Early tests question 733MHz Power Mac peformance
G4 faster than G4 Plus, just as PIII faster than P4?
Is the 733MHz Power Mac G4 no better than the 533MHz version? Some benchmarks posted on the Web this week suggest it isn't, others that it should be. And while Apple remains silent on the issue, confusion reigns supreme.
The claim emerged at the weekend when benchmarks run on a new 733MHz Power Mac and what was said to be a single-processor 533MHz machine were posted on xlr8yourmac.com's Web site, apparently copied from a posting on an Apple support forum.
A mix of tests, the benchmarks showed the 733MHz box beating the 533MHz part on just a couple of them: MacBench 5.0's CPU test and AltiVec/Velocity Engine fractal generator.
We say 'said to be a single-processor' because the 533MHz box's system bandwidth rating was nearly double the one-chip 733MHz Mac's rating, which leads us to wonder if it's really a dual-processor box that the tests were run on.
Certainly, you might well expect a dual-processor box to run faster that a single-CPU machine on code designed to take advantage of the second chip, even with a 200MHz difference in clock speed.
That's certainly what the second set of benchmarks suggest, which do compare a dual-processor 533MHz box with the 733MHz Mac.
The 733MHz machine is based on Motorola's PowerPC 7450, aka the G4 Plus, rather than the 7410 used in the 533MHz system. At the same clock speed, we'd expect the 7450 to be a little slower than the 7410 thanks to the design trade-offs Motorola made to get the chip to run at clock speeds above 600MHz.
Motorola added more steps to the 7450 instruction processing pipeline to keep the chip busy at higher clock speeds - with fewer stages, at high clock speeds, the processor spends to much time idle, so processing efficiency tails off. Adding stages allows the chip to process more instructions at any one time. That's at the cost of design simplicity and efficiency, but upping the clock speed and improving the chip's various instruction and data caches, nicely counters that.
So while the original PowerPC 7400 - the G4 - has a four-stage pipeline, the 7450 has seven stages. In contrast, the Pentium 4 has a 20-stage pipeline to get it past the 1GHz mark.
And now we have 733MHz benchmarks from MacInTouch. Tested against 500MHz dual and single-processor box, the new Power Mac is a clear winner. Now, we trust MacInTouch's methodology rather more than any of the other benchmarks, since we know these guys know how to test Macs properly and methodically. Given the gap between the 500MHz machines and the 733MHz, we'd expect to see the latter running faster than 533MHz boxes too.
If not, there are still issues relating to support for the 7450 and its three-level cache implementation in the benchmark test code which may be obscuring the chip's true level of performance, but since no ones' applications will have yet been compiled with the 7450 in mind, perhaps that's just as well.
The initial feedback certainly suggests that the 733MHz isn't as fast as it could be, particularly when running existing applications. It's a similar problem that Intel is facing with the P4, which in some cases runs more slowly than the older Pentium III. Intel blames that on un-optimised benchmark code, but was still forced to admit that productivity apps will show little if any improvement when running on a P4 rather than a PIII.
All this will change with future revisions of application code, but that won't come as much consolation to anyone wondering whether to buy a 733MHz box or a dual-533MHz machine. MacOS X, with its better support for multiple CPUs than MacOS 9 offers, will confuse matters further, though again users will still have to wait for revised apps - in this case, for Carbon support - to see the benefit. ®