[REVIEW] HIARCS 6.0 Macintosh (long)
Richard A. Fowell
Here's a review of the strongest chess program for the Macintosh
(at tournament time controls vs. other chess programs, at least).
fow...@netcom.com (Richard A. Fowell)
======================================================================
Program Name: HIARCS 6.0 Mac
Type: Chess Playing Program (Commercial)
Requires: 68020+, System 7.1+, 640x400 pixels+, CD-ROM, 3MB RAM
List Price: $150
Street Price: $116 US ($58 US if upgrading from HIARCS 1.0 Mac)
Copy Protection: No
PPC Native: Yes
Publisher: Applied Computer Concepts Ltd.
HIARCS 6.0 Mac is the Macintosh version of the top PC program on the
November 1997 SSDF rating list [1], and HIARCS 6.0 runs even faster on
Power Macintoshes than on Pentiums [2]. HIARCS 6.0 Mac should be about
130 rating points stronger than (almost twice as strong as) the second
strongest Mac chess program, Chessmaster 4000 Mac [3]. HIARCS 6.0 Mac
was released in October 1997.
HIARCS 6.0 Mac isn't for everyone. Even the freeware MacChess provides
a master-strength opponent that few people can beat
(<http://members.aol.com/Macchess/>), and for $40, Chessmaster 4000
provides a nearly undefeatable opponent for most of us, with varied
opponent styles, tutorials, etc. (See review:
<http://www.imgmagazine.com/issues/vol5_issue5/chessmaster_review.html>)
Unlike Chessmaster 4000, HIARCS has no tutorials, a single 2D set, and
no 3D set. However, unlike the unhappy plethora of Chessmaster sets,
the HIARCS set is the legible "diagram-style" style from chess books.
The increased strength and special analysis features of HIARCS make it
worth considering for serious chess players, however. For post-mortem
analysis and analysis of pet opening variations, an engine that is 130
points stronger than Chessmaster for something less than the cost of
entry to a major chess tournament is not unreasonable. HIARCS has
several special analysis features that Chessmaster lacks:
- Analyse W or B: A key reason for buying the best engine you can get
is to analyse your games to see where you can improve.
This task requires only the best move for your side
(White OR Black) in each position. Both HIARCS and
Chessmaster will analyse a game from a given position
to the end, and log the results to a file. However,
HIARCS gives you the option of analyzing for only one
side, while Chessmaster always analyzes for both,
which takes twice as long.
- Monitor mode: you move the pieces manually, and every time you move,
HIARCS immediately starts analyzing.This mode also
enables the "k-var" option and the opening tree display.
- "k-var" option: (HIARCS6:variations) Here, HIARCS displays the best
"k" continuations (k=1,2 or 3, set by the user). This
shows you not only the "best" move, but the best
alternatives as well. Of course, the search depth for
any single variation is less, since the computer has
to work several variations in parallel.
- Opening display: In "monitor mode", HIARCS displays (in addition to
its analysis) the list of responses in its opening
book for that position, followed by (what I assume are)
numbers representing the relative frequency that HIARCS
will choose these moves. HIARCS also displays the
opening name and ECO code whether in "monitor mode" or
not. Chessmaster displays opening names, only. Since
the HIARCS books is advertised as having 143,000
positions in it, there is a lot of information here.
- Next best move: This makes HIARCS take back its last move and look for
the next best alternative. This is a more "hands-on"
alternative to "k-var", and also an interesting
way to force HIARCS to play a different opening line,
or to handicap HIARCS.
The HIARCS 6.0 interface follows the Macintosh guidelines for the most
part. Two exceptions: it uses cyclic (changing) menu items rather than
hierarchical menus for several options, and the position setup uses menu
commands for many things usually handled by buttons. However, it is much
more "Mac-like" than Chessmaster's baroque buttons, sliders and
non-defaulting dialog boxes. The interface is a slightly improved version
of the HIARCS 1.0 Mac interface. The HIARCS 1.0 Mac demo is available at:
<ftp://mirrors.aol.com/pub/info-mac/game/brd/hiarcs-chess-10-demo.hqx>.
(There is no demo version of HIARCS 6.0 Mac). The documentation consists
of a 4-page Mac-specific pamphlet, a quick reference card, and the
43-page HIARCS PC manual.
