Author: K. Burcham
Date: 07:08:20 07/01/01
Go up one level in this thread
very interesting reading.
if we assume he is refering to the 1000 mhz pentium 3 in this paragraph.
and i think he is.
the pentium 3 gets about 900 kns or 900,000 nodes per second.
in this paragraph with the "reverse bitboards concept", he says this
could increase to 10,000,000 nodes per second.
it seems this would be a 10x increase in nodes per second, without any
hardware improvements. amazing......
Hyperbola - Coming 2001
Well, the new use of bitboards to accomodate simple 64-bit operations
promises very high-speed calculations for Hyperbola, especially considering the
decreased reliance on memory tables. The large tables just for sliding piece
calculations under current programs can't even fit into the Level 2 cache of the
Coppermine P-III, although most recently-used addresses would remain cached.
However, even an access to the L2 cache degrades performance: the Coppermine,
with the full-speed ATC, still has approximately a 9-clock latency. Since the
16K L1 cache can only hold 512 cache lines, it is probable that most of the
recently used table values will get bumped to the L2 cache, or even to memory,
which has a 12-clock latency even with PC800 RDRAM.
Hyperbola is a long way away from first release or even alpha testing, but the
technological advancements, such as the concept of reverse bitboards explained
in detail above, promise to give this new engine a significant edge. It is
possible that, even with a relatively thorough evaluation function, the nps
(Nodes Per Second) rate could exceed 10,000,000.
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.