Author: Robert Hyatt
Date: 22:05:03 03/07/03
Go up one level in this thread
On March 08, 2003 at 00:51:46, Matt Taylor wrote: >On March 05, 2003 at 15:28:26, Vincent Diepeveen wrote: > >>On March 05, 2003 at 13:30:05, Anthony Cozzie wrote: >> >>the basic problem of SMT/HT is that the P4 is not the ideal cpu to use SMT/HT. >>the k7 is a much better cpu for SMT/HT usage. therefore the speed wins for >>SMT/HT are not so convincing. > >And why is that? Because the K7 is efficient? Somehow I thought the only reason >the P4 gained much was because it was inefficient... > >Also, seems like the K7 has a faster bus at present, which is the other place >where HT is useful. > >>Another problem of SMT/HT in general is that the OSes fire threads and processes >>at a speed which is just having a too big latency. For vaste majority of >>databases it is not true to say that SMT/HT helps there, because the weakest >>chain there is the speed/latency of the harddisks, not the IPC of a processor. > >I don't know many high-end databases that run off of hard disks, do you? > >-Matt This "latency" issue is a crock for chess programs. If a CPU is idle, there is _no_ latency from the time a process is marked "ready to run" until it is "running" except for the limits imposed by the hardware for setting up to do the context switch to the newly unblocked process. IE load regs, set up the memory map, floating point stuff, and then "go". Then you have to do cache line fills to get the process into L1/L2 if it isn't already there. Which for a chess program is usually not a problem since it is all that is running. If you run _lots_ of threads (more threads than CPUS) then there is latency by design, otherwise you do too many context switches per unit of time and that wastes processor cycles. Vincent is confusing the latter with the former.
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.