Meet the Xeon: Inside the Mac Pro's processor
Although Intel has been using the Xeon name since 1998, the dual-core Xeon 5100 processors that power the Mac Pro and Xserve lines are entirely new chips, based on Intel’s 64-bit Core architecture platform. This platform was initially introduced to the public late in 2005, with chips continuing to roll out throughout this year; it includes the Core Duo and Core Solo processors used in the Mac Mini and the iMac, as well as some other chips that haven’t shown up in Macs—at least not yet.
Code-named “Woodcrest," the Xeon 5100 debuted at the end of June, and was designed to offer top-speed computational throughput with better power efficiency than Intel’s previous high-end processors. The company terms it a “server” chip, although most hardware vendors, Apple included, will use it in professional systems like the Mac Pro. That’s due to the fact that the chip was designed to excel at the processing of huge amounts of data, like those found in real-world applications like movie production with Final Cut Pro, or image editing in applications like Aperture or Adobe Photoshop (once Photoshop is Intel-native on OS X).
The Xeon 5100 powering the Mac Pro comes in three speeds: 2.0GHz, 2.67GHz, and 3.0GHz. Like the Core Duo and IBM’s PowerPC 970MP used in the current Power Mac G5, the 5100 has two microprocessor cores built into each chip. Both cores run at the same rated speed, with a 1.33GHz frontside bus (which connects the processor to the rest of the system) and share a 4MB Level 2 cache, which helps keep the processors humming during compute-intensive tasks.
Unlike the cache in the Core Duo and PowerPC 970MP, either processor core can utilize the entire cache if necessary, which gives a performance boost in crucial data-processing tasks, especially with legacy, non-multithreaded applications that aren’t designed to take advantage of multiple processors.
One other reason, aside from performance, that the Xeon 5100 was the perfect chip to use in Apple’s flagship Mac was because it is the only Core chip that can currently be used in a dual-processor configuration similar to the Power Mac Quad G5. The 5100 can only be used in single- or dual-processor configurations, so we won’t see any Xserves with four Xeons in it, although Intel is expected to announce a quad-core successor to the 5100 series some time in 2007.
How much better than a G5?
In synthetic tests, Apple claims that the 3GHz Mac Pro, with its twin Xeon processors, offers more than twice the integer performance of the previous top of the line, the 2.5GHz Power Mac Quad G5, and 1.6 times the floating point performance. Of course, real-world benchmarks will tell the true story of the Mac Pro as it compares to its predecessor, but the Xeon 5100 should have a distinct performance edge over the older PowerPC chip, as we would expect in any more modern chip design with a faster front-side bus and larger and more flexible cache. (Each core in the PowerPC 970MP chip has 1MB of dedicated Level 2 cache.)
But, with the Xeon 5100, Intel claims that it has eliminated one huge performance advantage held by the PowerPC architecture: vector processing. Known to Mac users as AltiVec, or the Velocity Engine, this technology increased vector-based processing significantly on the PowerPC machines, and was one of the reasons that applications like Photoshop were able to manipulate such large image files with ease, even when other parts of the Mac subsystem weren’t as fast as comparable Intel-based PCs.
In the Core architecture, Intel has a feature called Advanced Digital Media Boost,. That may not roll off the tongue as cleanly as AltiVec, but Intel claims that it achieves the same end. For Mac users, the important thing to note about Advanced Digital Media Boost is that it executes 128-bit vector-based instructions in one clock cycle, instead of the two clock cycles taken by previous Intel designs. This theoretically doubles the performance of vector operations, and brings Intel to parity with AltiVec. It also should provide enhanced performance of native graphics applications, especially when working with large amounts of data. (If you would like a much more in-depth (i.e. geekier) explanation, check out ArsTechnica’s excellent analysis of the Core architecture, written by Jon Hannibal Stokes.)
Intel and IBM don’t use the same metrics or nomenclature when discussing power efficiency, so it’s hard to compare the true power efficiency of the Xeon 5100 series with the Power Mac G5’s PowerPC 970MP processor. The fact that Apple has been able to eliminate so much of the Power Mac’s cooling apparatus says quite a bit about the power requirements of the 5100, however. Apple told us that the Mac Pro system as a whole only pulls 980 watts, versus 1,000 watts for the Power Mac Quad G5, and a lot of that power is routed to the PCI Express bus.
The 5100 is only the beginning
The last two years have seen chip manufacturers push away from the “speeds and feeds” mentality of the previous decade, largely a result of the fact that, as chips have increased in speed, their power consumption has increased as significantly. The result is that companies like Intel, Advanced Micro Devices, and IBM have moved to multi-core designs that don’t offer huge jumps in raw performance speed (as measured in gigahertz), instead providing tangible increases in performance (through multiple processor cores on a single chop) without linear increases in power consumption.
One of the great things for Mac users is that the Xeon 5100 used in today’s Mac Pro is only the beginning of a line of high-performance chips that will be released in the next year—as noted, we should see quad-core chips, as well as chips better tuned to run in portable systems, and chips that offer notable increases in performance for low-end desktops.
[ Rick LePage is Macworld ’s editor-at-large. ]