Arduous Victory against Chess Player 2150

This forum is for general discussions and questions, including Collectors Corner and anything to do with Computer chess.

Moderators: Harvey Williamson, Steve B, Watchman

Forum rules
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
Martin Hertz
Member
Posts: 76
Joined: Wed Oct 07, 2015 7:08 pm

Post by Martin Hertz »

spacious_mind wrote:Thanks and interesting. Depth 3 it should have found maybe at 30 seconds per move. But I also see you are at 7.1 MHz. 68,000 at 8 Mhz is the equivalent of 6502 with 4 MHz. You took if I read it correctly 22 seconds. I suspect 30 seconds on the Atari is not enough time to find it. I suspect that the 80186 has similar speed as a 6502 therefore I suspect if I tried it again Atari ST would have probably needed around 40 seconds to find this move.

Depth 4 maybe not enough time it took you 11+ minutes to find it. Whittington's programs were always slow searchers.
Yes, the CP2150 is one of the slowest searchers indeed, but the 80186 is speedwise similar to the 8086 and should not be
faster than a 68000 with the same clock speed, probably slower, because of the better suited instruction set of the 68000
for chess programming. I've found a speed benchmark of the old CPUs using the n-queens problem thats very near to chess
for comparison: http://www.hpmuseum.org/cgi-sys/cgiwrap ... i?read=700

I think only a test on a real Atari will show the actual speed difference.

The typical search depths of some other programs at 30sec/mov at midgame on the HP-1000CX for showing how slow the CP2150 is:

CP2150 2 plies (7 plies on the i7 thats about 5000 times faster)

REX Chess 4 plies

Fritz1 5 plies

Fritz2 5 plies
User avatar
spacious_mind
Senior Member
Posts: 4018
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Martin Hertz wrote: Yes, the CP2150 is one of the slowest searchers indeed, but the 80186 is speedwise similar to the 8086 and should not be
faster than a 68000 with the same clock speed, probably slower, because of the better suited instruction set of the 68000
for chess programming. I've found a speed benchmark of the old CPUs using the n-queens problem thats very near to chess
for comparison: http://www.hpmuseum.org/cgi-sys/cgiwrap ... i?read=700

I think only a test on a real Atari will show the actual speed difference.

The typical search depths of some other programs at 30sec/mov at midgame on the HP-1000CX for showing how slow the CP2150 is:

CP2150 2 plies (7 plies on the i7 thats about 5000 times faster)

REX Chess 4 plies

Fritz1 5 plies

Fritz2 5 plies
Hi Martin,

I have a couple of Atari St. Don't ask me why, maybe I am nuts, but I enjoy chess computers and programs of all kinds on old Home computers and DOS and I hate to be stuck by not being able to play alone 2 chess programs of the same system.

When I get back to playing with Atari and Commodore I will test CP2150 again. As I really think that both systems are slower than your 80186 and a lot slower than 68,000 chess programs on dedicated chess computers.

To confirm this suspicion, I did some research around the internet. For the Atari ST there is very little regarding speed comparisons but there are several references on the internet about the ST 8 MHz = 1 MIPS.

Amiga has the same problem there are references of it being 0.7 MIPS for the Amiga 2000=same as 500 and also as low as 0.54 MIPS for it's 7.16 MHz system as you can see from this chart:

