Cell and Java/.net
- Thread starter Tsmit42
- Start date
Java will run on anything you've a VM for. As for performance, there's a PPC like CPU running at 4 GHz (in the ISSCC implementation anyway) so it'll be as quick as anything else out there.
So why is it so hard for Cell to move into the desktop arena if java/.net application could run on it? Given that it a 4ghz ppc, it should run pretty fast even without utilizing the apus like you said. Even if it don't cause a big buzz at first, it will cause some developers into thinking about writing their programs in java/.net. If cell could give an performance advantage will running those programs, it could cause a real stir. Imagine Microsoft Word coded in java/.net and running on cell/mac/pc/linux.
BOOMEXPLODE
Regular
Java and .net are not at all the same thing...
Sonic said:
I thought so, I was not trying to be rude or anything, but just saying that they worked the pretty much the same way. I know they are not exactly alike but close enough to ask the same question for. They both used mananaged libraries and programs are compiled into a imtermediate langauge for jit compilation. If the platforms are not interchangable in relation to the question, please fell free to enlighten me.
Basically the power core should be similar in speed to any other similarly specced core. Probably about the same as a current High end Mac.
The APU's would be significantly more difficult to utilise. Java programs can be large, and existing runtimes are large so any reasonably sized program is unlikly to fit inside the APU's 256K of local memory.
Given that it's going to be cheaper to omit the APU's if your primary goal is running Java or .Net apps. At which point why use Cell at all.
The APU's would be significantly more difficult to utilise. Java programs can be large, and existing runtimes are large so any reasonably sized program is unlikly to fit inside the APU's 256K of local memory.
Given that it's going to be cheaper to omit the APU's if your primary goal is running Java or .Net apps. At which point why use Cell at all.
I think talk of Cell not making the move into the desktop is more a case of price and software (Windows) compatibility and not the inability of the processor. Sure the APU's might not be the most useful things for wordprocessor, but the PE will be. However if the PE won't run Windows, you won't run Office, so no-one will want to use a Cell PC as it hasn't the software.
If there were a huge library of portable office apps, written in a JIT based language like Java, it wouldn't take a lot to move those apps onto Cell. But there isn't and until MS want to port their apps people won't buy Cell PCs, and MS won't port their apps until there's enough Cell PCs to sell them to.
If STI can keep the price of Cell down maybe Media PCs will become an option, with a variation of Linux or something as OS and Linux based apps, and fearsome video editing performance for specialist Cell-enhanced code. It won't happen any time soon though, if ever, while CPUs specifically targetted at that arena will offer enhanced Windows performance.
If there were a huge library of portable office apps, written in a JIT based language like Java, it wouldn't take a lot to move those apps onto Cell. But there isn't and until MS want to port their apps people won't buy Cell PCs, and MS won't port their apps until there's enough Cell PCs to sell them to.
If STI can keep the price of Cell down maybe Media PCs will become an option, with a variation of Linux or something as OS and Linux based apps, and fearsome video editing performance for specialist Cell-enhanced code. It won't happen any time soon though, if ever, while CPUs specifically targetted at that arena will offer enhanced Windows performance.
I think that if a cell pc can show much improved performance over a wintel pc with applications like video editing, graphic applications, sound editing, etc. or even everyday applications like mp3 ripping/encoding, compression programs, etc. , users of these programs will be inclined to give a cell pc a try. Linux don't give their users a much improved performance over a windows pc and the mac performance has been an ever closing gap. Macs are also very expensive compared to their pc counterparts, and with a cell processor that is going to be hugely massed produced and put into a $300-$400 console (along with ram, blu-ray, and a gpu), you can expect the price to be competitive.
Also note that MS has made office for the mac, so it is not far-fetched that they could make office using their much beloved .net technologies. Given that Office is the only program that MS has of any value since people dont care what browser they use to surf the web(IE/firefox), or what plays their music as long as it plays(media player/winamp).
IF they could come out with a good interface and easy to use OS, I think a cell pc could have lots of potential.
Also note that MS has made office for the mac, so it is not far-fetched that they could make office using their much beloved .net technologies. Given that Office is the only program that MS has of any value since people dont care what browser they use to surf the web(IE/firefox), or what plays their music as long as it plays(media player/winamp).
IF they could come out with a good interface and easy to use OS, I think a cell pc could have lots of potential.
Tsmit42 said:
You are living in a fantasy land. If Window will not run on the Cell "pc", then why do you think everybody will switch over even it is more "powerful"? People will not just jump over to linux. People want to use what is familiar to them and if it isn't familiar, they won't use it. Also, Cell isn't "hugely" mass produced compared to processors from Intel or AMD. Development of these cell PCs will take a LOT of money on top of what is already spent unless the cell workstations have drive controllers, firewire, usb, etc. Also, you are mildly retarded if you think Office is the only program MS has of any value. Windows?? Also, people DO care what they surf the web with. Not all sites allow the use of Firefox because of ActiveX controls and VB Script.
a688 said:
Fantasy land? Even I know this will not happen over night. I already gave you the reason why people are not jumping all over linux, even though some have, is because it don't offer improved performance and is not very user friendly. I know lots of people that jump up and down over something that lets them get their most time consuming computer task done 10x-20x faster if the cost is about the same.
As far as mass production, of course the cell is not on the same level as all of intel, but much ahead of them on any one chip. 100 million(200 million if 2 PEs) is just with the ps3 and not counting what Toshiba and IBM has planned. Cell PCs would also raise that number.
