< Earlier Kibitzing · PAGE 287 OF 287 ·
|Mar-14-07|| ||zarg: <TheGladiator: <zarg: <Of course I know that even 40x increase (100Mhz vs 4Ghz) in clock speed doesn't amount to very much deeper searches on average (due to the exponentially increasing number of nodes to search), but I still think this is a major factor in the improvement of the chess programs.>
The IBM Power2 Super Chip of Deep Blue, was running at low clock speed (135 MHz), but you cannot say a 4 GHz CPU of a different architecture is 40 times faster.>|
I wasn't referring to Deep Blue anymore in the quote above; I was talking about faster CPUs on (basically) the same architecture.>
OK, but context of this discussion, was comparing current desktop CPU with Deep Blue hardware. Anyway, even when comparing different models of the same CPU family, care must be taken, benchmarking is better than looking at just the clock speed.
<Wasn't Pentium 2 (first time around) 166/200/233 Mhz?>
IIRC, the slowest Pentium 2 ran at 233 Mhz, you might be thinking of Pentium Pro. Pentium 133 MHz sound like 1995 hardware. :)
<I think I still would prefer using fritz 8.>
Which year did Fritz 8 arrive? Then I suggest you run Rybka on hardware available at that point.
You might get in trouble running Rybka on a very old Pentium box, because MMX instructions was not available before Pentium Pro (AFAIK). I don't know if Rybka uses such speedup trick though.
<If you don't mind, I'll use two of my old boxes for rybka and run fritz 8 on "modern" hardware. Is your bet still on Rybka?>
I will still bet on Rybka....
IF you run it on the "best" available desktop PC hardware, when Fritz 8 hit the market
Fritz 8 is run on the "best" available desktop PC hardware, when Rybka 1.2 hit the market. :)
|Mar-14-07|| ||slomarko: May i ask again at what time controls will this Fritz-Ryba match be?|
|Mar-14-07|| ||TheGladiator: <TheGladiator: I wasn't referring to Deep Blue anymore in the quote above; I was talking about faster CPUs on (basically) the same architecture.>|
<zarg: OK, but context of this discussion, was comparing current desktop CPU with Deep Blue hardware. Anyway, even when comparing different models of the same CPU family, care must be taken, benchmarking is better than looking at just the clock speed.>
Just for reference, here is what I responded to when I gave that 40x comment:
<zarg: Rybka running on a laptop, has a great chess playing strength too, the main reason for this is not 2-4 GHz CPUs, but the improved algorithms, e.g. better evaluating function, used today.>
And then I suggested the comparison:
<Regarding the quote above, I'm not sure that the most important difference in playing strength between a current Rybka/Fritz 10 on current hardware with 1 GB ram, 2-4Ghz CPU and higher bus bandwidth and e.g. fritz 8 on say a Pentium 133Mhz box w/64Mb ram actually is the heuristic algorithms. Intuitively, based on general knowledge of computer chess and the importance of seeing some plies deeper on average, I think a major part of the increase in playing strength is due to the increase in "brute force elements" of the hardware.
Of course I know that even 40x increase (100Mhz vs 4Ghz) in clock speed doesn't amount to very much deeper searches on average (due to the exponentially increasing number of nodes to search), but I still think this is a major factor in the improvement of the chess programs.>
The 40x referred to the difference between a 133Mhz pentium and a ca. 4 GHz Pentium 4 - I rounded one down and one up, just to get a "pretty" number.
From my previous comments, it should be pretty clear that I know that general benchmarking needs to take more into account than just raw clock frequency (e.g. bus bandwidth, speed of memory, CPU caches, other architectural choices, etc.), games or other graphical applications (like CADs, Mathematica and such) are bound by GPU(s), on-board memory, bus bandwidth and PC RAM, IO-intensive work is mostly bound by stuff like storage-solutions (disks, RAIDs, NAS/SAN, etc) and communication infrastructure, but also by OS; there are for instance huge differences in the time various *nixes need to create, copy and delete files, even when low-level IO-scheduling is tuned for maximum performance for a given purpose. Lately, I've worked professionally with performance issues related to IO on both Windows platforms and Linux, Solaris (SPARC and x86), AIX and HP-UX (both PA-RISC and Itanium).
While you can assume that your particular interests in computing provide for some knowledge that I don't have, I'd appreciate it if you also would assume that I have a satisfactory working knowledge of most issues a senior software engineer with several years of experience in different domains is supposed to cope with.
|Mar-14-07|| ||TheGladiator: <zarg: Which year did Fritz 8 arrive? Then I suggest you run Rybka on hardware available at that point.>|
I hope you understand that I won't purchase any new hardware to make this test - I'll run it on some of the boxes I've got access to, period. My purpose wasn't to check who "work faster", the hardware engineers or the chess engine programmers. It was to check "improved algorithms" versus "faster CPU", inspired by your comment <the main reason for this is not 2-4 GHz CPUs, but the improved algorithms>. Since I can't run a program with worse heuristics (e.g. Fritz 8) on something 10-20 times faster than a 2-3 GHz machine, I had to propose the opposite. How much slower, is a parameter we can adjust (to some degree) - it's not really about "winning a bet".
<<Wasn't Pentium 2 (first time around) 166/200/233 Mhz?>
IIRC, the slowest Pentium 2 ran at 233 Mhz, you might be thinking of Pentium Pro.>
Actually, I was thinking of Pentium MMX, introduced in 1997.
<Pentium 133 MHz sound like 1995 hardware. :)>
And this is indeed the case :)
<because MMX instructions was not available before Pentium Pro (AFAIK). I don't know if Rybka uses such speedup trick though.>
I've got a 166Mhz Pentium MMX in the attic, so it was indeed available before Pentium Pro. I have no idea if Rybka uses MMX instructions, but I thought MMX was mainly geared at improving multimedia performance?
|Mar-14-07|| ||AAAAron: So after Linares, what's the next big super tournament? Will Toppy get his rematch against Drawnik any time soon?|
|Mar-14-07|| ||TheGladiator: <zarg: benchmarking is better than looking at just the clock speed.>|
It just occured to me that you might were thinking of specialized chess engine benchmarks (like fritzmark) here...?
|Mar-14-07|| ||slomarko: <Will Toppy get his rematch against Drawnik any time soon?> i dont think so, Kramnik speciality is... avoiding the matches.|
|Mar-14-07|| ||zarg: <TheGladiator: <zarg: benchmarking is better than looking at just the clock speed.>
It just occured to me that you might were thinking of specialized chess engine benchmarks (like fritzmark) here...?>|
Yes, for example Fritz Chess Bechmark is highly relevant. Some figures are given here:
where we see that P4 2.6 GHz has relative speed of only 1.51 (vs P3 1.0 GHz).
|Mar-14-07|| ||zarg: <TheGladiator: I thought MMX was mainly geared at improving multimedia performance?> |
I'm not up to date with the recent Intel Optimizing manuals, but when MMX came out.. it looked very interesting for general-purpose optimizing of inner loops too, as long as you didn't need to access the floating-point unit.
MMX had 8x 64-bit registers, with very fast instruction set.
|Mar-14-07|| ||zarg: <TheGladiator: I have a satisfactory working knowledge of most issues a senior software engineer with several years of experience in different domains is supposed to cope with.>|
The initial response to <demondays> looked very odd, for someone with expert knowlege. Your advice, on looking up clock speed and number of CPUs of home PC's today, gave the wrong impression.
However, I don't think <demondays> was right, when he said Rybka would at best be equal to Deep Blue. In fact, I would not be surprised if Rybka on a home PC today, smoked Deep Blue anno 97 in a chess match.
The main reason for this, is not the advance in HW, but rather the advance in software.
|Mar-15-07|| ||TheGladiator: <zarg: where we see that P4 2.6 GHz has relative speed of only 1.51 (vs P3 1.0 GHz).>|
Actually, this isn't very surprising and not something one would not anticipate. But one could also say that a 2,5x faster CPU giving a 1,5x higher fritzmark (where I think nodes searched is an important marker) shows that the fritzmark number increases much faster than "playing strength".
<The initial response to <demondays> looked very odd, for someone with expert knowlege. Your advice, on looking up clock speed and number of CPUs of home PC's today, gave the wrong impression.>
I didn't write any technical response to <demondays> - regarding home PCs I thought he'd might be surprised how much faster and more advanced PCs had become since the construction of Deep Blue started. But I probably misunderstood his use of the word "engines", where I thought he might referred to full blown (software) chess engines.
Therefore I told about the custom hardware in Deep Blue (probably exactly what he meant by "engines", coined "chess processors" somewhere) and the most important things it did (for instance move generation and heuristic calculations/algorithms). A point to me, was that these engines weren't complete chess engines, but rather helper chips (like math coprocessors) where the most repeated operations in a chess engine were programmed in hardware, not software.
But (hopefully) unlike a PC chess program that goes to market, the central software in Deep Blue wasn't complete or tested enough at the time of the match, so that they actually during the match (between games) had to work on correcting pure technical software bugs which was discovered during play. So Kasparov kind of played alpha software which in some cases even prevented "normal operation" of Deep Blue - a fact that I also found worth mentioning. Of course, Kasparov thought they "adjusted Deep Blue to fit his style of play" ;) Without improving heuristics, just by removing software bugs to make it work like intended, Deep Blue might have been even stronger than it was.
<The main reason for [Rybka beating Deep Blue], is not the advance in HW, but rather the advance in software.>
I still would like to do a comparison like the one I mentioned - increasing CPU speed by 2,5 times is one thing, increasing it by 20-30 times (together with other advances in hardware) sure amount to something, particularily in computer vs computer matches where a difference in horizon of 3-5 plies might actually mean something (and more than in a match against a top human player, where sometimes the human understanding (which heuristics try to immitate) is superior to the computer's basic brute force approach).
