I see the difference as one of "how it's connected".. If we look back at things like the Floating Point Systems boxes that did fast FFTs hanging off a PDP-11, that's a box where you send a bunch of data, kick it to go, and wait for the results. I see that as an accelerator. (and perhaps a modern GPU fits in that model).
I see a coprocessor as being more tightly coupled. Take as an example the Floating Point Unit. Often integrated on chip, in the past it was an optional widget that worked sort of in parallel with the main CPU. That is, you'd have instructions like "FMUL".. if you had the FPU hooked up, the main CPU stopped, the FPU did the calculation, and the main CPU resumed. If you didn't have the FPU hooked up, you got an instruction trap, and some software Floating point routine emulated the effect of the FPU. That is, the coprocessor is "synchronous" with the main CPU in some way. (even if you can parallelize the pipelines, for instance). Often, for instance, the FPU can "see" the registers of the main processor, even if it implements separate floating point registers. Likewise,the main CPU can "see" the FPU registers (and woe to the person who forgets to store and restore those FP registers in an ISR that changes the state of those registers) Jim Lux From: beowulf-boun...@beowulf.org [mailto:beowulf-boun...@beowulf.org] On Behalf Of amjad ali Sent: Sunday, March 10, 2013 9:36 AM To: Beowulf Mailing List Subject: [Beowulf] difference between accelerators and co-processors Hello All, Is there any line/point to make distinction between accelerators and co-processors (that are used in conjunction with the primary CPU to boost up the performance)? or these terms can be used interchangeably? Specifically, the word "accelerator" is used commonly with GPU. On the other hand the word "co-processors" is used commonly with Xeon Phi. Thank you. Regards, Amjad
_______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
