Author: Tord Romstad
Date: 13:59:50 04/27/05
Go up one level in this thread
On April 27, 2005 at 16:30:33, Steven Edwards wrote: >On April 27, 2005 at 12:04:58, Tord Romstad wrote: >>On April 27, 2005 at 11:29:07, John Sidles wrote: > >>>Is this the newest "ultimate" desktop chess computer? >> >>Absolutely not. The G5 is a rather slow processor for chess programs, and for >>integer computations in general (its strong side is floating-point operations, >>which are not important for chess programs). > >To take advantage of a 64 bit architecture, a program has to be recompiled for >it. Merely running unaltered 32 binaries is not going to show that much of an >improvement. For a bitboard program, having a full range of 64 instructions >means a vast performance increase. I compile Crafty (and all other chess programs) as G5-optimized 64-bit exeuctables, of course. This results in a noticable, but not very dramatic speed increase. IIRC, it was somewhere in the range 10-15%. >>My iMac G5 1.8 GHz, for instance, is clearly slower than the old Pentium IV >>2.4 GHz in my office, for all the chess programs I have tried. Not even >>64-bit programs like Crafty are faster on the G5. > >Maybe, but were the programs correctly recompiled for the 32 bit pointer/64 bit >register model? Yes. See above. >For bitboard programs, the NextSq/FirstSq/CountSqs macros (and >related low level functions) should be recoded appropriately. This helps a little, but not much. Even the portable C versions of these functions are rather low on Crafty's (and similar programs') profile graphs. Even if you could reduce the cost of these functions to zero, the speedup wouldn't be very spectacular. And of course, it is possible to use inline assembly language to speed up such functions on the Opteron as well. >Also, the compiler should be told to use G5 instruction scheduling instead of >the default universal mode. Of course. >Of interest are anecdotal reports that the new optimization strategies in gcc >4.0 are giving 20%-30% speed increases in general purpose code on G4/G5 targets. > This is to me the only big reason to buy Mac OSX 10.4 Tiger. I hope this is true (I have already ordered OS X 10.4, and will install it as soon as the Cocoa bridge in OpenMCL works with the new version), but until I see it I remain skeptical. Apple advertised better gcc optimization in previous Mac OS X upgrades as well, but I never noticed any improvement. >>A dual 2.7 GHz G5 is of course much faster than a single 1.8 GHz, but it still >>not >>even remotely close to a dual opteron. > >There are significant differences between the iMac and PowerMac system bus >implementations. That's one of the reasons the PowerMac costs a lot more. Yes, but in my experience bus speed is not very important for the performance of most chess programs. >>The PowerMac G5 is an awesome machine, but if the "ultimate desktop chess >>computer" is what you are looking for, you should look elsewhere. > >If someone is merely interested in running free downloadable binaries (nothing >wrong with that), then I'd suggest an x86 machine, not a Macintosh. But for >developers, it's a different story. I agree entirely. My iMac does everything I need (with the exception of running amateur chess engines) as well as or better than any Linux or Windows machine. I don't care that my chess program runs a little bit slower than it could; it beats me easily anyway. If I were looking for a new computer (desktop or portable) right now, I wouldn't even consider a non-Apple machine. Tord
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.