Hyper transport lives on

Electronic News, Nov 19, 2001 by Tom Murphy

So what has happened to HyperTransport, the chip-to-chip interconnect phenomenon that lost out to 3GIO as the next I/O standard for PCs? Well, the technology is making headway, getting implemented in silicon and forging some endorsements from industry heavyweights.

While HyperTransport is not likely to be implemented in the same volume of chips and systems that 3G10 stands to be once it becomes a standard, perhaps it still can play an important role as system designers increasingly seek ways to drive more bandwidth out of the box.

Just ask Gabriele Sartori, president of the HyperTransport Consortium, which jus recently announced the addition of Silicon Graphics Inc. as one of the technology's supporters. That list appears to be growing by one company per week, Sartori said, as other companies begin to see the value of an interconnect technology which, Sartori said, can scale up to 12.8Gbytes/sec. The technology its also getting recognition for its low latency, a factor that becomes more important in communications boxes, Sartori said.

Probably the greatest market potential for HyperTransport is in routers, switches and hubs, said Shane Rau, an analyst for IDC in Framingham, Mass.

"I think it will have a place," Rau said. "The best thing they have done with HyperTransport was to form the consortium to get support and organize their efforts."

HyperTransport's big win to date is on the board with Nvidia Inc.'s Nforce chipset, which allows an Advanced Micro Devices Inc. (AMD) Athlon processor to link with double data rate SDRAM memory. HyperTransport also gained prominence at the Microprocessor Forum as AMD, the company which fostered HyperTransport, announced that the technology would be used in the upcoming Hammer series of microprocessors as an on-chip interconnect.

But the biggest move for HyperTransport was getting endorsements outside the microprocessor community, Sartori said. AMD, which started development of the technology in 1998, had credibility in the PC world but not in the communications industry.

Soon, though, communications heavyweights such as Broadcom, PMC-Sierra, API Networks, Xilinx and Altera joined the HyperTransport Consortium after recognizing the technology's potential, Sartori said. HyperTransport may wind up as the technology used in communication system backplanes, which link different cards and blades together inside a box and allow them to transfer information.

Sartori said that consortium supporters are equally divided among computation and communications camps. The next step in establishing HyperTransport as a technology that won't go away is to get third-party intellectual property suppliers to recognize it as a standard. As soon as that happens, HyperTransport could be driven into the foundry industry as fabless ASIC designers look to select a controller technology for interconnect.

Sartori said that members that chose to join the consortium will have free access to the interconnect standard as long as they implement it in a conventional chip-to-chip fashion. A more embedded use could incur royalty fees, such as AMD's use of HyperTransport on the Hammer microprocessors.

HyperTransport is even showing some promise in server platforms, Sartori said. Companies are looking at ways to build RISC microprocessor-based systems at PC costs. HyperTransport is providing some companies with that potential.