Author: Robert Hyatt
Date: 21:32:15 07/04/03
Go up one level in this thread
On July 04, 2003 at 11:25:24, Vincent Diepeveen wrote: >On July 04, 2003 at 11:18:58, Vincent Diepeveen wrote: > >>On July 03, 2003 at 13:57:02, Dann Corbit wrote: >> >>>On July 03, 2003 at 12:28:05, Ralph Stoesser wrote: >>> >>>>Dear chess programmers, >>>> >>>>What are your personal experiences with the MTD(f) search introduced by Aske >>>>Plaat some years ago? >>> >>>It does not work for me as well as it does for some others. >>> >>>I think success will depend very much on your particular engine. >>> >>>Andrew Williams has a successful implementation. >> >>Claims to have a successful implementation is more near the truth. >> >>Hopefully getting a position from hashtable doesn't eat 4 >>microseconds with him. > >For those who do not understand what i mean, let's be clear for the non experts. > >Aske Plaat claimed a successful implementation of MTD in Cilkchess. However i do >not see how it is possible to implement MTD in a supercomputer program. I wish you would improve your vocabulary. For example, the Cray T90 is definitely a "supercomputer". It is a full SMP machine with 32 processors. Not NUMA. And mtd(f) would work just fine on that "supercomputer". Use the term right. You are talking explicitly about NUMA-type machines or pure clusters. There _are_ other types of "supercomputers" for which your comments make no sense at all. > >Without hashtables you in advance cannot use MTD. Too expensive researches you >need. Average latency at supercomputers is like 4 microseconds - 30 >microseconds. > >In short if you plan to get more than 15000 nodes a second at a cpu, forget it >at a supercomputer with MTD. 15000 nps means already hashtable eats 50% of your >system time with MTD and in reality it will eat more as you need more than 2 >references a node with MTD. 1 read and 1 write. Write is way more expensive than >load. I did significantly better than 15000 NPS per CPU on the T90, so again, be more careful with your terminology. I did 20K+ per second on 1980- level supercomputer single cpu. > >This is why cilkchess searched with CILK around 5k nps a cpu and at his laptop >without cilk it searched 200k nps. The supercomputer chip was higher clocked >than his laptop. Also the supercomputer chip was 64 bits and cilkchess as we all >know is a bitboards thing. > >How can you claim MTD to be succesful then? If it produces decent speedups and plays decent chess, it _must_ be successful. > >Best regards, >Vincent
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.