Nodes per second

[Don]

I'm preparing some slides for a presentation and I need historical data on how many nodes per second the typical programs were achieving over the past 4 decades or so.

I know some of you have some of the old constellations and other machines - can you tell me how many nodes per second these programs/machines are reporting?

I also wonder if there is historical data on Crafty running on various hardware over the years. And what did Bell do, as well as the old Northwestern program chess 4.7 (I think it was called) when it was dominate. Knowing the year would be good too if anyone has that information.

[bob]

I'm preparing some slides for a presentation and I need historical data on how many nodes per second the typical programs were achieving over the past 4 decades or so.

I know some of you have some of the old constellations and other machines - can you tell me how many nodes per second these programs/machines are reporting?

I also wonder if there is historical data on Crafty running on various hardware over the years. And what did Bell do, as well as the old Northwestern program chess 4.7 (I think it was called) when it was dominate. Knowing the year would be good too if anyone has that information.

Here are some numbers.

Northwestern on the Cyber 176 was doing 2600 nodes per second, circa 1976.

final version of Belle was doing around 160K between 80 and 84 or so.

Cray Blitz was doing about 10K nps in 1983 when we won the WCCC, and about 160K in 1986. Our fastest ever speed, circa 1994, was on a 32 CPU Cray T90 where we peaked at around 7M nps.

Deep thought was all over the map, ending at their peak NPS of 1B (average of 100M or so) in 1997.

If you go back to 1976, "blitz" was doing 100 nodes per second on a 1.3MIPS computer. Around 70-72 we were doing 1-10 nps depending on the version and the hardware.

Crafty started in 1994 (Dec) and was hitting around 30K nps by the time the pentium 133mhz processor was out. The pentium pro took me to 70K, and then a quad pentium pro 200 hit around 300K or so. We broke 1M on the quad xeon when it was released, but I am not sure when that was released. My box started at 4 x 400mhz and ended up at 4 x 700 with later versions of the original xeon.

[Don]

I'm preparing some slides for a presentation and I need historical data on how many nodes per second the typical programs were achieving over the past 4 decades or so.

I know some of you have some of the old constellations and other machines - can you tell me how many nodes per second these programs/machines are reporting?

I also wonder if there is historical data on Crafty running on various hardware over the years. And what did Bell do, as well as the old Northwestern program chess 4.7 (I think it was called) when it was dominate. Knowing the year would be good too if anyone has that information.

Here are some numbers.

Northwestern on the Cyber 176 was doing 2600 nodes per second, circa 1976.

final version of Belle was doing around 160K between 80 and 84 or so.

Cray Blitz was doing about 10K nps in 1983 when we won the WCCC, and about 160K in 1986. Our fastest ever speed, circa 1994, was on a 32 CPU Cray T90 where we peaked at around 7M nps.

Deep thought was all over the map, ending at their peak NPS of 1B (average of 100M or so) in 1997.

If you go back to 1976, "blitz" was doing 100 nodes per second on a 1.3MIPS computer. Around 70-72 we were doing 1-10 nps depending on the version and the hardware.

Crafty started in 1994 (Dec) and was hitting around 30K nps by the time the pentium 133mhz processor was out. The pentium pro took me to 70K, and then a quad pentium pro 200 hit around 300K or so. We broke 1M on the quad xeon when it was released, but I am not sure when that was released. My box started at 4 x 400mhz and ended up at 4 x 700 with later versions of the original xeon.

That will be immensely useful.

I am also interested in the very earliest sargon programs, such as sargon on the z80 Jupiter and sargon 2 for the TRS-80 and so on. Does anyone have any data on that? I can probably figure out the year.

Other interesting data points would be for the various versions of Fritz over the years and other good commercial programs as well as chess challenger 7, 9 and so on. I think most of the 6502 program running at 4mgz were within a factor of 2 or so of each other and I only need approximate numbers.

[JVMerlino]

Other interesting data points would be for the various versions of Fritz over the years and other good commercial programs as well as chess challenger 7, 9 and so on. I think most of the 6502 program running at 4mgz were within a factor of 2 or so of each other and I only need approximate numbers.

IIRC, The King (Chessmaster) was a relatively slow searcher, nps-wise, and when the US dev team was laid off in November 2002 it was getting around 100-150K on my moderately slow work machine. I'll see if I can verify from some old e-mails I might still have available.

jm

[mhull]

I'm preparing some slides for a presentation and I need historical data on how many nodes per second the typical programs were achieving over the past 4 decades or so.

I know some of you have some of the old constellations and other machines - can you tell me how many nodes per second these programs/machines are reporting?

I also wonder if there is historical data on Crafty running on various hardware over the years. And what did Bell do, as well as the old Northwestern program chess 4.7 (I think it was called) when it was dominate. Knowing the year would be good too if anyone has that information.

Here are some numbers.

