: Would still be nice if...
By Aikouka on 4/24/2007 1:08:12 PM , Rating: 2
How would you combat the lack of branch prediction in the SPEs?

Personally, I see the biggest issue of Cell as being splitting up the tasks into jobs that the SPEs are adequate for. They're quite good at pure number crunching, hence why the Folding@Home performance is good.


RE: Would still be nice if...
By exdeath on 4/24/2007 6:44:20 PM , Rating: 2
The same way branch prediction is handled on the ARM7 in the Gameboy Advance: you don't.

SPEs are not that deeply pipelined, and fetching is from the on die local store. Something as simple as a branch delay slot and branch hint mechnism would suffice (which the SPE already has if I remember)


RE: Would still be nice if...
By saratoga on 4/24/2007 6:53:14 PM , Rating: 4
quote:
The same way branch prediction is handled on the ARM7 in the Gameboy Advance: you don't.


Branch latency on ARM7 is 3 clocks as I recall. Its 17 clocks on the SPE according to RWT . I'd say they're not really comparable. ARM7 has a 2 cycle latency on load words from it's L1 cache. Branches are practically free. Conversely they're ungodly expensive on the Cell.

quote:

SPEs are not that deeply pipelined, and fetching is from the on die local store. Something as simple as a branch delay slot and branch hint mechnism would suffice (which the SPE already has if I remember)


I don't know how deep the pipeline actually is, but the ARM7 core you mentioned hits about 150MHz at 90nm using 3 stages. Cell hits at least 3.2GHz. Thats at least a 15x difference, so theres got to be a fairly deep pipe involved.

Regarding a branch delay, that'd be a trick to cover a 17 clock delay.


RE: Would still be nice if...
By exdeath on 4/24/2007 10:04:10 PM , Rating: 2
That bad? Guess I made too many assumptions on the simplicity of the SPEs and the speed of their local stores.

My point was though, it would have been preferable to have all the cores be the same, instead of having two architectures on one chip. If they had scraped the PPE they could have added better branch handling to the SPE cores AND still added 2 more SPEs.

Programming for different architectures on the same platform is a pain in the ass from a tool chain logistics point of view. On the PS2, at least the VU programs and the IOP .irx modules were purpose specific modules that were uploaded on initialization and started for the duration of the main application such that the VUs and the IOP were treated as hardware ASICs. Main application code from then on was single threaded and multi tasked by the EE.

If you want to write efficient multi threaded main application code on the Cell, you basically have to write everything form the perspective of the SPE being your primary method of execution. Forget the PPE.


RE: Would still be nice if...
By Samus on 4/25/2007 3:56:48 AM , Rating: 2
*caugh* pwned!

haha, just HAD too ;)


RE: Would still be nice if...
By Aikouka on 4/25/2007 8:23:32 AM , Rating: 2
Here's an interesting little quip I found on Wikipedia about Cell's SPE:

"The Cell is designed to compensate for this with compiler assistance, in which prepare-to-branch instructions are created."

In a discussion about Cell's lack of branch prediction when compared to modern desktop processors.

In this section, they talk a bit more about what they do to "organize" the multitasking in cell and how they use the PPE... not too sure that you'd want to organize it using a SPE.

http://en.wikipedia.org/wiki/Cell_Broadband_Engine...


RE: Would still be nice if...
By bdewong on 4/24/2007 1:15:13 PM , Rating: 2
I really wonder how difficult it is to program for the Cell/BE. I haven't looked into coding for it but I can't imagine that it would be that much more difficult than parallel programming in general, at least for programmers. I should probably take the time to go look into that.

Anyway, this is great news for programming, as I think most programs are going to have to switch to parallel. If any of you have a chance to take a parallel programming class, take the opportunity.


RE: Would still be nice if...
By saratoga on 4/24/2007 2:32:13 PM , Rating: 1
Cell is basically a parallel DSP system. So it'd be about as hard as programming a typical multiprocessor DSP system.


