AMD aims data-transfer technology at industry standard

The company aims to increase the performance of PCs with an internal connection that increases the speed and amount of data that get ferried between chips.

John G. Spooner Staff Writer, CNET News.com
John Spooner
covers the PC market, chips and automotive technology.
John G. Spooner
4 min read
Advanced Micro Devices aims to increase the performance of PCs and other computing devices with an internal connection that increases the both the speed and amount of data that get ferried between chips.

The new connection, dubbed HyperTransport, will be licensed royalty free to manufacturers and, so far, the concept is gaining interest. Cisco Systems, Sun Microsystems, Microsoft and others are looking at it, according to AMD.

Routing data between chips inside computing devices is full of hitches because many peripheral devices must share a single input-output connection. AMD aims to offer each device its own connection with HyperTransport and make that connection faster.

AMD won't only offer HyperTransport to PC makers. The chipmaker intends to offer the technology to companies that manufacture networking hardware, set-top boxes, handheld computers and even game consoles. So far, AMD has signed up several big names, including Sun and Cisco.

HyperTransport was originally developed as a new input-output method for multiprocessor servers based on AMD chips. However, the company now believes HyperTransport has a much broader appeal because of its ability to free up performance bottlenecks inside PCs, servers and networking equipment by establishing a wider pathway for data to move between chips.

AMD's plan is to spread the technology as widely as possible throughout the industry, establishing it as a standard.

Should it become wildly successful, HyperTransport could open new doors for AMD's chip business. Not only will the chips offer built-in compatibility with the technology, but AMD's work to establish HyperTransport as a standard will expose it to a number of potential new chip customers.

IDC analyst Roger Kay cautioned it might take some time to establish HyperTransport.

"Even though there's a lot of positive buzz about this...it's still going to be an uphill battle against entrenched architectures," he said Wednesday. Nonetheless, AMD has an opportunity to parlay its partnerships with Cisco and Sun into a major play, he added.

"I think it's a big win for AMD if it gets to place some processors in some infrastructure equipment" such as Cisco routers or Sun servers, Kay said. Such a move, Kay added, would give AMD "some credibility with enterprise customers," which might help the company in its long-running battle to get corporations to purchase computers that use AMD chips.

If accepted as a standard, HyperTransport could eventually replace the PCI (Peripheral Component Interconnect), the current method of sharing data between chips inside PCs.

"We designed (HyperTransport). And when we looked at what we had, the basic conclusions were that it had broader applications and that its basic adoption would create new opportunities for AMD," said Chris Neuts, manager of "technology evangelism" for AMD.

To that end, AMD has established the HyperTransport Consortium and has begun freely licensing the technology.

The consortium will work to establish the HyperTransport as an industry and to tackle specific issues related to the technology using working groups.

In addition to Cisco and Sun, AMD has signed up ATI Technologies, Broadcom and Nvidia to use the technology. API Networks, which helped develop HyperTransport, is also working on a series of components that will help companies build devices based on the new input-output method.

"The mind share we're getting from working with partners of this quality has been invaluable to us," Neuts said.

AMD says that the first implementation of HyperTransport will offer a peak data rate of 6.4GB per second. Additionally, the technology is up to 24 times faster than the current PCI bus. PCI offers data rates of 266MB per second. HyperTransport was also designed to be compatible with current operating systems and applications.

"None of the operating system or driver software has to be changed. It's really a way to get more bandwidth and reduce latency by using the same programming model," said Ruth Shelley, vice president of engineering at API.

However, not all of the products based on the first implementation of HyperTransport will offer 6.4GB per second data rates initially. Follow-on implementations of the current HyperTransport technology will offer even higher data rates, AMD said.

API's hardware products will, for example, offer 2GB per second of total peak bandwidth. HyperTransport uses separate inbound and outbound paths, which means API's products will support 1GB per second each for inbound and outbound paths. By contrast, PCI uses a shared connection, meaning that multiple devices must share its bandwidth.

AMD says that HyperTransport products should begin shipping in about the third quarter of this year.

HyperTransport is "seeing a tremendous amount of interest," Shelley said. "I think you'll see products later this year or early next year."

AMD, for its part, will use HyperTransport technology as the interconnect technology in systems based on its forthcoming Hammer family of 64-bit processors for desktops and servers.