Apples Response Apple questioned the CPU focus of the tests and said that critics need to look at the overall performance enhancements to the new systems, including its faster system bus (167MHz, compared with 133MHz in earlier models); twin ATA hard-drive controllers, one ATA/100 and one ATA/66; its direct PCI bus; and the ATI Radeon 9000 Pro graphics card included in the high-end models.In addition, Boger said, the Bare Feats benchmarks dont reflect the DDR Macs potential performance increases in bandwidth for direct memory access, which lets system elements send and receive data directly from main memory without going through the CPUs. Boger said the new systems represent a better price/performance ratio than the January model ($2,499 vs. $2,999) and urged users to look to the 1.25GHz Power Mac before drawing conclusions about how far the platform has advanced, speedwise, in the intervening months. "For a real comparison, theyll have to wait until the dual-1.25 comes out," he said. Dennis De Mars, a Mac OS X shareware developer from Fractal Domains, also contests Bare Feats claims. De Mars said that most of the tests are CPU-intensive, and that improvements in bandwidth from DDR memory to CPU wouldnt necessarily improve these benchmarks dramatically. "For these tasks, much more time is spent in CPU processing than in reading or writing memory," he says. "That isnt to say that the DDR implementation Apple is using isnt advantageous in general system usage, it is just to say that these benchmarks will not reveal those advantages." De Mars also notes that independent of the Power Macs DDR improvements, the new machines have a 25 percent faster system bus. The faster bus didnt demonstrate any improvements in Bare Feats tests, which he sees as proof that the tests arent measuring effective RAM bandwidth. Nick dePlume is the editor in chief of Think Secret.
Tom Boger, Apple director of Power Mac product marketing, pointed out that in the Bare Feats benchmarks, the DDR model includes 1GB of L3 cache, compared with 2GB in the SDRAM model. He also said that the tests place most of the stress on the CPU, not the performance of the system as a whole: "Theyre spending most of their time in the processor(s)," he said. "Since the systems are running at the same clock speed, its not too surprising there isnt more of a gap."