Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty multithreading

Author: Robert Hyatt

Date: 06:20:23 05/23/98

Go up one level in this thread


On May 23, 1998 at 09:11:14, Ernst A. Heinz wrote:

>On May 22, 1998 at 16:52:31, Robert Hyatt wrote:
>>
>>On May 22, 1998 at 15:12:19, Ernst A. Heinz wrote:
>>
>>>On May 22, 1998 at 13:19:04, Robert Hyatt wrote:
>>>
>>>Bob,
>>>
>>>Great WAC results -- my congratulations!
>>>
>>>>Note that the ALR machine is running approxmiately as fast as the 500mhz
>>>>alpha machine we used in Paris, for comparison.  On that machine we were
>>>>hitting around 250K nodes per sec...  on this machine roughly a little
>>>>faster, but about 1/4 to 1/3 of the machine is "wasted" due to search
>>>>overhead...  So that the newer alpha machines would actually be a bit
>>>>faster on one processor, and way faster with more than one...
>>>
>>>Sounds a bit strange to me if you look at the SPECint95 ratings of
>>>a PPro 200 MHz = 8.7 (i.e., 4x PPro 200MHz = 4x 8.7 = 34.8) and a
>>>500MHz Alpha-21164a = 15.0 ...
>>>
>>>Are you sure about comparing the speeds on the same positions?
>>>
>>>=Ernst=
>>
>>yes.  We ran several tests comparing a single pentium pro/200 to the
>>alpha/500 we used in Paris, and got 3.1X speedup on that machine.
>
>Hmh, for "Crafty 14.10" Kai Luebke's WWW pages mention the following
>benchmark speeds:
>
>CPU             MHz  hash   hashp  nps     n/M  Tester
>--------------  ---  ----   -----  ------  ---  ------
>Alpha 21164     500  96.0m   5.0m  187549  375  Michel van der List
>Pentium II      300  48.0m   5.0m  122460  408  Michel van der List
>Pentium II      233  48.0m  10.0m  111475  478  John Perry
>Pentium Pro    (233) 24.0m   5.0m  107949  ---  Kai "AMD-Killer" Luebke
>AMD K6          233  12.0m   5.0m  101103  434  Werner Muehlpfordt  (1)
>HP PA8000       160  12.0    640k  100485  628  Michael Kalisz
>AMD K6          233  12.0m   5.0m   93735  402  Djordje Vidanovic
>Pentium Pro     200  24.0m  10.0m   87143  ---  Kai Lübke
>
>According to this list, "Crafty 14.10" runs roughly 2x faster on a
>500MHz Alpha-21164a than on a 200MHz PPro which corresponds quite
>well to the relative SPECint95 ratings of both CPUs ...
>
>=Ernst=


it seems to depend on the machine.  IE when Bruce ran crafty on his
polywell, it didn't do nearly so well.  But in the case of Crafty on
the DEC 500mhz machine, it did fly right along...  Jason had a benchmark
he used (now in the bench command in crafty, which is much better test
than the "a3" test used in the chart you gave above) that produced about
80K on the P6/200, and (finally) 250K on the alpha in Paris.  When he
left the USA, we were running roughly 2.5X faster on the alpha, but he
played with the optimizer between rounds and reached 3.1X as the best
optimizations he could find.  You can find out what optimizations he
used
from jdeines@mindspring.com if you are interested...

I believe the machine we used had a 1M L2 cache...  not sure about the
memory however...  and as I mentioned...  on the problem set of WAC,
crafty
is running about 2.2X faster than with one processor...  in a game, it
seems
to be running about 3-3.2X faster when there is continuity from move to
move to help move ordering...  next chance I get, I'll run wac on a
single
alpha and compare/report the results...

Bob

BTW, several versions of Crafty have hit 299/300 on a P6/200 one
processor,
although most do 297.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.