I know the developers have written off GPGPU capabilities, but Mercury systems has a neat comparison between the i7, 8641, and the RV770.

http://www.mc.com/uploadedfiles/Comparison-of-embedded-solutions.pdf

Edit : I guess no one else finds the relative comparison of GFLOPS between these processors as interesting as I do.

[ Edited by Jim on 2010/8/20 23:43 ]

@Jim,

The day someone develops an OS, file system and other applications that can all be run on just a GPU, or on a system that basically has a very basic FPGA motherboard with a PCIe slot for a powerful graphics card that the OS, file system and all applications run on, I will pay more attention to the speed difference between the two. (I know someone is probably going to jump in here and show me that it is being done, or has already been done and I am just not aware of it)

But as long as the actual availability of such a system is only theoretical, or the number of applications that can run on such a system is even less than the number of native applications now available for MorphOS2.x, I doubt many people are going to be too interested in such comparisons. (way over my head)

There is a moderator on A.org (Karlos) that likes programming to his PC's GPU, he would be interested in such a comparison.

Quote:

Yes, I'm familiar with Karlos. He sent me a Linux 64bit X86 flash package that works fairly well.

I don't think the parallel architecture of a GPU renders it a good replacement for current CPUs. But as that article pointed out, there are some functions that a GPU can perform many times faster than a CPU.

Personally, I think GPU computing will make a nice addition to CPU programming. Functions that work well on a GPU could be moved there freeing CPU cycles for other tasks.

I love such biased and half documented results...

For example, they don't explain the exact benchmark used, they only speak about a "1k FFT single-precision". If the size of input data is 16-bits and output is 32-bits can make a big difference in results if the input is in fact 32 bits, etc...

Sure, a GPU is always better than a CPU in the computational domain but speed factors annouced shall always to be checked twice (or more)!

And I love also the buzz around GPU: it's quite not new, a GPU is just a DSP where we've removed all analogic/digital convertors

Finally, CPU are much better than GPU in the control domain (like OS)!

I absolutely agree that the CPU is still the essential core of the computer.
In fact, in some code (like code with heavy branching) a CPU would outperform a GPU.
It's going to require some judgment to decide what is appropriate for which processor. Still, with such massive parallel processing being made available cheaply, why not use it? After all, video card GPUs are now getting ridiculously powerful. As I have no interest in Crysis under MorphOS, all those stream processors could be put to good use.