Soft Cores in a Hard World |
|||
Home Porting lcc >> << Altera, Xilinx Announce
Usenet Postings |
Newsgroups: comp.arch.fpga,comp.arch.embedded,sci.electronics.design Subject: hard FPGA CPU cores do not moot soft cores Date: Mon, 21 Aug 2000 08:31:39 -0700 Peter Alfke wrote > As you may have read, Xilinx will put PowerPCs in the Virtex-II chips. That > might make part of this discussion moot. I most respectfully disagree. A fast hard CPU core will be a welcome development, but it won't moot the utility of slower, more versatile, programmable logic soft cores. For example, a fast hard processor core could prove to be a nice management interface coprocessor for a Virtex-II-based network router building block chip made from an array of 32 or more routing-optimized (*) 100 MHz soft processor cores. For example, the 2M "gate" XCV2000E has 80x120x4 logic cells + 160 block RAMs At 600-800 logic cells + 2-4 block RAMs per CPU (educated guess), this indicates a 32-processor chip-MP with 30% LUTs uncommitted. Just think of what you might do with a 10M "gate" Virtex-II. Soft processor cores will play a huge role in future FPGA SoCs designs. Jan Gray Gray Research LLC www.fpgacpu.org (*) routing-optimized: multithreaded, integrated DMA, special instructions, function units, tables, etc. Even the lowly xr16 core has an integrated 15-channel DMA engine, at a cost of only 16 LUTs to the datapath.
Copyright © 2000, Gray Research LLC. All rights reserved. |