Code: Select all

System...........OS.......CPU...(MHz)...MIPS(v1.1) 
 -----------.------------.------.-----.------ 
 ATT.PC6300+.MSDOS.3.1....80286...6,0....0,81 
 IBM.PC/AT...VENIX/86.2.1.80286...7,5....0,75 
 IBM.PC/AT...PCDOS.3.1....80286...6,0....0,71 
 IBM.PC/AT...PCDOS.3.0....80286...6,0....0,69 
 ATT.PC6300+.MSDOS.3.1....80286...6,0....0,69 
 IBM PC/AT...PCDOS 3.0....80286...6,0....0,62 
 IBM PC/AT...PCDOS 3.0....80286...6,0....0,61 
 IBM PC/AT...VENIX/86 2.1 80286...6,0....0,57 
 Amiga 1000..AmigaDOS.....68000...7,16...0,54 
 Macintosh...Mac.Rom......68000...7,8....0,52 
 Fast Mac....------------.68000...7,7....0,51 
 IBM PC/AT...PCDOS 3.0....80286...6,0....0,48 
 ATT PC6300..MSDOS 2.11...8086....8,0....0,44 
 NEC PC9801F.PCDOS 2.11...8086....8,0....0,44
Since I really think that there is not that much difference between and Atari ST and Amiga I think the truth probably lies somewhere between 0.54 MIPS and 1 MIPS for the Atari ST.

Now this is important as a reference as in all cases your 80186 is faster. 80186 is quoted as 1 MIPS for 6 MHz. Below is a chart showing all the scenarios I researched and projected it forward on how much time it would take to find the correct 3 ply move which your 80186 found in 22 seconds.

Code: Select all

PROCESSOR			MHZ			MIPS			COMPMHZ			TOTALMIPS	FACTOR		SECONDS
AMIGA 500			1.000			0.0754			7.16			0.54		0.41			54
AMIGA 500			1.000			0.0978			7.16			0.70		0.53			42
ATARI ST 68000		1.000			0.1250			8.00			1.00		0.76			29
INTEL 80186			1.000			0.1670			7.91			1.32		1.00			22
Therefore I do really think that both Atari ST and Amiga 500 with their Motorola 68000 processor are slower than the 68000 chips in the Mephisto Computers. The reason being is that the former are generic computers whereas the latter is optimized just to play chess. There is lots of reference of Amiga performance equaling an Intel 8086 at 5 MHz.

From this research I created also this table below:

Image

Best regards
Nick
User avatar
spacious_mind
Senior Member
Posts: 4018
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Fernando wrote:Yes, there is such previous musing about general features of the position, what we should do, where are the weaknesses of both sides, the imbalances, etc. At last you pick a thread and then begin the tactical search. Clearly, we do not do that as the comp. That previous philosophical part if just ours. And it is in this part where the professional player has advantage due to his experience. We have some, bit not that much...

Fern
Hey Fernando,

I think I have a way that would work in seeing how much strength of a Grandmaster belongs to hard study, intuition and experience. And how much strength is based on search during a match.

I think it is really quite easy to establish.

If I took Magnus Carlsen for example. I could take 5 games from a Blitz tournament he played this year and rate them to get his average rating where he has almost no time to think.

Then I could take 5 games from his Worldchampionship and rate them as well.

Using the same methodology I use for the rating tests. The difference between the two final ratings would be his strength improvement from taking time to study and search into the positions and almost not thinking at all.

Easy eh!

Just like computers you could then figure out his improvement if any for doubling of time (doubling of speed on computers) :P

If it were to turn out that the strength improvement is zero, then of course there would be an argument for why bother to play long games at all if you are not any better by playing them :P

Best regards
Nick
Martin Hertz
Member
Posts: 76
Joined: Wed Oct 07, 2015 7:08 pm

Post by Martin Hertz »

Hello Nick,

I've checked your MIPS comparison, but have to say that some results are not realistic if you
run the same chess programm on the different CPUs. The problem is that the MIPS of a CPU is not
that good for measuring the speed of a real problem.

The typical instruction clock cycles for a real 8-bit assembly written program on the Z80, 68000
and the 8086/80186 is roughly the same. A 0 wait state 80286, 80386, 68020, 68030 and the 6502 are
about 3-5 times faster and a 80486 or SH-3 risc is around 8 times faster. The Pentium is only slightly
faster than a 80486. You need at least a Pentium M to be twice as fast as the 80486 at the same clock
speed.

I'm not sure, maybe a chess program on the Amiga or Atari cannot use 100% of the CPU time compared
to the dedicated computers. A good test of a dedicated vs Atari and the HP-1000CX is Psion Chess on
a Test position, because we can be sure that all versions are assembly coded programs and not compiled.
It would be also interessing to see, if there is any difference in speed factor between Psion and CP2150.

