By using this site, you agree to our Privacy Policy and our Terms of Use. Close
kars on 31 May 2007
Edit Thread
Diomedes1976 said:
TheBigFatJ said:
 


But not videogames .When someone says the X360 CPU is better for "general purpose " they seem to be forgetting we arent talking about a computer here but a videogames machine ,and graphics sound and about everything represented in screen needs mostly single point operations ,so FLOPS (floating point operations per second )is the way to measure them .And the PS3 ,like it or not ,just has some more FLOPS of power that the X360 .Like three times ,to be exact .

Besides ,your 3X more general purpose power of the X360 would assume the 7 SPU of the PS3 have ZERO power applied to general purpose.Thats not the case ,if you have to calculate two times each operation without some help provided by the architechture it evidently reduces the efficiency of the proccesor in half ,but each SPU has as much power as one Core of the X360 (about 25 Gigaflops ) .That lends me to think you simply doesnt know what are you talking about .


 First of all: Programmers can only use 5-6 SPEs!

Second:  You know how you can quickly identify people that know nothing about Computers?

When they cite FLOPS. This says as much about a Processing unit as the cycles. Only with very special programs you can get the power of a cell and even they are far below these rates. There is no way to  avoid wait cycles. In fact if the SPEs are only waiting 50% of their time that's quite good! The more parallel systems you have the more wait cycles you will have. Add to it that many programs can't be executed in parallel and you have another dramatic slow down. 

Third: The Cell was especially developed for number crunching and pipelines. If a program needs access to a lot of memory while it only computes a little it can't be executed in a SPE! In fact even IBM advices to use the SPEs only for computing intensive functions. And even then the SPE might be the wrong choice, because it can only execute functions without a lot of branching in an effective way. There is no way to get around these restrictions! A multi core doesn't know these restrictions.

Fourth: While normal people didn't realize it but no one uses Assembler anymore for these systems. They are simply to complex to do it on such a deep level because programers need to much time and make to many errors and debugging in Assermbler is a lot of fun and takes even more time. You won't really go lower than C or even C++ and competitors and this means that you depend on compilers and compilers always prefer simpler environments, Less compiler bugs and better optimized code. T0o get a better grip on their programs it is advisable to rely on Libraries or even objects, but these structures make the optimization for the SPEs difficult.