A thorough list of the HIARCS 6.0 Mac features can be found on the Web at:
<http://www.acc-ltd.demon.co.uk/mach6.htm>. Those familiar with HIARCS
6.0 PC will note that these features are omitted from the Mac version:
- database filter and non-PGN import
(try the freeware ExaChess Lite 1.1 for this, available at:
<ftp://mirrors.aol.com/pub/info-mac/game/brd/exa-chess-lite-11.hqx>)
- variable hash table size (HIARCS 6.0 Mac uses a 256K hash table)
- book opening editor
- autosensor board support
- figurine notation support
- board/piece customization
(The adventurous can do this with ResEdit)
- colored text.
I put together a HIARCS 6.0 speed benchmark by taking 32 problems from
the LCTII test and running them for 4.5 minutes each to simulate
tournament time controls with thinking on the opponent's time. Note that
"Analyse EPD" only reports nodes used if the "bm" (best move) or
"am" (avoid move) EPD tag is used, and HIARCS stops when it first finds
the solution. Note that this can lead to falsely high scores on EPD
problem tests, since HIARCS might change its mind given more time.
For this test, I used false "best moves" in the test file to make HIARCS
think the full 4.5 minutes. Here are results for HIARCS 6.0 Mac and PC.
These nodes/second (nps) speeds are fairly consistent - when the speed
on each problem is normalized to a generic model created by averaging
the solution times across the machines, the standard deviation of the
nodes/second (nps) on a given machine was less than 4%. Note that the
second row here is a machine using a Newer Technologies upgrade card
(250 MHz "G3" or "Arthur" chip). Apple's new 266MHz G3 machines should
do pretty well here, too.
HIARCS 6.0 Speed on Macs and PCs
Machine Processor Mean nps Mean nps/MHz
================================================================
Mac 9600/350 350MHz 604e 39,489 nps 113 Sys 7.6.1
Mac 9600/250 250MHz 750 36,197 nps 145 Sys 8 + Speed Doubler
Mac 7300/180 180MHz 604e 22,184 nps 123 Sys 7.5.5
AMD K6/200 200MHz K6 21,396 nps 107
Mac 20th Ann. 250MHz 603e 12,591 nps 50 Sys 7.6.1
Pentium 120MHz P5 8,273 nps 69
Mac 6100/66 66MHz 601 5,651 nps 85 Sys 7.5.1?
Mac Quadra 610 1,280 nps Sys 7.5?
Comparison between HIARCS 1.0 Mac and HIARCS 6.0 Mac
If you own HIARCS 1.0 Mac, you may be wondering whether it is worth
upgrading. Between the new engine and increased speed, Power Macintosh
owners should see a strength increase of over 200 rating points (150
if Speed Doubler was used before). This increase is largely due to the
increased speed of compiling to PPC native code (7-10 times faster than
68K code without Speed Doubler), and the rest due to a stronger engine
(60 Elo stronger than the HIARCS 4.0 engine based on SSDF results).
The strength increase is much less impressive for older Macs - just
that 60 Elo. In addition, there are the improved analysis features:
- k-var mode
- opening display of ECO codes and books alternatives
- game analysis for White/Black only (this didn't work in version 1.0)
- improved PGN import/export features
- improved EPD analysis features (recognizes "bm" and "am" codes for
problem tests)
There are also a number of other new features: a "Max Time" option
(for fixed time/move, use Level:Infinite and this option), storage
of user annotations for each move, purge EPD file of annotations,
six levels of selectivity vs. three in v1.0, support for Fischer
Shuffle chess, English/German/French menus all standard, a larger
opening book, and a board pattern that displays better on B/W displays.
Conclusion
HIARCS 6.0 is the strongest Macintosh chess program, and has advanced
analysis features its closest rival lacks. HIARCS is expensive, but
cheaper (and stronger) than a good stand-alone chess computer. For
serious chess analysis, HIARCS is the best Mac chess program available.
Where to buy HIARCS 6.0 Mac
A list of HIARCS distributors is on the publisher's web site:
<http://www.acc-ltd.demon.co.uk/dist.htm>
The U.S. distributor ICD/Your Move Chess & Games is at
(800) 645-4710, <http://www.icdchess.com/>, email I...@ICDchess.com
Another distributor to consider is the Gambitsoft site in Germany,
which quotes a 7-10 day delivery to the USA, and sells HIARCS 6.0 Mac for
$116 US ($58 US for upgrade from HIARCS 1.0 Mac) including shipping.
The Gambitsoft site is at: <http://www.gambitsoft.com>
Notes
[1] The SSDF rating list can be found at:
<http://home3.swipnet.se/~w-36794/ssdf/>
Note that, statistically speaking, HIARCS 6.0 is only "first amongst
equals" - the difference between it and Rebel 9 is statistically
insignificant. Also, the newest versions of MChess, Fritz, Nimzo,
and others are not yet rated - they may be better.