I will try to find a good test position for Psion.
User avatar
spacious_mind
Senior Member
Posts: 4018
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Martin Hertz wrote:Hello Nick,

I've checked your MIPS comparison, but have to say that some results are not realistic if you
run the same chess programm on the different CPUs. The problem is that the MIPS of a CPU is not
that good for measuring the speed of a real problem.

The typical instruction clock cycles for a real 8-bit assembly written program on the Z80, 68000
and the 8086/80186 is roughly the same. A 0 wait state 80286, 80386, 68020, 68030 and the 6502 are
about 3-5 times faster and a 80486 or SH-3 risc is around 8 times faster. The Pentium is only slightly
faster than a 80486. You need at least a Pentium M to be twice as fast as the 80486 at the same clock
speed.

I'm not sure, maybe a chess program on the Amiga or Atari cannot use 100% of the CPU time compared
to the dedicated computers. A good test of a dedicated vs Atari and the HP-1000CX is Psion Chess on
a Test position, because we can be sure that all versions are assembly coded programs and not compiled.
It would be also interessing to see, if there is any difference in speed factor between Psion and CP2150.

I will try to find a good test position for Psion.
I agree with you that MIPS are not optimal but it is one of the few things that are available for comparing across different hardware. Notice I also deliberately showed Amiga 1200 which is a 68020 with a different MIPS.

PSION is partially fine as it is available on the ST but not on Amiga, but there are at least 2 versions of Psion so we would have to check that. Sargon III also works but Sargon III is disappointing on the 68,000 Amiga and Atari, not as strong as expected when compared to what it already does on C64, Apple II and Atari 400/800. But then again that may be exactly for the same reason. 68,000 is just not as strong on Atari and Amiga.

Colossus X is weak compared to Colossus IV so probably not perfect either. And Colossus IV is not available on ST or Amiga or PC, which is a shame as it would have been perfect since you have it on Sinclair Spectrum, C64, Apple II, Amstrad and BBC.

CP2150, CP2175 and Chess Simulator are also available for both PC and Amiga and ST but not on 6502's and Z80's.

Best cross checking software in my opinion might even be Sargon III since you even have that on a C64, Apple II, Atari 400/800, Amiga, ST as well as PC.

ps... the lowest speed I can check on a PC is a 286-10 Mhz. But then again I think my Tandy 286-12 MHz laptop has a turbo button I think (can't remember) which if switched off I think runs at 6 MHz.

Best regards
Nick
User avatar
spacious_mind
Senior Member
Posts: 4018
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Hi Martin,

I have been digging some more about MIPS, from what I can gather an 8 Cycle instruction on 8MHz CPU would be calculated as 8,000,000 / 8 = 1,000,000 instructions per second. = 1 MPS

Which is what I see quoted for the Atari ST M68,000 8 MHz.

Other pieces of information that I found show the following:

"The fastest instruction on the ST takes 4 clocks. It's possible to hit 2 mips using 4 cycle instructions. The Blitter on the ST does steal cycles from the CPU.

50% on an async transfer or 90 - 100% on a synchronous transfer. 90% using the restart method or less if there are interrupts during a transfer."

If I read that correctly and accept that 1 MIPS for the Atari ST is about right then with whatever else the CPU does in the background I can easily see that programs as CP2150 are playing at less than 1 MIPS.

In the chart that I had posted yesterday with 1 MHz 6502 being 2.966 times faster than a Z80, this is also pretty much correct.