Northwestern on the Cyber 176 was doing 2600 nodes per second, circa 1976.

final version of Belle was doing around 160K between 80 and 84 or so.

Cray Blitz was doing about 10K nps in 1983 when we won the WCCC, and about 160K in 1986. Our fastest ever speed, circa 1994, was on a 32 CPU Cray T90 where we peaked at around 7M nps.

Deep thought was all over the map, ending at their peak NPS of 1B (average of 100M or so) in 1997.

If you go back to 1976, "blitz" was doing 100 nodes per second on a 1.3MIPS computer. Around 70-72 we were doing 1-10 nps depending on the version and the hardware.

Crafty started in 1994 (Dec) and was hitting around 30K nps by the time the pentium 133mhz processor was out. The pentium pro took me to 70K, and then a quad pentium pro 200 hit around 300K or so. We broke 1M on the quad xeon when it was released, but I am not sure when that was released. My box started at 4 x 400mhz and ended up at 4 x 700 with later versions of the original xeon.

That will be immensely useful.

I am also interested in the very earliest sargon programs, such as sargon on the z80 Jupiter and sargon 2 for the TRS-80 and so on. Does anyone have any data on that? I can probably figure out the year.

Other interesting data points would be for the various versions of Fritz over the years and other good commercial programs as well as chess challenger 7, 9 and so on. I think most of the 6502 program running at 4mgz were within a factor of 2 or so of each other and I only need approximate numbers.

Steve B can give you actual numbers from period dedicated units. IIRC, the Fidelity 68000 Excel (Spracklen's) averaged about 2k nps on a 12 MHz MC68000 chip. That machine was circa 1988. I used to have a Champion Voice Challenger (zorched now) but I don't remember it reporting an NPS. It used a 6502 at 3 MHz, circa 1983.

[jhaglund]

These might help?

http://immortalchess.net/forum/showthread.php?t=5863

http://mlngveld.home.xs4all.nl/craftybe ... hhash.html

http://www.jens-hartmann.at/Fritzmarks/

[Aser Huerga]

In 1970, was held in New York the first Computer Chess North American Tournament organized by ACM (Association for Computer Machinery). Chess 3.0, based on Shannon's B strategy, and written by three Northwestern University students (David Slate, Larry Atkin and Keith Gorlen), stand out on one's own. There were some Chess version in the next nine years, being the last Chess 4.8 (Chess 4.7 played against IM David Levy [1.5-3.5], and had an estimated ELO of 2.200).

Chess 4.8, running on Cyber 176 (Control Data Corporation) reached 18 million instructions per second, almost 4.000 positions per second.

Source: Computer Schach (1980) by Ludek Pachman (badly translated by me, sorry ... hope it helps)

[Don]

In 1970, was held in New York the first Computer Chess North American Tournament organized by ACM (Association for Computer Machinery). Chess 3.0, based on Shannon's B strategy, and written by three Northwestern University students (David Slate, Larry Atkin and Keith Gorlen), stand out on one's own. There were some Chess version in the next nine years, being the last Chess 4.8 (Chess 4.7 played against IM David Levy [1.5-3.5], and had an estimated ELO of 2.200).

Chess 4.8, running on Cyber 176 (Control Data Corporation) reached 18 million instructions per second, almost 4.000 positions per second.

Source: Computer Schach (1980) by Ludek Pachman (badly translated by me, sorry ... hope it helps)

Good stuff! Thanks.

[Rémi Coulom]

Hi Don,

You'll find a lot of data on page 12 of this document:
http://archive.computerhistory.org/proj ... 028.sm.pdf

An otherwise fun read, with caricatures of all participants

Rémi

[wgarvin]

According to its chessprogramming wiki page, in 1995 Darkthought was hitting 60,000 nps on a 175Mhz Dec Alpha 21064, but I'm not sure where that number comes from...

This page by Heinz suggests it was more like 20 K to 40 K on a 275Mhz Alpha, in 1995-1996. The same page has more info:

The fourth and now current DARKTHOUGHT version is an evolution of its predecessor featuring lots of speed optimizations as well as dramatic changes in the bitboard move-generator and various fundamental data structures (e.g. new chess-board representation). On a 500MHz DEC Alpha-21164a, the DARKTHOUGHT version as of July 31, 1997 routinely reaches 200K nps in the middlegame while peaking at over 650K nps in the endgame (see Section 1.2.8).

Edit: Also, from this chessbase article from 2002:

In February 1996 Garry Kasparov beat the computer Deep Blue in Philidalphia. Deep Blue ran on an IBM SP/2 server with 25 special micro-channel cards, each of which had eight special chess processors. These did all of the brute calculations, allowing the system to generate and evaluate a cool 100 million positions per second. Deep Blue was six hundred times faster than the fastest microcomputer programs – at least with regard to positions per second (at the time Fritz running on a 150 MHz Pentium was searching at 170,000 nodes per second).