Under the Desktop: All PowerPC to the People

CreativePro Staff

Performance enhancement is often on the minds of content creators, whether from application and system software, or in hardware, such as the graphics display card, storage devices, or memory configuration. Each component shares the load, which can either improve or impair the overall speed of your system. Considering the compute-intensive applications and large file sizes common to creative work, system performance is critical to the daily workflow of most designers and producers.

Of course, a significant part of the performance picture is the processor, so I looked forward to the processor industry’s annual powwow, In-Stat/MDR’s annual conference, Microprocessor Forum, held in San Jose, Calif. There, several manufacturers discussed the future gains in performance of chips destined for high-end graphics workstations. Still, the operative word here is "future," which in this case means at least a year or even longer.

Known in the trade simply as MPF, the conference often features the introduction of new chips for all types of applications, from mainframe computers to $300 PCs, as well as the low-power processors found in consumer digital devices, such as phones and handheld devices. In addition to great up-to-the-minute information, attendees receive an eBay-worthy industry collectable: a portfolio with actual processors embedded in the cover (see figure 1).

For content creators on the Macintosh platform there was big news indeed: IBM introduced the PowerPC 970, a chip that appears destined for the next-generation Apple servers and Power Macintosh desktops. The reason for the equivocation is that IBM and Apple are completely mum on the subject. No surprise there. Despite the official silence, it’s clear that this processor is slated for a future Mac.

The announcement is perhaps the most important piece of processor news since the introduction of the PowerPC G3 in 1997 and has fueled plenty of speculation in the Mac community over the 970’s origins and performance potential.

64-bit Detour
Before moving into the specific products, we would do well to briefly discuss 64-bit processing, since the technology is a source of some hubbub.

All of today’s desktop computers use a 32-bit processor, whether Mac or Windows. The 32-bits refer to length of the chip’s addressing and instructions. These processors can handle a maximum of about 4 gigabytes of memory at a time (4,294,967,296 bytes, but who’s counting?). There are ways of extending this limit, but it’s tough. Yet, most of us rarely consider 32-bit addressing much of a drawback, since until recently, the cost of 4 GB of RAM was astronomical, and our operating systems and applications are designed to run within these constraints — we just accept them. But as mentioned in previous columns, content-creation applications can certainly use more RAM and speed.

However, some high-end servers and workstations used in the CAD (computer-aided design) engineering and scientific visualization market run 64-bit processors, which can handle much more RAM and larger data sets. The 64-bit addressing can tackle very large operations, such as rendering each pixel of a huge image, or by mapping lots of memory to hard drives in a virtual memory scheme.

According to chip gurus, a 64-bit processor could theoretically manipulate up to 16 exabytes of data. However, ignoring the pie-in-the-sky specifications, these workstations can easily use 8 GB or more RAM in real-world applications.

These workstations and their associated processor architectures are available from a variety of vendors, including HP (PA-RISC), IBM (POWER), SGI (MIPS), SUN Microsystems (SPARC); Intel and HP teamed up to create the Itanium processor, a.k.a the IA-64, distinct from the IA-32 family that includes the 32-bit Pentium 4 and Xeon processors. Most of these machines use their own homegrown 64-bit-savvy OS (some version of Unix) and applications must be coded specifically to support the 64-bit capabilities.

Mac My Day?
The PPC970 is significantly different than its forebears. Rather than an evolution of the current lines of PowerPC architecture, it’s a 64-bit processor based on IBM’s POWER architecture used in the company’s Unix workstations and servers. The 970 is not on the old PowerPC design roadmaps (see figure 2), still, it can run current 32-bit applications.

The processor also combines a SIMD (single instruction, multiple data) vector engine for accelerating multimedia. From the description, it sounds suspiciously like the Velocity Engine found in the current PowerPC G4.

One way that chip vendors increase frequency (the chip’s internal speed) is by breaking down a job into sets of smaller ones, thereby increasing the number of steps needed to do a task. This ordered series of steps is called a pipeline. This sounds as if it would be less efficient, however, since the chip marches to the strict beat of a clock, the more little things that can be done in a single cycle sometimes the better. Intel has done this to great effect in the Pentium 4.

The 970 will run at a faster clock rate than the current PowerPC G4 and IBM’s specification mentioned 1.4-GHz and 1.8-GHz versions. In addition, these initial speeds could be increased as the chip enters production in the next year. The clock speed will always lag behind the Pentium 4, which currently stands at 3 GHz.

"The 970 makes absolutely perfect sense — the only question is why did they wait so long?" asked Keith Diefendorff, the former architect of the Velocity Engine at Apple, and now vice president of product strategy at MIPS Technologies. He said that while Apple has done an excellent job of reducing frequency as an issue in the marketplace, the speed of the PowerPC needed to pick up the pace.

More megahertz doesn’t automatically mean greater performance when comparing different chip architectures. For example, the PowerPC’s RISC architecture does more with each step, and while running at a slower clock rate when compared with the Pentium 4, it can provide equal or better performance. As Diefendorff mentioned, Apple (and now even PC-chip vendors such as Advanced Micro Devices) point to this fact as the "Megahertz Myth."

Some tasks benefit from higher integer speeds, while others can take better advantage floating point operations or multiple processors. Vendors prefer to compare performance with artificial benchmark tests, even though the results may be better or worse when the chip is working in a real computer, with a real application and in your particular workflow.

In his presentation, Peter Sandon, IBM Senior architect, provided a couple of figures for the 970 using the Standard Performance Evaluation Corp. 2000 (SPEC2000) tests for integer and floating point performance. He said a 1.8-GHz 970 rated a 937 and 1,051 for the integer and floating-point tests, respectively; in comparison, according to InStat/MDR, a 2.5-MHz Intel Xeon currently holds 893 and 878 ratings, respectively.

The Velocity Engine can provide even more power to the floating point performance. "The 970 is a floating-point monster," said Peter Glaskowsky, editor in chief of the Microprocessor Report.

In addition, the processor features a very, very fast 900-MHz bus to move data on and off the chip, which is useful for multiprocessor configurations and large RAM caches. It also will have the "elastic" ability to wait a number of cycles for data to arrive — a necessary feature since most current system buses run way slower than 900-MHz (the current Power Macintosh G4 has a 167-MHz system bus).

It’s difficult to know whether the first flavors of 970 will be used in a PowerBook. Analysts at the conference expected the chip first in Apple’s Xserve server and in Power Mac desktops. The current G4 chip runs about 30 Watts when really crunching hard on data. My reading of the 970’s spec sheet showed the 1.8-GHz 970 will be in the mid-40-Watt range. However, it also includes the current frequency reduction routines to reduce power consumption and heat. The analysts expected later versions of the 970 to be designated for notebooks.

1 2 Next