Here is a different way to look at the chart:

Image


Below is a rule of thumb that has typically been used for dedicated computer comparisons historically, where it is assumed that the programs are configured optimally on dedicated computers.
6502 5 MHz 1,6 MIPS
68020 20 MHz 4 MIPS
68030 33 MHz 11,9 MIPS
68030 50MHz 18 MIPS
68040 40MHz 44 MIPS
68060 66MHz 94 MIPS
68060 77MHz 110 MIPS
Dedicated computer people consider 68,000 as having half the MIPS of a 68020 therefore a Mephisto 12 MHz 68,000 computer would have approx. 1.2 MIPS which would make it slower than a 6502 5 MHz. But most people also say 5 MHz 6502 dedicated computer equals a 12 Mhz 68,000. Therefore even using the above dedicated chart an Atari ST 8 MHz would be 0.8 MIPS which is very close to the Amiga 0.7 that gets quoted a lot. And, that's assuming that what ever interrupts are happening in the background are not slowing it down even more. ie.. 0.8 MIPS is probably best case.

I don't think that the charts I provided are wrong, they are Drhystone MIPS results which were done all the same for all these systems and relatively speaking match the MIPS that have always been considered for dedicated computers from a comparison between systems perspective.

The difference is more about what really happens in a chess program when it calculates compared to these DMIPS tests and it seems to me there is loss in all cases when chess programs calculate as seen from the DMIPS table the dedicated computer processor list.

ie Motorola 12 MHz = 2.1 MIPS & 6502 5 MHz 2.15 MIPS on chart I provided compared to dedicated 1.6 MIPS versus 1.2 MIPS used by most people as a rule of thumb.

Also here is some information from Wikipedia:

https://en.wikipedia.org/wiki/List_of_I ... processors

8086 = 8 MHz with 0.66 MIPS
80186 = 6 MHz with > 1 MIPS
80286 = 6 MHz with 0.9 MIPS

So would say yes 186 and 286 are similar but not 8086 which seems to be half the speed.

It's all very interesting.

best regards
Nick
Martin Hertz
Member
Posts: 76
Joined: Wed Oct 07, 2015 7:08 pm

Post by Martin Hertz »

Thank you for the explanations. I've made some tests on the HP-1000CX for better comparison:

Chess Player 2150 needs 00:01:32 for finishing depth 3 after f3.

Sargon III needs 00:00:53 for finishing depth 5 after f3.

Psion 2.13 needs 00:04:15 to find d6 at the BT2450-18 test, where the Amsterdam needs 00:02:57.

I hope this helps a bit. Would be interessting to see the 286 results.
Martin Hertz
Member
Posts: 76
Joined: Wed Oct 07, 2015 7:08 pm

Post by Martin Hertz »

spacious_mind wrote: ...
Also here is some information from Wikipedia:

https://en.wikipedia.org/wiki/List_of_I ... processors

8086 = 8 MHz with 0.66 MIPS
80186 = 6 MHz with > 1 MIPS
80286 = 6 MHz with 0.9 MIPS

So would say yes 186 and 286 are similar but not 8086 which seems to be half the speed.
The Wikipedia information about the speed of the 80186 seems not to be correct. In the meanwhile I have tested
the 80186 @ 7.9 MHz compared to the 80286 @ 10 MHz with Rexchess. The 80286 is around 3 times faster, what is
also visible after pressing alt+h. The TimeRef on the 80186 is 2150 vs 726 on the 80286.
User avatar
spacious_mind
Senior Member
Posts: 4018
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Martin Hertz wrote:
spacious_mind wrote: ...
Also here is some information from Wikipedia:

https://en.wikipedia.org/wiki/List_of_I ... processors

8086 = 8 MHz with 0.66 MIPS
80186 = 6 MHz with > 1 MIPS
80286 = 6 MHz with 0.9 MIPS

So would say yes 186 and 286 are similar but not 8086 which seems to be half the speed.
The Wikipedia information about the speed of the 80186 seems not to be correct. In the meanwhile I have tested
the 80186 @ 7.9 MHz compared to the 80286 @ 10 MHz with Rexchess. The 80286 is around 3 times faster, what is
also visible after pressing alt+h. The TimeRef on the 80186 is 2150 vs 726 on the 80286.
Hi Martin,

Sorry I have forgotten all about this post. I am on vacation next week, when I get the chance I will start my 80286 and do the same test and we can compare.

Best regards
Nick
Post Reply