Familiarity is all in inteface design. If a cell pc OS can provide a simple easy to use interface that is comparible to Windows, then the users will be okay. Don't give them endless options and configs just to run a web browser, let it be a click and go process.
Only thing stopping me from linux right now is games, and with ps3 games that will be easy to port to a cell pc I don't think that would be a problem.
sigh...one can only dream of a world without Microsoft and Windows... 
I for one, use windows mostly because of games. If I could have a cell linux workstation , capable of playing PS3 games, I woudn't think twice about dumping Windows. But that's just my opinion.
About .NET and Java, well....NET is Microsoft, so in my book, it's suspicious as a multi-platform solution. Java on the other hand, still have a long ways to go before it turns into something useful. My problem with Java, is the lack of integration with the GUI environment , which usually results in slow and ugly pieces of software.
I for one, use windows mostly because of games. If I could have a cell linux workstation , capable of playing PS3 games, I woudn't think twice about dumping Windows. But that's just my opinion.
About .NET and Java, well....NET is Microsoft, so in my book, it's suspicious as a multi-platform solution. Java on the other hand, still have a long ways to go before it turns into something useful. My problem with Java, is the lack of integration with the GUI environment , which usually results in slow and ugly pieces of software.
Tsmit42 said:
100 million over what, 5 years? 200 million if 2 core? Nice marketing talk. 100 million 2 core processors != 100 million processors. No OS will provide the interface Windows does besides Windows itself. There are already versions of linux that are point and click but people do not use them as much as they use windows. Also, why will people be porting ps3 games to this cell based workstation? There is a reason why more games don't show up on PCs and its not because they are powered by your dream cell processor.
a688 said:
Yes over 5 years or so? What does that matter 100 million is 100 million. And the possible configuration of cell processor if they are 2 PEs could be 2 seperate chips, we don't know.
No OS provides the interface windows does besides windows? Of couse not exactly the same. Linspire has a nice interface, but that is not the point. Ibm/sony/toshiba/etc. could make a easy to use linux interface, it is not that hard considering what they already done.
And can you stop bringing up why linux failed? It failed for many reasons, and that is why I'm not talking about a fedora port to a cell pc.
Improved performance in multimedia apps and good games would draw many people like me and Alejux away from a windows PC.
Matey, we see your points, and they would be valid if we weren't living in this dimension. But taking over a market where there are more than 1 Billion machines running Windows, all more or less compatible with each other (notice the "more or less"). It's just not gonna happen anytime soon, or else Apple would be there already.
It would be good, but in the end i'm not sure how Cell will trump all over other processors as far as Windows-related applications are concerned. Cell was designed first and foremost to calculate an insane amount of floating point maths, and it might do that well, however the "power" of a computer is not all there.
As it stands now, if MS doesn't support something, that something is dead even before trying to get into the PC market. Not saying it's right (God forbid!), it's just the way it works today and for a long time in the future...
It would be good, but in the end i'm not sure how Cell will trump all over other processors as far as Windows-related applications are concerned. Cell was designed first and foremost to calculate an insane amount of floating point maths, and it might do that well, however the "power" of a computer is not all there.
As it stands now, if MS doesn't support something, that something is dead even before trying to get into the PC market. Not saying it's right (God forbid!), it's just the way it works today and for a long time in the future...
The SPE/APUs will not be suitable for general purpose programming if only because they can only directly access 256k of RAM. I don't think they will be able to implement automatic virtual memory paging even if you tried. In theory you could work like it was a stupidly fast 286 and manually emulate segmented memory by DMAing 16K chunks, but that would more than cancel any potential benefits. The 4GHz PowerPC would probably out perform all 8 4GHz SPEs executing general purpose code. General purpose code references memory all over the place, has lots and lots of branches and rarely coordinates 4-floating point multiply-adds in succession.
As much fun as it is to theorize about porting your favorite interpreted language to the SPEs you must face the fact that they are very special purpose processors designed for stream processing. That means you must structure your code around
1) Lots of identical but completely independant tasks that each only reference tiny, contiguous regions of memory
2) Math-intensive processing
3) Very limited if any branching
If you tried to design a language around that you would probably end up with something like Cg/HLSL.
As much fun as it is to theorize about porting your favorite interpreted language to the SPEs you must face the fact that they are very special purpose processors designed for stream processing. That means you must structure your code around
1) Lots of identical but completely independant tasks that each only reference tiny, contiguous regions of memory
2) Math-intensive processing
3) Very limited if any branching
If you tried to design a language around that you would probably end up with something like Cg/HLSL.
Panajev2001a
Veteran
corysama said:
1.) General Purpose code also suffers from the illness known as locality of reference, of course this happens in various forms depending on the kind of code being executed. There are patents from IBM dealing with code and data exceeding the bounds of local storage and using overlays (code and data would be broken in appropriate modules including a global, always there module which contains instructions and data that are shared between all the modules). There would still be benefits. You would be running 9 things at once even if the PPE was busy doing something else: the PPE could deal with the foreground task while the CELL OS could schedule the background tasks between the SPE's.
2.) 4-way FP MADD is not the only instruction that can be executed: integer processing is not that slow on SPE's and they do have instruction operating on scalar values.
3.) Itanium 2 seems to be able to do fine things with if-then-else conversion: predication. Sure SPE's ISA is not fully predicated as EPIC/IA-64 is, but as the Alpha guys showed with demos on the now defunct EV6x line a good CMOV can suffice.