RE: Would still be nice if...
By exdeath on 4/24/2007 6:55:09 PM , Rating: 3
It's not that it's difficult, just annoying.

On every other platform, you write your C++ program like you normally would, with basic multithreading provisions to make your functions re-entrant and thread safe. Compile once, with one compiler, and you take advantage of 2 cores of 32 cores with one .obj file that can run equally on any core multiple times. This is true for PC, Xbox 360, etc.

With the PS3 the biggest pain is that the SPEs architecture and memory model are completely different from the PPE.

You as the developer are constantly aware of what is SPE code and data and what is PPE code and data, and the boundary between the two, and the separate tool chains, including SPE programs as data in your PPE programs, etc. Think if your multi core PC had 1 x86 core and 1 68030 core. It's a lot more annoying and time consuming than if you just had two identical cores of either one or the other, as you have to deal with two entirely differing memory models, architectures, etc. You can't just have a function that runs on any available core that writes it's answer back to universally accessible memory. Instead of you have worry about marshaling code and data back and forth across different architectures.

My advice to beginning Cell programs is focus on the SPEs and don’t really worry too much about the PPE until later. Any code compiled for the PPE is guaranteed single threaded which defeats the purpose of multi core.


RE: Would still be nice if...
By exdeath on 4/24/2007 6:56:53 PM , Rating: 2
correction: "take advantage of 2 cores or 32 cores"

Also, with the VUs in the PS2, they weren't so bad, because you just coded a set of fixed functions for them such as transform, lighting, subdivision surfaces, etc, upload them once for the life of the program, and treat the VUs like ASICs that are programmable for a application wide specific task.

Not so with a multi core generic CPU.


Silly Promotional Stunt
By Flunk on 4/24/2007 11:59:23 AM , Rating: 2
A course that teaches students about parallel processing is great but just because they created projects able to run on cell in just a month does it mean that they were utilizing the chip to it's potential. This is just a marketing stunt that really doesn't mean anything.




RE: Silly Promotional Stunt
By FITCamaro on 4/24/2007 12:24:10 PM , Rating: 5
Well while they definitely didn't fully utilize the processor it does give the students an introduction to threaded and parallel programming which will be invaluable to them.

As another said, this is good for the Cell as a chip, not as a processor in the PS3.


RE: Silly Promotional Stunt
By saratoga on 4/24/2007 2:34:37 PM , Rating: 2
quote:
Well while they definitely didn't fully utilize the processor it does give the students an introduction to threaded and parallel programming which will be invaluable to them.


Again, why do you think this? I'd expect them to pick projects that parallelized easily across the SPEs, and thus essentially drove the CPU as hard as it was going to go. At least it wouldn't make sense to practice on highly serial problems.


RE: Silly Promotional Stunt
By saratoga on 4/24/2007 2:33:18 PM , Rating: 1
quote:
A course that teaches students about parallel processing is great but just because they created projects able to run on cell in just a month does it mean that they were utilizing the chip to it's potential.


Why not? Its not exactly rocket science to do so for typical DSP tasks.


Developers are cry-babies....
By Goty on 4/24/2007 4:01:32 PM , Rating: 1
It seems to me that developers with years of experience with programming (albeit single-threaded programming)that have been complaining about programming for the Cell B.E. need to man up and get to work. These students are going to get out of MIT and whoop some ass.




RE: Developers are cry-babies....
By saratoga on 4/24/2007 6:56:54 PM , Rating: 2
Making the hardware far more difficult to work with at a time when the software complexity has already strained the budgets and talents of the best in the industry has nothing to do with the need to "get to work".


RE: Developers are cry-babies....
By Aikouka on 4/25/2007 8:29:20 AM , Rating: 3
You can't just deem all developers as "cry babies" when they claim that programming for Cell isn't worth their time. Cell's SPEs are the most efficient when it comes to pure number crunching, which is not applicable to every project. Also, not every project in the world wants or needs that much floating point processing power or even desires that much parallel processing ability.