[2] Based on a thorough (33-problem, 2.5 hour) HIARCS 6.0 speed benchmark.
On my 7300/180 (180MHz 604e), HIARCS 6.0 runs 10% faster than a 200MHz
AMD K6. The nodes/second/CPU MHz speed for HIARCS 6.0 on machines with
256K L2 cache was: Mac 604e: 123, AMD K6: 107, Mac 601: 85,
Pentium P5: 67.
[3] The Selective Search rating list at
<http://www.elhchess.demon.co.uk/ehss.htm>
estimated these Elo ratings on Pentium 100s:
HIARCS 6.0: 2536 +/- 16, Chessmaster 4000: 2408 +/- 51.
The HIARCS Mac menus:
FILE EDIT ANALYSIS VIEW
============================================================================
New Game [N] Undo [Z] Monitor [M] Show Analysis
Save Game [S] Cut [X] Takeback [T] Flip Board
Restore Game [R] Copy [C] Forward [F] Coordinates
List Game [G] Paste [V] Mark Position Sound
Export EPD Position Previous Position Flash
Import EPD Abort Edit Next Position Replay
Analyse EPD Clear Board Goto Move Language
Purge EPD Toggle to Add Goto Start [[] Game Details
Export PGN Move Number Goto End []]
Import PGN Castling Rights Goto Keypoint
Delete PGN Commit White Edit [W] Analyse Game
Save Preferences Commit Black Edit [B] Add Annotation
Quit [Q] Delete Annotation
GAME LEVEL HIARCS OPTIONS
============================================================================
Play Instant Move Now [ESC] Etiquette
Pause Average Resp. [J] Give Hint [H] Your Name
Swap Sides Adaptive Best Line Set Piece Chars
Fix Time White Preset TimeCtrl Auto Play Short Algebraic
Fix Time Black Change TimeCtrl Next Best Set Draw
Offer Draw New TimeCtrl Permanent Brain Move Rule
Resign Incremental Clk [K] Selectivity Max Depth
Adjudicate Game Infinite [I] Style Max Time
Optimise Best Play Book
Hash Table Current Rating
Variations Clear Rating
Shuffle Chess
Junior McCarthy
Richard,
Thanks for the excellent review. I have the 1.0 version and have had it run
out of time when set for overnight analysis. This is sort of aggravating
when it hasn't analyzed the key point in the games. Have you ever run into
this before and is it likely to have been corrected because of the
increased speed on my PPC?
Regards,
Scott Cole <sc...@sbt.infi.net>
Richard A. Fowell
> Thanks for the excellent review. I have the 1.0 version and have had it run
> out of time when set for overnight analysis. This is sort of aggravating
> when it hasn't analyzed the key point in the games. Have you ever run into
> this before and is it likely to have been corrected because of the
> increased speed on my PPC?
My copy of version 1.0 did have a tendency to use much too much time on
some moves. This was a major annoyance to me in version 1.0, since I used
"analyse game" a lot, and it often didn't finish in the time it should have.
In HIARCS 6.0, however, you can force HIARCS to use a fixed
time per move by setting "Level" to "infinite", and "Max Time" to the
number of seconds you want HIARCS to think on each move. This approach
has been perfectly reliable for me.
fow...@netcom.com (Richard A. Fowell0
Bernard Pulham
Thankyou for the very thorough review, perhaps ignored by those
satisfied with the dos version! I expect the commercial pressures will
continue to ensure only the standard release will be showcased in
tournaments, wather than one which runs on different hardware. But with
both dos and Mac appearing on one CD (we are told) perhaps a few will be
curious enough to try their copy out on a fast mac if they know a friend
who has one...
Does the mean nps/mhz always stay constant for a particular processor,
(enabling us to interpolate nps for a 200mhz P5 for example)?
It must be said that there are a few more non-standard mac interface
aspects that you don't mention.
Firstly the windows are neither resizable nor collapsible.
Secondly clicking the movelist scroll bars achieves the play-forward and
play-backwards functions, (totally unexpected by mac users!).
Your nodes per second comparisons demonstrate that HIARCS6ppc is easily
the strongest implementation of HIARCS6. So where does that leave all
the other chess programmers who only have one platform?
HIARCS MacTeam
Bernard Pulham <bern...@enterprise.net> wrote in article
<3485B6...@enterprise.net>...
> It must be said that there are a few more non-standard mac interface
> aspects that you don't mention.
> Firstly the windows are neither resizable nor collapsible.
Fair comment !
> Secondly clicking the movelist scroll bars achieves the play-forward and
> play-backwards functions, (totally unexpected by mac users!).
