Stalemate For Five Years, Then X86 Accelerates Ahead
In late spring of 1994 the Power Mac was in fine shape. The PowerPC 601 consumed less power, than Pentium and generally matched it on integer benchmarks, while easily surpassing it on floating point benchmarks. It achieved all this while running its system bus at half the processor frequency (i.e. from 30 to 40 MHz) compared to the then brutal 60 or 66 MHz bus frequency required by Pentium based systems. However those halcyon days for the Power Mac proved short lived.
The most simplistic indicator of processor performance, and unfortunately the one most often used by personal computer marketers and purchasers alike, is clock speed. The highest clock speed for shipping x86 and PowerPC processors between 1994 to present is shown in Figure 2.
Figure 2. Clock Rate of x86 and PowerPC processors, 1994 to Present
The clock rate lead exchanged hands several times between 1994 and 1998 until x86 started to pull away fast starting in mid-1998. At the present, x86 has fully doubled the best PowerPC clock rate (1 GHz Pentium IIIs and Athlons vs 500 MHz G4). Clock rate isn’t performance but the fact is the two are strongly correlated. This can be seen in the relative integer and floating point performance of x86 and PowerPC MPUs between 1994 and the present as shown in Figures 3 and 4 respectively.
Figure 3. Relative Integer Performance, 1994 to Present
Figure 4. Relative Floating Point Performance, 1994 to Present
The SPEC92 and SPEC95 benchmarks are used to gauge integer and floating point performance of x86 and PowerPC processors for purposes of comparison. Unfortunately, the use of the SPEC benchmark suite to compare the CPU performance of Wintel and Macintosh computers is somewhat problematical for several reasons. First of all it can be argued some of the individual programs in the suite are not representative of the most popular PC applications. Also, SPEC will fail to exploit special capabilities of the processors such as SIMD instruction set extensions like SSE and Altivec that may be quite useful for specific applications. Finally, SPEC will not run natively on the Mac OS.
The PowerPC SPEC figures are mostly taken from IBM Unix-based workstations and servers built around PowerPC processors, and the IBM and Motorola semiconductor division performance estimates for their products. That contrasts with the x86 SPEC figures that are almost exclusively measured on standard PC hardware. This exaggerates the performance of Power Macs because Apple systems tend to use cheaper and lower bandwidth memory systems and smaller caches than IBM’s workstation class products or the optimistic configurations found in PowerPC promotional literature.
Regardless of these problems, the SPEC CPU benchmark suite is the best impartial indicator of CPU performance available. Figure 3 clearly show that while the integer performance lead exchanged hands between x86 and PowerPC several times between 1994 and 1998, the current clock frequency gap has greatly diminished the competitiveness of Apple systems relative to x86 based PCs. The story Figure 4 show is even more compelling. The distinct advantage in floating point PowerPC initially enjoyed over x86 processors has been gradually eaten away to the point where the current clock rate gap has given x86 the clear lead.
Discuss (18 comments)