now that the ps3 is homebrew enabled, we should port dolphin to it (if this is possible) I think that the ps3 should be able to hit fullspeed on most games. The airforce uses it for cluster computing so i think its pretty powerful.
Dolphins core is a jit recompiler that recompiles powerpc instructions to x86. Cell BE is a completely different architecture! We would have to start from scratch. Also the Cell architecture relies on SPEs to provide the high level of ILP that gives it it's strength. Simple put even if we completely remade dolphin from scratch to run on the PS3 it would perform horribly since its practically impossible to design dolphin to take advantage of that type of architecture. Not to mention 256 MB of main ram and 256 MB of vdram is completely insufficient for this type of emulation. The PS3 doesn't even have half the amount of ram that dolphin usually uses. On top of that the PS3 uses a G77 (same chip used to make the geforce 7800 series cards for desktops) which is only enough to run most games at 640 x 480 smoothly.
Sony couldn't even get their official PS2 emulator working well enough on the PS3 and they have access to all the technical docs, schematics, SDK's etc.. Could you imagine trying to get something like Dolphin working on a PS3 with as little documentation as there is for the GC/Wii, not to mention that the GC/Wii are more powerful than the PS2 hardware?
Sorry to break it to you, but the Cell sucks compared to a desktop processor. It bottlnecks itself, as it has 1 core dedicated to checking the work of the rest of the cores.
It's used for cluster computing for 2 reasons:
It's cheap
Performance-per-watt
wanted to ask this question, but felt a bit shy.
however good question, and excellent answer
Quote:Sorry to break it to you, but the Cell sucks compared to a desktop processor. It bottlnecks itself, as it has 1 core dedicated to checking the work of the rest of the cores.
It's used for cluster computing for 2 reasons:
It's cheap
Performance-per-watt
That's not true. It massively outperforms any x86 chip in the market in heavily threaded floating point intensive math if it's done right (you need to write the program in assembly because the compilers for Cell all suck). A linear filter, vertex shader, or video encoding are all examples of something that could be done much more effectively on a 6 SPE Cell BE than any x86 out there right now. For certain applications it's a much better choice than any x86 cpu out their it's just that the majority of programs are integer based, single-threaded, or non-linear and would have no benefit from this type of architecture. Plus x86 chips are also a lot better than cell at a lot of stuff like branch prediction which most programs need more than ILP. It's not that it's not a potentially powerful chip it's that dolphin, like most applications, is the wrong type of program for it.
sure go ahead and do it if you want it that bad
Porting Dolphin to the PS3 wouldn't be even THAT hard anymore (thanks to the x86-less branch), but there are still several problems:
- Dolphin doesn't make use of that many cores (as pointed out before), so performance will be bad
- there's no OpenGL driver we could use (nouveau would need to be ported first), writing a software rasterizer is too time-consuming
- We don't have any non-x86/x64 JIT (just in time recompiler), yet, so one could only use interpreter mode...
- probably Dolphin won't even compile properly on any PS3 toolchain available...
- probably something else I can't think of right now
(09-03-2010, 02:36 PM)NaturalViolence Wrote: [ -> ]Quote:Sorry to break it to you, but the Cell sucks compared to a desktop processor. It bottlnecks itself, as it has 1 core dedicated to checking the work of the rest of the cores.
It's used for cluster computing for 2 reasons:
It's cheap
Performance-per-watt
That's not true. It massively outperforms any x86 chip in the market in heavily threaded floating point intensive math if it's done right (you need to write the program in assembly because the compilers for Cell all suck). A linear filter, vertex shader, or video encoding are all examples of something that could be done much more effectively on a 6 SPE Cell BE than any x86 out there right now. For certain applications it's a much better choice than any x86 cpu out their it's just that the majority of programs are integer based, single-threaded, or non-linear and would have no benefit from this type of architecture. Plus x86 chips are also a lot better than cell at a lot of stuff like branch prediction which most programs need more than ILP. It's not that it's not a potentially powerful chip it's that dolphin, like most applications, is the wrong type of program for it.
I'm pretty sure the Core i- line dominates it in x86. Cell is a 5-year-old architecture. (started 9 years ago, actually)
Depends on what your doing. For a non-linear function that needs branch prediction then yes, but cell still beats even the mighty i7 980X several times over in certain tasks. Example:
http://www.tomshardware.com/reviews/lead...,2523.html
For something like software ray-tracing or rasterizer it's still a better choice than any x86 cpu out there. RISC can be so much more efficient for some things than CISC x86 (not surprising considering how archaic x86 is, can't believe we're still using an architecture that started in 1978). Honestly the architecture kind of reminds me of a gpu in some ways. Like a halfway point between x86 cpus and a g80 based gpu. Versatility of a cpu with the ILP of a gpu.