Communications of the ACM
And Then There Were None: A Stall-Free Real-Time Garbage Collector For Reconfigurable Hardware
Credit: City of Greenwood, S.C.
The end of frequency scaling has driven architects and developers to parallelism in search of performance. However, general-purpose MIMD parallelism can be inefficient and power-hungry, with power rapidly becoming the limiting factor. This has led the search for performance to non-traditional chip architectures like GPUs and other more radical architectures. The most radical computing platform of all is reconfigurable hardware, in the form of Field-Programmable Gate Arrays (FPGAs).
FPGAs are now available with over one million programmable logic cells and 8MB of on-chip "Block RAM," providing a massive amount of bit-level parallelism combined with single-cycle access to on-chip memory. Furthermore, because that memory is distributed over the chip in distinct 36Kb units, the potential internal bandwidth is very high.
No entries found