STI (Sony-IBM-Toshiba) has said all along that they plan to see Cell commercialized far beyond the upcoming PS3, so it was no surprise when Mercury Computer Systems announced back in June that they planned to use the Cell in unspecified aerospace, defense, and medical imaging applications. Today, Mercury unveiled the first fruits of those plans, in the form of a dual-Cell blade server.
Does the Cell processor support out-of-order execution? There was an article a while back where developers talked about what they thought about the next-gen consoles and a developer mentioned that none of the new CPU’s supported it.
With out-of-order execution being the norm, do compilers optimize for multiple pipelines or do they leave it for the CPU?
Cell is a radically different design. The core is based on the PowerPC specification, but not the same as any existing cores. The eight SPUs have their own opcodes, and maximizing performance requires a completely different way of thinking. It’s a new dog that doesn’t do old tricks.
i would be interested in finding out fast how the cell compiles code, the fact the ps3 may ship with a linux kernel has me interested.
if this can cut down on my gentoo compile times i’d be happy, failing that some numbers on video transcoding would also be nice.
hopefully one of the hardware sites will be able to get hold of one and benchmark the cell to see what its actually capable of.
“hopefully one of the hardware sites will be able to get hold of one and benchmark the cell to see what its actually capable of.”
The benchmarks will be underutilizing the SPUs if they attempt generic code approaches. Unless they do SPU code, all they will be doing is benching the PPE. I’ll say straight up – benchmarking the PPE will look like Cell is slow but ignores that “Go really fast” button.
Seems a little early you would think IBM would release a product first. Either way it takes more then a port to really optimize for a new platform.
Does the Cell processor support out-of-order execution? There was an article a while back where developers talked about what they thought about the next-gen consoles and a developer mentioned that none of the new CPU’s supported it.
The PPE supports it; the SPEs don’t, they’re essentially static issue PPC601-like CPU cores.
Nope, the PPE is In-Order I’m near 100% sure.
The PPE and SPEs are in-order.
Modern compilers have lots of scheduling optimizations, even for OOO processors.
You’re never going to run a compiler on the SPEs. Compiling code on a PPE will be slower than on a recent desktop PC, but this is only a problem for Gentoo users who waste their lives compiling. Serious Cell developers can just get a G5 to do their builds on.
It’s funny how people manage to create problems where they weren’t. When I was young, one guy built it and you just *distributed* the compiled version to everyone. Damn, that was efficient!
How simple that was… I wonder what the Gentoo people where thinking about. The only difference from running Seti@Home is that at least you can work while the cpu is being used for some hopeless repetitive task.
When I was young, one guy built it and you just *distributed* the compiled version to everyone. Damn, that was efficient!
I’m sure you’re trolling, but I’ll bite anyway. What kind of system did that one guy build the executables for? Probably i386, because who needs those fancy new instructions and scheduling for later processors, right?