Author: Steven Edwards
Date: 00:21:36 11/19/03
The other day I was able to visit an Apple Store to try out my chess toolkit on a dual 2GHz PPC G5 machine. Running a single thread instance of the toolkit showed that there was a nearly exact linear speed up of a factor of 2.5 when compared to performance on a 800 MHz PPC G4. Note that there was no real testing of 64 bit operations as the code was compiled for 32 bit mode and at this time, the G5 operating system Mac OS 10.3 still presents only a 32 bit space to the user. Only the virtual memory mamngement in the kernel knows about the 64 addressing available in hardware. I am not sure at this point if 64 bit operations can be enabled in user space in the current OS and even if enablement is possible, it wouldn't make much sense as long as the entire system API is still a 32 bit implementation. It looks like we will have to wait until next year and Mac OS 10.4 when I assume that a full 64 bit API and a corresponding tool chain will be in place to support complete porting of current 32 bit applications like the toolkit. This will be a big win for bitboard programs, although separate binaries (or "fat" binaries) will be needed for backward portability. The Apple rumor mill suggests dual 2.4 GHz PPC G5 machines in early 2004 and dual 3 GHz PPC G5 desktops by late summer the same year. My own testing with chess applications shows that PPC hardware has about a 20% performance advantage over x86 machines for a given clock speed based on PPC G4 vs Pentium III experiments. If IBM can keep on their POWER architecture roadmap moving towards 8 GHz chips by the end of the decade, then the days of x86 dominance of the desktop market may be limited.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.