Is it Virtex? Or is it HyperTransport?

The clever reader could pen a parody on the above title to the tune of Cheryl Wheeler’s “Is it peace, or is it Prozac?”, but I don’t have time for such folly. The question of the title was raised because of an announcement July 28 from Commex Technologies Ltd. of Tel Aviv, who claimed a throughput speed of 27 Gbits/sec from a network interface card utilizing Xilinx Virtex-5 FPGAs with HyperTransport connections.  (It’s important to stress that Commex is aiming at NICs and classification ICs, not at data-center switches or routers, as its end product.)

Of course, I really don’t think it’s an either/or at all. FPGA developers have been boasting about the speed of internal library cells, as well as the advantages of efficient serial transceivers and serdes devices. The most interesting part of this speed equation to examine may entail a close look at when parallel traces are used, and when serial I/O is employed, for designs utilized in the highest-speed data centers and metro telecom devices. We may not have heard a lot about HyperTransport of late, but it remains an important inter-chip and intra-chip interconnect to allow for milestones similar to those claimed by Commex.

Expect to see many alternatives remaining in the market – Aurora, Interlaken, PCI Express, RapidIO, etc. And expect to see FPGAs ready to take on the challenges of 40-Gbit and 100-Gbit Ethernet. We simply have to decide on the multiplexing of the interface, and groups like Optical Internetworking Forum are providing plenty of choices for system designers.