No one needs to "man up and get to work"; people who think Cell is applicable to every task to need man up and get an education.


RE: Developers are cry-babies....
By Goty on 4/25/2007 4:53:13 PM , Rating: 2
For those of you saying I'm in the wrong calling these developers cry-babies, just take this quote from Yosuke Hayashi, project leader and director for Ninja Gaiden, Ninja Gaiden: Black, and Ninja Gaiden: Sigma for the PS3.

quote:
I think all those developers who are saying, “We don’t want to do a PS3 game,” or “It’s really difficult to do it,” should shut up and make their games. If you have time to complain about it, then you should be spending your time working on getting the most from the hardware.


Gee, I guess having the same opinion as an actual developer makes me wrong.


First Prize
By JustKidding on 4/24/2007 10:57:31 PM , Rating: 3
quote:
BM is also currently hosting the Cell University Challenge programming contest for students in 25 different countries, offering cash prizes and awards for the most innovative applications on the Cell/B.E.

Just found out that first prize is $10,000 and a Nintendo Wii.




:0
By DarthKojima on 4/24/2007 11:05:44 AM , Rating: 2
well that's cool.. lomg live MIT.. proud to be a ME.




Interesting optimization opportunities
By fsam on 4/24/2007 11:15:28 PM , Rating: 2
I've been doing some algorithm design and evaluation research on the Cell BE for school. One thing people fail to realize is the local store on SPEs presents a HUGE optimization opportunity. On conventional processors, the developer has no control over the on-board cache. On the Cell BE, one has full control of the 256KB per core. Let's put it this way, no one understands the flow of memory accesses of a given algorithm than the developer of the algorithm. Thus, one can potentially design a cache system that has much higher cache hit rate than a cache that's fixed by the architecture. Replacement strategies and prefetching may be far more sophisticated to suit the needs of the particular algorithm.




Things like this will make PS3 a success
By BPB on 4/24/07, Rating: -1
RE: Things like this will make PS3 a success
By d0gb0y on 4/24/2007 12:00:50 PM , Rating: 5
I see this as a way to help the Cell be a success, but as the course is to open up the Cell to other uses then gaming, I don't see this really helping the PS3. It's neat that the PS3 can do other stuff, but it should excel at gaming to live long and prosper.


By Garreye on 4/24/2007 2:41:33 PM , Rating: 5
I agree with you on that for sure. The thing is I think that might be what IBM wants. PS3 isn't the only place where cell processors are used, and if IBM can get more uses out of cell that they are going to sell more of them...here's an example:

http://www.dailytech.com/article.aspx?newsid=6866

One of the big things about putting cell in the PS3 is being able to bring the price down. When they can manufacture such a large number of cell CPUs as the do with the PS3 they can produce them much more cheaply, and can use that price advantage to sell cell CPU in areas other than the PS3.


RE: Things like this will make PS3 a success
By Hydrofirex on 4/24/2007 7:42:12 PM , Rating: 3
Totally wrong. One of the major strikes against the PS3 was the "overly difficult programming" that the cell chip required. By proving that the chip is in fact not so monstrously difficult to program for they are sending a clear message to the game developer community. Also, consider that cell is still a General purpose CPU - just like the Xeon inside the Xbox (Used in tens of thousands of outdated CPU's). Finally, some outside sales could only help Sony and the PS3 by helping to justify the development costs of the this part and providing a little black ink to go with all the red they're going to have to eat while the market catches up with them in the console arena.

HFX


By robber98 on 4/24/2007 9:08:44 PM , Rating: 2
I think that's what d0gb0y said

quote:
I don't see this really helping the PS3


RE: Things like this will make PS3 a success
By giantpandaman2 on 4/24/2007 12:01:21 PM , Rating: 4
Compared to development tools and momentum of x86 the Cell stuff is very weak. Just like few people develop for IA64 I don't see a whole lot of people thinking, oh, let's develop for Cell. I think it's pretty clear that new CPU architectures have a HUGE barrier to entry. I mean, that's why x86 rules almost everything now.

