I believe Fruit is a 32 bit program and glaurung on this platform has been reworked to be 32 bit. Rybka is a 64 bit program.
There is no question that 64 bit chess programs take a big hit on 32 bit operating systems. My own program is almost 2X slower on a 32 bit system. I could probably optimize a little of that away if I were motivated, but not a lot as this is a program designed specifically to run well on 64 bit machines. When you develop a program for a specific platform, all bets are off for other platforms. Another issue is running a 10 year old program on modern hardware - not entirely fair because compilers were less capable back when those programs were originally compiled, and even if they weren't, the compilers were designed to optimize instructions for older processors. And the developers themselves tested and optimized for the platform they were developing on, using tricks that worked well on those platforms. I think by itself this can make a big difference. My cilkchess program was designed on a 64 bit alpha architecture and ran well on that. There were occasions when I would run demo games on an intel laptop (this was about a decade ago) and the difference was enormous. I basically blew out the cache on the laptop and took a hit much greater than what one would expect from a normal benchmark of the two platforms. It's easy to imagine that if Vas were hell bent on making a strong Rybka program for the ARM and willing to dedicate a few hundred man hours to the task, there would be a lot of reworking of the code, even if he didn't make a single change to the chess playing heuristics. - Don On Wed, Jun 10, 2009 at 4:15 PM, Ian Osgood <i...@quirkster.com> wrote: > > On Jun 10, 2009, at 12:44 PM, Gian-Carlo Pascutto wrote: > > steve uurtamo wrote: >> >>> But here is someting interesting: In the case of computer >>>> chess it has been estimated that the progress in software >>>> has been roughly the same as the progress in hardware. >>>> Modern chess programs are truly amazing, and not just >>>> a result of faster hardware. There is no reason to think >>>> that this won't be true of computer go. >>>> >>> >>> This makes me wonder... so how slow (and RAM starved) >>> of a computer could you use and still get grandmaster level >>> chess play? >>> >>> In other words, how far back could we go in time if we had >>> today's software and expect a computer to play chess as >>> well as humans? >>> >> >> Assuming something like Rybka 3 is 3100 human-ELO on a 1 x 3Ghz Core 2: >> 3100 - 2500 = 600. >> >> Assuming 70 ELO for a doubling: >> 8.5 doublings >> >> 3Ghz/(2^8.5) = 8Mhz Core2. A Core 2 is a pretty nice CPU, so let's >> assume we lose a factor of 2 with a more ancient design: >> >> 16Mhz ARM or MIPS >> >> Very roughly, maybe an order of magnitude wrong. >> >> -- >> GCP >> > > > We have evidence against going this low: Rybka and several other modern > engines were ported to the dedicated computers Resurrection (203 MHz > StrongArm) and Revelation (500 MHz XScale). Rybka's rating in the SSDF pool > on these platforms are 2497 and 2634, respectively. Fruit 2.3.1 on a > handheld 400MHz Xscale attained 2656 SSDF, and will probably soon be > surpassed by Pocket Fritz 3 (HIARCS) and Glaurung on the same platform. > > Prior to these systems, the strongest dedicated computer was considered to > be the TASC R40, which ran on a 40MHz ARM and attained a rating of about > 2350 in this same pool. > > http://ssdf.bosjo.net/list.htm > > To conclude, it appears that 500 MHz (embedded: poor cache performance) > with little memory for transposition tables is the lowest you can go, while > still staying at grandmaster level. > > Ian > > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ >
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
