In February, I wrote that "several NPU vendors will endorse HyperTransport by the end of this year." Oops, my bad. But HyperTransport continues to make headway as a chip-to-chip interconnect for networking ICs.
Early-adopter Broadcom has been sampling its HyperTransport-based 1250 central-processing unit for several months and will have a production part in the first quarter. At that time, PMC-Sierra will begin sampling its HyperTransport-based RM9000x2 processor. MIPS CPU vendor Sandcraft is also developing HyperTransport parts.
The Broadcom and PMC processors are the most popular today for networking control-plane sockets. So far, they have had few HyperTransport chips to connect to, but that is changing.
Instead of network processor vendors jumping on the HyperTransport bandwagon, initial adopters have been vendors of networking-support silicon.
Several startups are building HyperTransport interfaces into their classification coprocessors, traffic managers and TCP offload engines. Unfortunately, those vendors have not yet announced their products.
A key factor behind this design activity is Cisco Systems. Though the networking giant has publicly endorsed both HyperTransport and its rival, RapidIO, Cisco is putting its muscle behind HyperTransport. As a result, many startups eager to gain design wins at Cisco are adding HyperTransport interfaces to their products.
RapidIO, on the other hand, has been slower to market. Motorola has announced plans to add RapidIO to its PowerPC line in a product due to sample in the second half. It will add RapidIO to its NPU line, but not until 2003. IBM will also support RapidIO, but hasn't said when.
We recently polled the 2,000 subscribers of our e-mail newsletter, NPU Wire, about which bus would be most successful in networking applications. HyperTransport was the clear winner with 45 percent of the vote, followed by RapidIO with 29 percent and Arapahoe (3GIO) with 15 percent. Internet polls are not the most reliable devices, but this response concurs with my own research.
RapidIO backers claim technical superiority, and I can't disagree. But the advantages, such as supporting thousands of devices, are not very relevant when trying to connect a few chips on a line card. With Motorola's backing, RapidIO appears to be making inroads into communications equipment. Arapahoe will thrive in PCs. For networking equipment, HyperTransport is emerging as the best choice.
Linley gwennap is founder and principal analyst of the linley group (www.linleygroup.com) in Mountain View, Calif.
RSS
RSS| 
Digital| 
Mobile
