Author: Robert Hyatt
Date: 16:53:25 09/16/05
Go up one level in this thread
On September 16, 2005 at 11:40:03, Joshua Shriver wrote: >A friend of mine uses myrinet on his opteron cluster. I've heard many good >things about it with respect to having a relatively low latency. > >Kind of OT, but I remember reading that you once used crafty or another engine >on a Cray before. Mind talking about it? > >From what I understand about Cray's, they're like a cluster in that they have a >lot of cpu's all working together instead of one massive processor. Though it >used some kind of special logic board for connecting them all together so no >network like ethernet/myrinet/etc between nodes was needed (or something to that >effect). wrong machine. You are talking about the T3 family. The machines I ran on were, in order, the Cray-1, the cray-XMP, the cray YMP, the Cray C90, and finally the Cray T90 with 32 processors... All were pure SMP boxes, memory and processors connected with a crossbar, no message passing or anything like that... > >With today's extremely fast processors, it seems the super computing market has >died down, or at least switched. A couple years ago I went to a seminar at the >Pittsburgh Super Computing Center and at the time they where almost entirely >cluster based (Win NT or Linux). I talked with one of the gentlemen in charge >after one day and was shocked to hear they had a Cray (1? 2?) just sitting in >storage unused. Apparently it cost over a $1M just for the electricity bill, and >it required some kind of special coolant that was expensive even for a small >amount. > >So compared to super fast mainframes or supercomputers... it seems clusters >give you the more bang for the bug. No special hardware needed, and cheaper >costs. > >But what does this mean for computer chess? > >If money wasn't an issue, what would really be ideal hardware for the best >computer machine? Biggest SMP (shared memory) machine one could build. But it would be expensive, as scaling to even 32 processors without resorting to NUMA is costly... > >Top of the line IBM mainframe? 200 node Dual core-dual opteron cluster? >Assuming that the code could utilize the hardware. The NUMA machines are interesting, because they scale better than the pure cross-bar SMP boxes. And before you-know-who hijacks the "scale" word, the correct usage here means that NUMA architectures are more cost-effective as the numbers of processors is increased, because they are generally a nearly linear-cost-increase, while SMP cost increases as the square of the number of processors... They have their interesting issues that have to be addressed. Hopefully I'll get to test on a quad-core NUMA box late this year or early next year. That will have a new set of issues to deal with, just as the dual-core boxes had their unique issues this year. > >Sorry for the ramble :) just something on my mind. > >Josh > >> >>Yes. gigabit ethernet is high bandwidth, but still long latency. We have a new >>128 node dual xeon cluster in the department using myrinet, which is lower >>latency. Our old cLAN switch was the lowest latency I have ever seen, but it >>was pricey as all hell... It was also not TCP/IP based...
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.