That's why MS was so smart to go with its XNA dev. tools for the X360. Though the 360 is powered by PowerPC chips, you develop almost like you would for a x86 architecture. Unfortunately, I don't think that can be done with Cell. The architecture of Cell is just too different. So Cell is going to run up against the same barriers that all other architectures have run up against when competing with x86. And it'll lose, just like all other architectures have.


RE: Things like this will make PS3 a success
By tmw1735 on 4/24/2007 1:02:22 PM , Rating: 2
When I was in college back in the 80s, everyone where using DEC VAX-11, IBM s370 mainframe, etc. The computer science boys were playing with a new system called Sun Workstations running Solaris on Motorola 68030 processors. The CS people love it although no body else outside knew what it can do, and brushed it off as "just another new system".

Fast forward 15 years Sun Microsystems were everywhere and you know the history.

Can't say the Cell/BE will follow the same path, but don't under estimate what a group of college kids can do.


RE: Things like this will make PS3 a success
By theapparition on 4/24/2007 1:53:44 PM , Rating: 2
What is Sun using these days? x86


RE: Things like this will make PS3 a success
By BPB on 4/24/2007 1:59:48 PM , Rating: 2
And mainframes are still the backbone of many, many large corporations. Interesting history, but it doesn't mean it has to repeat itself.


By tmw1735 on 4/24/2007 2:16:53 PM , Rating: 2
Sun Microsystems went from Motorola 68030 processors, to Sparc, to x86 processors, etc. But the importance of the story is the process - Sun Micro chartered the Unix/Open System/Client Server model and it started small by a few graduate students from Standford University.

Cell/BE *may* get popular from a few students from MIT. Just a thought.


RE: Things like this will make PS3 a success
By BPB on 4/24/2007 1:55:10 PM , Rating: 2
I think if you were to meet the big wigs at MS and Sony you'd find that MS really wants the Xbox 360, and Sony really wants the PS3, to succeed beyond the gaming world. Both companies are trying to create media center to be actual centers of media in homes. They both want folks to tie their gaming machines into their PCs and entertainment centers. If they can get people to see the potential of this, especially with things like 80MB to 100MB FIOS connections for homes where it's available, the sky's the limit. I think that if Sony keeps to this vision the PS3 will over time be bigger than the PS2 could have dreamed. The openness of the platform means people will find all kind of uses for it beyond gaming. My buddy's getting one and he doesn't game one bit. He wants to use the cell processing and the blu-ray. I'm sure he's not alone.

As for development tools, they'll come. It's not like Sony is some sort of start-up just beginning to get its feet wet. The big question is: does Sony stick to its guns and give the PS3 time, or give in and go another route?


RE: Things like this will make PS3 a success
By kalak on 4/24/2007 2:43:01 PM , Rating: 1
quote:
MS really wants the Xbox 360, and Sony really wants the PS3, to succeed beyond the gaming world.


NOPE. They are VIDEOGAMES... They really want you to adopt their Videogames, they want to sell a "idea", but this will not work in real world and is pure propaganda....

quote:
if Sony keeps to this vision the PS3 will over time be bigger than the PS2 could have dreamed.


When and IF it happens the time to PS3 was over and then we have PS4 (or PS3 Ultra, whatever...). The fact is that this Videogames (yes, they ARE videogames ! Meant to play GAMES on it ! They are NOT PCs !)have a short life cycle...
I can upgrade my PC to adapt for new technologies, but I can't "upgrade my PS2 to a PS3".... I have to buy a NEW Videogame...