I'm curious - what would be an expected behaviour ?
There is a menu to achieve this for those who so wish.
Not being defensive - it is just good to know for future reference.
Best regards,
Clive
Komputer Korner
Not true, you forget that the Hiarcs Mac is crippled with an extremely low
hash table limit (256K) . The effect is to nullify its hardware advantage
to the point that a Pentium II -300 Mhz PC is the fastest platform when
equipped with 64 Mb of RAM running from a DOS boot disk.
--
- -
Komputer Korner
The inkompetent komputer
If you see a 1 in my email address, take it out before replying.
Please do not email both me and the r.g.c.c. at the same time. I read all
the postings on r.g.c.c.
Bernard Pulham <bern...@enterprise.net> wrote in article
<3485B6...@enterprise.net>...
> Thankyou for the very thorough review, perhaps ignored by those
> satisfied with the dos version! I expect the commercial pressures will
> continue to ensure only the standard release will be showcased in
> tournaments, wather than one which runs on different hardware. But with
> both dos and Mac appearing on one CD (we are told) perhaps a few will be
> curious enough to try their copy out on a fast mac if they know a friend
> who has one...
>
> Does the mean nps/mhz always stay constant for a particular processor,
> (enabling us to interpolate nps for a 200mhz P5 for example)?
>
> It must be said that there are a few more non-standard mac interface
> aspects that you don't mention.
> Firstly the windows are neither resizable nor collapsible.
> Secondly clicking the movelist scroll bars achieves the play-forward and
> play-backwards functions, (totally unexpected by mac users!).
>
Richard A. Fowell
In article <3485B6...@enterprise.net> bern...@enterprise.net writes:
>Thankyou for the very thorough review, perhaps ignored by those
>satisfied with the dos version! I expect the commercial pressures will
>continue to ensure only the standard release will be showcased in
>tournaments, wather than one which runs on different hardware.
Actually, HIARCS Mac on a 275MHz G3 competed in the German tournament
quite recently.
> But with
>both dos and Mac appearing on one CD (we are told) perhaps a few will be
>curious enough to try their copy out on a fast mac if they know a friend
>who has one...
>
>Does the mean nps/mhz always stay constant for a particular processor,
>(enabling us to interpolate nps for a 200mhz P5 for example)?
While there are always other things (such as L2 cache size, memory
bus size, L1 cache MHz/CPU MHz ratio) that impact this, nps/MHz does
seem fairly stable. For example, note that in my table, the
180MHz 604e has a nps/MHz of 123, and the 350MHz 604e has a nps/MHz of 113 -
not bad consistency for a factor of almost two on clock speed.
Always better to actually run the test, if possible.
>
>It must be said that there are a few more non-standard mac interface
>aspects that you don't mention.
>Firstly the windows are neither resizable nor collapsible.
Oops. Too used to it, I guess. And after I lambasted CM4000 for its
interface sins (CM4000 is actually pretty good in that regard, despite
it's other sins.)
>Secondly clicking the movelist scroll bars achieves the play-forward and
>play-backwards functions, (totally unexpected by mac users!).
True again. Personally, what I like to see in a move list is all the
features supported by, say, ExaChess:
- "VCR-style" buttons for the non-cognoscenti (not in HIARCS 6.0)
- Click on move takes you to that move (not in HIARCS 6.0)
- cursor keys move you forward/backward (not in HIARCS 6.0).
-
>
>Your nodes per second comparisons demonstrate that HIARCS6ppc is easily
>the strongest implementation of HIARCS6.
Well, only if the hash table issue that KK raises is a non-issue.
The old rule of thumb was "8 rating points per hash table doubling".
If that is really true, then the 256K hash table will be giving 64
rating points benefit to a 300MHz Pentium II with 64Mb hash, which
will, as KK suggests, overwhelm the speed advantage of the Mac, which
I would guess is only something like 25-30 rating points benefit, assuming
the Pentium II has the same nps/MHz as a 200MHz K6 (I'd REALLY like a
benchmark on a Pentium II/300, rather than this guessing!).
Conventional wisdom seems to be "the more hash the better". When I
raised this issue with the HIARCS team during the beta test, they
didn't seem concerned about this (for example, they could have trivially
increased the table size to, say, 1 Mb), and I had got one response in
the past saying that they felt that larger hash tables did not seem to
increase strength enough to pay for the increased overhead of maintaining
them. Robert Hyatt seems to like lots & lots of hash table space, though.
One thing that would be a nice data point here - if any SSDF folks are
reading this - what hash table size do the various programs in the SSDF
use? If HIARCS is at the top of the list with smaller hash tables than
the competition, that might support a contention that huge hash tables
aren't overwhelmingly important in SSDF competition.
