Author: Vincent Diepeveen
Date: 05:36:14 08/05/05
Go up one level in this thread
On August 04, 2005 at 10:55:28, Robert Hyatt wrote: >On August 04, 2005 at 05:18:43, Vincent Diepeveen wrote: > >>On August 03, 2005 at 13:34:09, Robert Hyatt wrote: >> >>>On August 03, 2005 at 10:21:20, Gian-Carlo Pascutto wrote: >>> >>>>On August 03, 2005 at 10:13:45, Sedat wrote: >>>> >>>>>Hi there, >>>>> >>>>>Does anybody has any information about this processor ? >>>>> >>>>>-Can i run engine-matches with ponder on ? >>>>> >>>>>I mean: >>>>>-Does the kns of the engines will fall down ? >>>>> >>>>>And if its possible to run ponder on matches : >>>>>-is it enough just one processor or i need to buy two processors ? >>>> >>>>A single dualcore processor behaves almost exactly like a 2 processor machine. >>>> >>>>-- >>>>GCP >>> >>>This needs a _lot_ more testing before saying that so positively. I've been >>>testing on a quad dual-core box, and there are most definitely "issues" to deal >>>with that I/we have not yet solved. There are some memory issues that I am >>>working on quantifying, probably related to two cores sharing a memory bank and >>>the associated bus contention. First cut on the quad 875 box produced some >>>really ugly SMP results for me, with the NPS "scalability" only reaching 4X >>>generally, where on the quad 850 I last tested on, it scaled perfectly for 1-4 >>>processors... >>> >>>More as I work out the glitches (I hope).. >> >>This is because crafty doesn't scale. >> >>Not a hardware issue. > >Vincent, please go away and come back when you have some clue about what is >being discussed. Crafty, using two cpus, on a quad single-core opteron, scales >perfectly. Crafty, using two cpus, on a quad dual-core opteron, scales >horribly. It isn't a "crafty issue". Oh yes it is a crafty issue. Just get Ubuntu and a default vanilla SMP kernel upgrade which takes just a few seconds to upgrade and 1 reboot and run some decent software instead of crafty and you'll see that software scales perfectly. > >More when we find out exactly what it is... > >Crafty scales perfectly on dual opteron, quad opteron, and 8-way opterons. But >the dual-core is adding a new problem that is currently undiscovered, but is >probably an issue of two cores with one memory path that is shared. Or it is >related to the MOESI cache coherency message bandwidth... Aha so that Diep works perfectly on dual cores and scales perfectly on dual cores (see www.sudhian.com) is because the dual cores are 'broken'? So not crafty is the problem but the dual core is the problem? Comeon. GROW UP! Learn some decent SMP programming, instead of blaming the hardware. It works fine for everyone else. The latency just increased from 133ns to 234 ns back to 200ns-220ns for 2.2Ghz dual cores. The real reason is how crafty has been SMP programmed. Multithreaded to mention one thing, though that doesn't need to be a handicap as Nalimov indicated. With all kind of artificial tricks you can reprogram multithreaded programs to multiprocessed programs :) Of course you still lose that one register, but well that's the last of your worries. The real problem is CRAFTY doesn't scale well at NUMA hardware above 2 cpu's. At 4 cpu's you still hardly feel it, but at 8 cpu's you sure do. > > > >> >>Memory latency is 234 ns to get 8 bytes of TLB trashing memory from 250MB >>buffers (in total 2GB ram for total testblock). > >Has zero to do with anything... > >> >>Compare with 400 ns that your own dual Xeon needs to deliver the same >>and compare with 700 ns that 8 processor Xeon needs. > >Has zero to do with anything... > >> >>I guess the central lock structure in crafty breaks it at 8 cpu's. > >You are guessing wrong. It's already run on 8-way (and beyond) single-core >boxes with zero problems... the dual-cores are experiencing problems at the >moment... Show us the output from a NUMA 8 way machine where you ran at. 8 processor Xeon, or that 8 way which you could get for world champs 2004 but didn't get as 'crafty didn't run on it well' will do. > >> >>Diep is not central locking, of course tested to work at ugly latencies >>until 500 cpu's and has zero problems with quad opteron dual core 1.8Ghz >>at which i play at. >> >>Please note the latency for 2.2Ghz dual cores is far better because the >>latency of each memory controller is somewhat dependant upon the speed of the >>processor. > >Latency is no better/worse for dual-cores than single-cores. We are swapping See results. Latency of single core at a dual is 111ns, versus 147ns for a single cpu test at a dual core. >them back and forth (same everything except for cpus) with zero problems and >memory latency is not changing one iota... > > > >> >>So the problem is not the hardware at all, but software issues within crafty. > >Nice to be able to debug something by reading tea-leaves. I'll report the >_real_ problem(s) as it(they) are discovered... Software issues in crafty :) > >> >>Any default x86-64 core 2.6.10 or later by default already is NUMA and works >>perfectly. No need to compile your own core. >Wrong answer. Default most recent redhat kernel crashes with numa=on. We just >built a new kernel to fix this yesterday... still not scaling correctly on >dual-cores, but scaling fine on single-cores... You compare a 4 way crafty versus 8 way. I just told you, the problem is 8 way crafty at a NUMA machine. > >> >>I installed Ubuntu at quad, upgraded to x86-64 kernel (thanks to Mridul >>Muralidharan for his big help!) and it worked fine. >>Ubuntu is the superior distribution nowadays. >> >>Vincent
This page took 0.01 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.