By BPB on 4/24/2007 4:28:48 PM , Rating: 2
Yes, there's a huge amount of money to be made on games and gaming hardware. Still, I don't think anybody can expect gaming only consoles to exist much longer. Yes, the Xbox could play music CDs and such, but there's a reason for MS and Sony to create the current generation such as they did. They know people want to simplify their entertainment. They don't want receivers and amps and pre-amps and dvd players and on and on. I think the current gen consoles are the beginning part of the process to simplify it all. I like the fact that the 360 ties in so nicely to Vista. It opens up so much as bandwith grows and programmers and hardware makers get more creative. I think the PS3 will try, is trying, to do similar things. Processors of all stripes are getting so powerful hardware makers simply don't need to create boxes dedicated to one function. Earlier on my PC I was recording a TV show and playing LOTR ROTWK at 1600x1200. I want to be able to do that and more with a console. It's coming, I'm sure of it. Think TiVo and the PS3 combined, only better.

Oh well, I like what Sony is trying to do with the PS3, I hope they succeed. I think we'll all be happy if it does. If they don't succeed with the PS3 then I hope the next gen has the same intentions, but maybe uses a platform that's more widely accpeted.

I think years down the road we'll be remembering the good ole days of single function game consoles.

quote:
They are NOT PCs !)have a short life cycle...
I can upgrade my PC to adapt for new technologies, but I can't "upgrade my PS2 to a PS3".... I have to buy a NEW Videogame...
Eventually we'll reach a point where you may not need to buy a new console. Maybe the next Sony goes beyond the openness of the current and makes not only a hard drive upgrade easy but other things as well, much like a PC. Time will tell.


By Ravenlore on 4/24/2007 5:45:17 PM , Rating: 2
Quote
NOPE. They are VIDEOGAMES... They really want you to
adopt their Videogames, they want to sell a "idea",
but this will not work in real world and is pure
propaganda....

Yea right and that is why Sony uses Linus as an OS and added a BluRay player and Memory card slot, Wifi... IT is ment to be the middle of the media center and they hope that you remember Sony as a brand and easy to use service also, for music, and movies...

And in a few years when HDTV sales explode and BluRay is all the rage PS3 will not only pass PS2 in sales but eclipes it as PS3 will last longer than the 6 years that PS2 did. And PS2 sales is still HUGE, even after 1 year of Xbox360 on the market.

Look at Sony's Home and "copy of live" network. which is not only a game matching but a 3D Myspace!!! And how big is myspace!! look foward to a connect to Google's Youtube or a copy of that as well connected to Home or another part of Sony's online plans.

PS3 will be hitting the mark with Price as HDTVs booms and BB spreads. And NO you can not upgrade a PS3 which is great as many people do not. And as many people found out with VISTA it is just a lot easier to buy a new computer with Vista than try to update to it.


RE: Things like this will make PS3 a success
By saratoga on 4/24/2007 2:30:23 PM , Rating: 2
quote:
When I was in college back in the 80s, everyone where using DEC VAX-11, IBM s370 mainframe, etc. The computer science boys were playing with a new system called Sun Workstations running Solaris on Motorola 68030 processors.


Given that neither Solaris nor the 68k was a particularly great success, I'm not sure what you're getting at.

quote:
Fast forward 15 years Sun Microsystems were everywhere and you know the history.


15 years would be about when Sun was in the toilet (2001 or so) and then starting carrying x86 and Linux systems. Really if anything this would suggest that those students didn't much care for Sun. At least not enough to make any of the products they pushed at the time a success.


By mars777 on 4/24/2007 8:32:36 PM , Rating: 2
quote:

quote:
Fast forward 15 years Sun Microsystems were everywhere and you know the history.

15 years would be about when Sun was in the toilet (2001 or so) and then starting carrying x86 and Linux systems. Really if anything this would suggest that those students didn't much care for Sun. At least not enough to make any of the products they pushed at the time a success.


Fast forward 15 years Sun Microsystems were everywhere

What didn't you understand? I am Croatian and not o good at english but what i read is:

In this 15 years Sun was everywhere. - and that is true.

It doesn't mean what has happened after this 15 years. But we are not looking what wil the PS3 be in 15 years but in the next 6...


"We don't know how to make a $500 computer that's not a piece of junk." -- Apple CEO Steve Jobs











botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki