13

What was the first chess engine that could beat the world chess champion when running on a standard desktop playing at standard speeds (i.e. not blitz chess)? For concreteness, say a $1000 PC.

ABCD
  • 22,419
  • 2
  • 43
  • 82
Simd
  • 369
  • 2
  • 9

2 Answers2

15

Deep Blue was a super computer. In the 2006 match, Kramnik was defeated by Deep Fritz that everybody could buy.

In a November 2006 match between Deep Fritz and world chess champion Vladimir Kramnik, the program ran on a computer system containing a dual-core Intel Xeon 5160 CPU, capable of evaluating only 8 million positions per second, but searching to an average depth of 17 to 18 plies in the middlegame thanks to heuristics; it won 4–2.[31][32]

(source)

ABCD
  • 22,419
  • 2
  • 43
  • 82
  • Thanks. Do you know what hardware Deep Fritz ran on and what the time controls were? – Simd Feb 03 '19 at 22:32
  • 4
    The comparison between a 1997 supercomputer and a 2006 standard machine isn't necessarily trivial. – Inertial Ignorance Feb 03 '19 at 23:20
  • 4
    @InertialIgnorance I don't know but I am sure you couldn't afford IBM Deep Blue. – ABCD Feb 03 '19 at 23:23
  • 2
    True, but I'm talking about the comparison between computational speeds. A mobile phone today is much faster than a supercomputer in the 50's, even though the latter costed way more at the time. – Inertial Ignorance Feb 03 '19 at 23:25
  • @Anush EDITED. Please review. – ABCD Feb 03 '19 at 23:28
  • 1
    @InertialIgnorance I don't know to be honest. Wikipedia stated Deep Fritz could run 8 million positions per second. What was the speed for Deep Blue? – ABCD Feb 03 '19 at 23:29
  • Personally i don't know as well, but to me it makes sense that a 1995 super computer is more or less equivalent to a 2006 mid-tier/high-tier comercial computer. Hardware does evolve pretty fast – Isac Feb 04 '19 at 10:22
  • @Isac The interesting detail is that in the last few years or so desktop PCs have not got much faster. Moore's law more or less ended. – Simd Feb 04 '19 at 10:30
  • 5
    @InertialIgnorance A mobile phone today would have made the supercomputer TOP 500 list in the mid 1990s, nevermind the 1950s! – J... Feb 04 '19 at 13:05
  • 2
    For reference, Deep Blue was evaluating 100 million (first version) to 200 million (updated) positions per second, but to a depth of only 6-8moves on average (to a max of 20 in some cases). Deep Fritz had better heuristics, allowing it deeper searches with fewer evaluations. – J... Feb 04 '19 at 13:13
  • I guess desktop CPUs stopped getting faster around 2010 maybe? So the speed of a 2006 desktop machine is not so dramatically different to a 2019 one as compared to 1996 versus 2006. – Simd Feb 04 '19 at 19:56
  • @J... I know, I was using an exaggerated example. – Inertial Ignorance Feb 04 '19 at 20:42
  • @Anush with >4 physical cores no longer a limit for mainstream desktops, growth for multi-threaded loads is resuming at a decent rate. (Although other than web browsers with multiple very resource hungry tabs open it's debatable if there are any commonish consumer workloads that really benefit from them but not being run on a GPU). – Dan Is Fiddling By Firelight Feb 04 '19 at 20:43
  • @Anush Desktop CPUs definitely did not stop getting faster. In 2010 the top shelf i7 was the [i7 970](https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-970+%40+3.20GHz&id=840), right now the direct comparison to that chip (6-core, 12-thread) is the [i7 8700K](https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-8700K+%40+3.70GHz&id=3098). The single and multi thread performance of the 8700K is double that of the 970. Certainly not the 8-12x improvement seen over the same span previously, but still progressing. – J... Feb 04 '19 at 21:18
  • @J... You are right of course, CPUs are just getting faster much more slowly. Interestingly the AMD 8350 from 2012 seems to be by far the fastest per dollar according to that chart. – Simd Feb 05 '19 at 05:49
  • @DanNeely This is a little off topic but there are also physical limits to the number of cores you can usefully have in a shared memory machine where all cores are considered to be identical. – Simd Feb 05 '19 at 07:10
8

A standard desktop today is significantly more powerful than whatever machine Deep Blue was running on in the mid-1990s against Kasparov. Since Deep Blue was the first engine to beat a world champion, that's the answer to your question.

Note that there may have been an engine before Deep Blue that, if it ran on a modern day desktop, could have beat Kasparov. But we never saw such a match happen so it's just speculation to say any earlier engine than Deep Blue.

Inertial Ignorance
  • 19,788
  • 22
  • 68
  • 5
    Thanks for this. It seems according to the wiki that Deep Blue was running at 11.38 GFLOPS which is roughly the speed of a cheap PC these days. However it's not 100% clear Deep Blue was better than Kasparov. The match was controversial. – Simd Feb 03 '19 at 20:01
  • 8
    Note peak FLOPS and achieved FLOPS are very different things. And I'm not sure that FLOPS is actually a reasonable measure of performance in this case - Ian the HPC guy – Ian Bush Feb 03 '19 at 20:30
  • 1
    Yes, the match was controversial. If you don't accept that Deep Blue was superior though, you could select the next engine that beat a world champion in a match (there's a list in the wikipedia page on "Computer Chess"). – Inertial Ignorance Feb 03 '19 at 21:35
  • 1
    I guess that’s more or less my question. – Simd Feb 03 '19 at 22:29
  • 8
    Deep Blue used significant amounts of custom hardware. It is not the answer to the question because it cannot "run on a standard desktop." – David Richerby Feb 04 '19 at 14:41
  • 3
    Due to the specific hardware used by Deep Blue I am not certain that the assumption about powerful holds. A 10 year old GPU can probably still outperform a modern CPU when rendering graphics. – Thorbjørn Ravn Andersen Feb 04 '19 at 16:08
  • Deep blue design, hardware and software, described by Feng-hsiung Hsu the architect and chief designer of Deep Blue: http://www.csis.pace.edu/~ctappert/dps/pdf/ai-chess-deep.pdf. "We based the Deep Blue system on an IBM RS/6000 SP supercomputer, which you could view as a collection of IBM RS/6000 processors or workstations connected through a high-speed switching network." – bishop Feb 04 '19 at 19:04
  • Full version of Hsu's book _Behind Deep Blue_: https://books.google.com/books?id=zV0W4729UqkC&lpg=PP1&dq=isbn%3A0691090653&pg=PA9#v=onepage&q&f=false – bishop Feb 04 '19 at 19:15