So, how much "faster" does a computer need to be to make up for the advance in heuristics between ca. 2001 (fritz 8) and ca. 2006 (the first commercial Rybka)? This amounted to about 200 rating points in the computer rating list I referred to.
|Mar-15-07|| ||zarg: <TheGladiator: I still would like to do a comparison like the one I mentioned - increasing CPU speed by 2,5 times is one thing, increasing it by 20-30 times (together with other advances in hardware) sure amount to something,>|
I would love it, if you took the time to run some tests on this! I fully agree that we talk about a very significant HW speedup.
IMO, if using Pentium 2 (166 MHz) and a Pentium 4 (3 GHz) as the test machines, the Fritz 8 vs Rybka match is not as interesting as Fritz 5 vs Rybka match would be.
My gut feeling is that Fritz 8 would have no problems winning, while we would see a more close fight if Fritz 5 is used.
If Fritz/Rybka has been highly tuned, a severe penalty my result when running the engine on a very new/old CPU. The cost-benefit of using memory has changed quite a lot, while numbers of transistors follow Moore’s law, the RAM speed increase only ca. 5% per year. This affect the design of hash tables and the importance of avoiding cache misses.
|Mar-15-07|| ||Ragh: Fritz vs Rybka matchup is being played by some users here on CG. Check out the The Kibitzer's Café.|
<Chessgames.com Members vs Machines Invitational> <Round 3, Saturday March 17, 13:00 EST (17:00 GMT)>
|Mar-15-07|| ||AAAAron: So what's the next big tournament, and who's going to be in it? <Anyone>?|
|Apr-13-07|| ||alphastrike20: <fromoort> He took the fide world title. BUT NOT THE CLASSICAL TITLE. HE DID NOT BEAT KASPAROV. WHO WAS UNDOUBTABLY NUMBER ONE.|
|Apr-13-07|| ||fromoort: <alphastrike20>ok.|
|Jun-26-07|| ||square dance: <Regarding "strong/weak" Linares, I had to go all the way back to 2006 to find something nearly as weak as the 2007 tournament...> this whole argument is a strawman because i never argued that linares 2007 had a weak lineup. what ive been arguing this whole time is that the first and third favorites finished last and you continually fail to demonstrate where that has ever happened before. quit changing the subject. your ratings based arguments are irrelevant. is ivanchuk definitively stronger now than he was when the april rating list came out? would adding ivanchuk to a tournament now instead of in march make the tournament stronger? did topalov and kramnik clearly get worse from april to july as demonstrated by their loss of three elo points?
<In 12 games, Kasparov only managed to win one game, against Vallejo, in a field where Kasparov outrated everyone but Kramnik with 100-170 points,> i dont care if kasparov underperformed by X number of points; he finished tied for 2nd, not tied for last! i dont even see how you think you're making a point for your argument. hmmm, +1 and finished =2nd, but did much worse than was statistically expected based on his previous tournament and match results, or -2 and =last. also did much worse than expected based on his previous match/tournament results, but within the normal range. i dont know. i would probably think the +1, =2nd score was a better showing for the worlds #1 player. |
<Was it excellent of Kasparov to score half a point more than 16-17 year old Radjabov rated 175 points below him?> no. who said it was? what could this possibly have to do with anything?
<I'd say 2004 was about as below par for Kasparov as 2007 was for Topalov.> you can say this all you want, but since he also scored +1 in linares 2003 i'd say that his 2004 score wasnt such a shocker.
|Jun-26-07|| ||acirce: <<Akavall: in that tournament they were weaker players than they normally are.>|
But by the same logic particularly Magnus and Morozevich, but also Svidler and Anand, were STRONGER players than they normally are, "making up" for Topalov and Leko "being weaker" than normally.>
This makes no sense and is totally illogical. Why on earth would they become stronger players "by the same logic"? Participant A and B were out of form so participant C and D must be stronger than they usually are? Bizarre. Of course there is nothing stopping the entire field from being weaker than usual.
|Mar-16-08|| ||TigerG: Where is Radjabov?|
|Mar-16-08|| ||whiteshark: At this very moment ?|
|Mar-16-08|| ||walker: I don't see any messages?!|
|Mar-16-08|| ||Augalv: neither do I!|
|Dec-28-10|| ||Penguincw: This tournament has so many drawn games.|
|May-15-11|| ||csmathReloaded: Thanks to God kramnik was not invited, otherwise he would have gave us a lots of short draws.|
|Nov-26-11|| ||Penguincw: 38/56 of the games are drawn. That's 67.9%. However, it is a high level of play.|
< Earlier Kibitzing · PAGE 287 OF 287 ·