Any volunteers with a Pentium II/300MHz and HIARCS 6.0, please send me email
and I'll send you the test file - only takes 15 minutes of your time
(and 2.5 hrs of machine time) to run.
> So where does that leave all
>the other chess programmers who only have one platform?
Probably still wishing they ran on a 600MHz Alpha, actually <grin>.
But maybe us poor Mac users will get a little more respect <not likely>.
And maybe someone will have enough interest to write the hand-coded
assembly routines for the Crafty bit-board stuff so the Crafty Mac nps
is respectable for a change.
********* NEWS FLASH! **************
I just got benchmark results on the new Apple desktop 266MHz G3!
Now we know why Apple retired the 9600/350 - this thing is even faster!
(well, for HIARCS, anyway). And a heck of a lot cheaper! Not too shabby
for a system with only 512K of L1 cache and that running at 1/2 the CPU MHz.
Check out that nps/MHz! (by the way, over the 32 problems used, the
standard deviation of the speed between the G3 and the 9600/350 was less
than 2%, which is tight, but which does mean that the 1% greater speed of
the G3 isn't really statistically significant).
Does anyone have one a 300MHz G3 clone with 1 Mb of 1:1 L1 cache
to benchmark?
Here's an updated table. I've also included nps for the 68K version of
HIARCS 6.0 (2nd to last row) so you can see what a difference the native
PPC code makes - a speedup of 8.5x here. Keeping in mind that HIARCS 1.0
Mac did not have native PPC code (though Speed Doubler helps narrow the
gap a good bit).
HIARCS 6.0 Speed on Macs and PCs
Machine Processor Mean nps Mean nps/MHz
================================================================
Mac G3 /266 266MHz 750 39,934 nps 150 Sys 8, no extensions
Mac 9600/350 350MHz 604e 39,489 nps 113 Sys 7.6.1
Mac 9600/250 250MHz 750 36,197 nps 145 Sys 8 + Speed Doubler
Mac 7300/180 180MHz 604e 22,184 nps 123 Sys 7.5.5
AMD K6/200 200MHz K6 21,396 nps 107
Mac 20th Ann. 250MHz 603e 12,591 nps 50 Sys 7.6.1
Pentium 120MHz P5 8,273 nps 69
Mac 6100/66 66MHz 601 5,651 nps 85 Sys 7.5.1?
Mac 9600/350 350MHz 604e 4,614 nps 13 *EMULATING 68000*
Mac Quadra 610 1,280 nps Sys 7.5?
fow...@netcom.com (Richard A. Fowell)
Visit the MacChess page at http://members.aol.com/Macchess/
Bernard Pulham
HIARCS MacTeam wrote:
>
> Bernard Pulham <bern...@enterprise.net> wrote
>
> > clicking the movelist scroll bars achieves the play-forward and
> > play-backwards functions, (totally unexpected by mac users!).
>
> I'm curious - what would be an expected behaviour ?
> There is a menu to achieve this for those who so wish.
>
The scroll arrows always (in my Mac experience) simply allow a long list
to be available, by scrolling it up and down, so you can click on
information that was out of view. I cannot think of any other Mac
program where as well as moving the information into view, the arrows
actually perform a function (in this case playing one move backward or
forward on the gameboard).
Put it another way: In most Mac programs the scroll bar buttons only
become available when the info has outgrown the window, until then they
are dimmed.
For example when I was near the beginning of a game, and all the move
info fitted in the move list window with room to spare, I felt no reason
to try the scroll arrows, and went searching in the menubar for move
forward and backward instead, wondering why there were no buttons
available. I was surprised (and pleased) to find the scroll arrows had
this function also, when I stumbled upon it later in the game when the
list had outgrown the window!
This probably seems esoteric, but I think this might well perplex us Mac
Users who rely too much on intuition....and would recommend two obvious
buttons like in the PC version!
Bernard
Bernard Pulham
Bernard Pulham wrote
> > Your nodes per second comparisons demonstrate that HIARCS6ppc is easily
> > the strongest implementation of HIARCS6.
Komputer Korner wrote:
>
> Not true, you forget that the Hiarcs Mac is crippled with an extremely low
> hash table limit (256K) . The effect is to nullify its hardware advantage
> to the point that a Pentium II -300 Mhz PC is the fastest platform when
> equipped with 64 Mb of RAM running from a DOS boot disk.
> --
> - -
I agree that the small hash tables are an issue, but what tests have
demonstrated that the increased nodes per second are "completely
nullified", by larger hash tables?
Bernard