> > > All the investment in the current bytecode and class files would be lost. > Some work would need to be done to cross compile the original source or > transform them with a tool to the new system. Not exactly a drop in > replacement... > > Again, this sounds just like an argument for only ever sticking to one processor. Tools exist at every level of the stack, and of course new ones are needed if you change your target. As with everything in business, there's a cost/benefit analysis to be made for any change of this nature.
> One of java's strenghts is the bytecode is an abstraction over the native > cpu which means the jit can do smart things depending on which generation of > a cpu it gets. This argument is actually completely wrong. Generally GPUs > have not been used for general puirpose programming tasks, they are > typically used for stuff thats naturally well suited to parallel execution > like processing an image which is very different from the programs most of > us write today and tomorrow. If you can give an example of a system that > uses GPUs instead of a regular cpu i would be very interested. From my part > of world and limited reading of the net i have yeet to hear of a system > without a core cpu and entirely powered by gpus. Even the old games machines > of yesteryear where all this started all have a core cpu and specialised > helpers. The Amiga had DMA which could do stuff faster than the cpu but > nobody ever wrote their programs entirely as a copper list w/out any 68000. > The same goes for SNES, MD, PSX, PS2 etc. > > I remember similar arguments being made when floating-point co-processors first came out. Of course, that logic has now been rolled back into the CPU, and we all freely use floating-point operations expecting reasonable performance from them. A (GP)GPU can just be seen as an SIMD co-processor. No doubt we'll come to take them for granted as time passes, and expect full language support. Just as we did for floating-point operations. I also imagine that other specialisations will become common as the computing world becomes increasingly parallel. And no, I'm not aware of any system ever running purely on coprocessors, although there are several blades and dedicated boxes now available, such as Nvidia's Tesla range. > Exactly so youve just jumped back to emulation of the bytecode, > classloading and more. Nothing is a first class citizen which means > everytjhing just became a lot slower. > There's already several layers between your source-code and the hardware. There are even layers within the hardware itself. Unless you're writing direct machine code, in binary, you're not a "first class citizen". Even assembly language is translated... Actually, even machine code is ultimately translated to a flow of electrons and quantum effects inside semiconductors. It's turtles all the way down... -- Kevin Wright mail / gtalk / msn : [email protected] pulse / skype: kev.lee.wright twitter: @thecoda -- You received this message because you are subscribed to the Google Groups "The Java Posse" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/javaposse?hl=en.
