The HyperTransport technology will make possible faster and less expensive computers that offer better overall performance, Gabriele Sartori, president of the HyperTransport Technology Consortium, told MacCentral.
“The PC architecture, its fundamental design, is pretty simple — and old, going back to 1982,” he said. “For these reasons many applications, such as those involving video editing, aren’t as fluid as they should be. Apple did better work on their systems than others, I think, since they’re a company with a vertical line of products. But it’s time to overcome some of the bottlenecks.”
HyperTransport interconnect technology, originally invented by AMD (Advanced Micro Devices), is a new high-speed, high-performance, point-to-point link for integrated circuits, developed to enable the chips inside of high-performance compute devices, networking and communications devices to communicate with each other faster than with existing technologies. The technology’s bandwidth of 12.8GB/sec purportedly offers up to a 48-fold increase in data throughput, compared with existing system interconnects that typically provides bandwidth up to 266MB/sec.
HyperTransport isn’t designed to replace technologies such as FireWire and USB, but to supplement them and handle chores that they’re not designed to handle. It’s a way to combine some network/shared device concepts, such as FireWire, with a faster serial port concept known as Low Voltage Differential Signaling (LVDS), and apply these design philosophies to a computer’s main bus. A standardized design for all the buses makes it less expensive to add more devices on a bus, Sartori said. Plus, you don’t have different types of buses to figure out and debug.
The technology, will purportedly enable the chips inside of high-performance computer devices, and networking and communications devices — such as those that power the Internet — to communicate with each other up to 24 times faster compared with existing technologies. HyperTransport complements externally visible bus standards like the Peripheral Component Interconnect (PCI), as well as emerging technologies like InfiniBand and 10GB Ethernet. And one difference between HyperTransport and FireWire/USB is an obvious physical one.
“The HyperTransport architecture is done on a board, inside a computer, while FireWire and USB are more external technologies,” Sartori said. “It’s designed to provide a smooth transition from existing technologies and to support the necessary scalability and bandwidth for future product innovations. It can deal with different bandwidths at different costs to cover the high end to the low end.”
Although initially developed for high-performance personal computer and server platforms, the technology is also gaining momentum in networking and communications devices, embedded applications and other non-PC devices. Multiple products integrating support for HyperTransport technology are in development to support desktop and notebook PCs, workstations and servers, and Internet communication devices.
For instance, graphics chipmaker Nvidia is incorporating HyperTransport data bus technology into its nForce platform processing architecture. The Nvidia nForce platform processing architecture combines the company’s graphics with the industry’s first implementation of the new Dolby Digital Interactive Content Encoder, a technology built in the MCP that encodes multichannel audio into Dolby Digital 5.1 in real-time.
“Nvidia will put into the market this year a very nice chip set with amazing performance and based on HyperTransport,” Sartori said.
However, there’s a perception that HyperTransport was beaten by Intel. A key industry group has approved an Intel technology to overhaul the innards of personal computers and overcome the PCI (Peripheral Components Interconnect) bottleneck, according to a ZDNet article (http://news.cnet.com/news/0-1003-200-6771562.html?tag=mn_hd).
Increasing PCI speed is becoming “prohibitively expensive,” and engineers have been searching for an alternative that will let computers keep pace with ever-faster central processing units, ZDNet reports. Rival chipmakers Intel and AMD each had designs that could do the trick, with HyperTransport further along than Intel’s, the story said. However, the industry group in charge of PCI leaned in favor of Intel’s technology, code-named Arapahoe, sometimes called 3GIO, and which will probably eventually be known as PCI 3.0.
Arapahoe could be used to speed the transfer of data to not only network cards but graphics systems as well, and to the subsystems that must funnel the huge amounts of data associated with digital images or video, Bala Cadambi, third-generation interconnect program manager at Intel’s Desktop Platform Group, told ZDNet. The goal is to make sure there’s a specification ready when products are ready by the second half of 2003, Michelle Leyden Li, platform initiatives manager in Intel’s Desktop Platform Group, added.
Today’s PCI transfers data across 32 or 64 parallel wires; Arapahoe will use fewer, higher-speed lines, and data won’t have to be synchronized across the collection, ZDNet said. AMD has positioned its HyperTransport technology as a way for one chip to communicate with another — for example, a CPU to communicate with a memory controller or a PCI subsystem.
Sartori said that Arapahoe and HyperTransport are complimentary technologies, and Intel understands this. Though analysts had said HyperTransport had the potential to split the industry, AMD didn’t position HyperTransport as a 3GIO competitor. AMD didn’t elaborate on why they chose not to do so.
The work on HyperTransport started at the end of 1997 and design work started in early 1998. Sartori said that it’s a solid technology that will be around for a “long, long time.”
Recently, a coalition of high-tech industry leaders — including Apple, Nvidia and Sun — announced the formation of the HyperTransport Technology Consortium, a nonprofit corporation that supports the future development and adoption of the HyperTransport I/O Link specification.
“Everybody has been looking for a solution to the bandwidth problem, so there’s interest from a lot of different parties,” Sartori said. “People need a pipe to carry data from the memory controller to the audio-video subsystem, and we provided the pipe.”