Debunking Hyper(Transport) Rumors

Macintosh rumor sites have cause quite a stir lately, having convinced most of us that flat panel iMacs and other whiz-bang products would be ready for Macworld New York. Unfortunately, what was predicted had little to do with reality.

MacOS Rumors has decided to move on from flat panel iMacs to hyping HyperTransport. Apple's having announced its involvement in the consortium has new rumors spreading like wildfire throughout the Mac community. However, most of the information being disseminated is factually incorrect. The folks at MacOS Rumors clearly do not understand HyperTransport, as evidenced in this smart sounding, but completely untrue statement:

Thusly, unless Apple manages to get a more consumer oriented processor ready for HyperTransport (perhaps the 7460 - the power-saving, lower cost version of today's high end G4), we will probably never see a Mac built on today's nForce chipsets.

It is helpful to understand a little bit about HyperTransport to understand why the aforementioned statement is misleading and shows a poor technical understanding of the technology. HyperTransport can be used to connect almost any two components and can be chained one device to the next. There are two derivations of HyperTransport at the moment, and there will most likely be many more in the future. Right now, one specification exists for connecting CPUs to systems and to each other, and another for general I/O connection. In the future one could perhaps expect a HyperTransport replacement for the AGP port - the specifications of HyperTransport are that flexible.

MacOS Rumors is implying two things in their article. First, it implies that Apple will or may use HyperTransport for its CPU interconnect technology. Second, it implies that Mac users will have to wait for an entirely new processor (7460 or later) to enjoy HyperTransport on the Mac platform. Both of these implications are false. There is no evidence that Apple is planning on using HyperTransport to connect CPUs together. This may be on the horizon many, many years from now, but it is not terribly likely. Apple already has a relatively efficient multiprocessor bus.

The second assertion, that Mac users will have to wait for a new CPU to utilize HyperTransport, is also factually incorrect. Lets apply some basic logic. The Nvidia nForce chipset connects to Athlons and Pentiums, which were designed five or more years ago. Why would Apple need to redesign the G4 in order to connect it to the nForce or any other HyperTransport chipset? The bottom line is they don't. The nForce uses HyperTransport between the Northbridge and Southbridge, not the CPU and the Northbridge. The Northbridge chip has the processor bus (GTL for the Pentium, EV6 for the Athlon), an AGP Bus, a Memory controller and an 800 Mbps HyperTransport link to the Southbridge (ATA-Controller, etc.) There is no reason why the nForce couldn't be easily adapted to the PowerPC bus. The fact that the nForce uses HyperTransport has absolutely nothing to do with incompatibility with the Mac platform.

If Apple isn't going to use it to connect processors together, why is Apple interested in HyperTransport? Two words: Digital Hub. HyperTransport allows you to chain PCI busses together for greater I/O bandwidth. Right now the PowerMac has a theoretical maximum I/O bandwidth of 264 MBps (megabytes per second) and sustains about 215 MBps under a peak load.

Think about all of the components in a PowerMac G4. Gigabit ethernet peaks at 128 MBps, almost half the peak bus bandwidth. FireWire peaks at 50 MBps and is soon to double to 100 MBps (FireWire and and Ethernet are normally rated in megabits per second, these figures are divided by 8 to get megabytes).

ATA-66 peaks at 66 MBps, and ATA-100 peaks at 100 MBps. If you add these together, you already get very close to the 264 MBps theoretical maximum of 64-bit PCI. When the PowerMac moves to faster disk and FireWire standards, the entire bus could be saturated.

HyperTransport would let Apple put all of the built in components on their own bus and leave plenty of bandwidth to spare. For example, a Northbridge with an 800 MBps HyperTransport bus could have three PCI 64 busses chained to it (264 MBps x 3). With Apple touting its hardware as a digital hub, and its encroaching on SGI in the professional graphics arena, I/O bandwidth is a paramount concern.

In order for Apple to grow, so must its I/O architecture. Apple controls the design of its Northbridge, so it is much more likely to appear there first. With HyperTransport, Apple is buying itself a bit of insurance for the future. HDTV and other extremely bandwidth intensive applications are right around the corner.

I don't purport to know exactly how Apple is using HyperTransport or whether they plan on using Nvidia nForce chipsets, but I think this is a far likelier scenario than the one presented by the technically ill-informed rumor sites.

Chris Lozaga is a technical writer and has documented software for the IBM SP super computer and the AIX Operating System. He is no longer an IBM employee; this article represents his opinion and his opinion only. It is in no way indicative of the views of his employers, past or present.