Author: Robert Hyatt
Date: 18:19:17 12/23/03
Go up one level in this thread
On December 23, 2003 at 12:11:26, Vincent Diepeveen wrote: > > >If old persons like hyatt deny such truths here time and again, what more can i >say? It's better to ignore this forum when he again is rambling such nonsense >that his $150 gigabit network card (price nowadays) is faster latency than a new >$1500 myrilnet work card. Not to mention even more expensive interconnects which >are put straight onto special supercomputer mainboards (with ECC of course). >Note that even the cheap $10k router used in Bob's 8 node cluster is eating 50 >ns from bandwidth. So adding the latency of the memory: 400 ns + 2x50 for the >router you're already starting at 0.5 usec and still must count the pci bus + >network card latency to that :) > Why don't you do your homework _first_? The cLAN stuff is _not_ called "gigabit". It is called "cLAN". It is 1.25 gigabits full duplex, 1.25 gigabits in _both_ directions simultaneously. The PCI cards are not $150. They were almost 10x that. The 8-port switch we have costs over $18,000 US. You might call the technology office at NSF and ask the CISE program director about this stuff. He always told me "Please try to use myrinet. It is much more cost-effective, even though the cLAN stuff is signficantly faster. But it is more than significantly more expensive." All you have to do is look. Dell sold this stuff, they could also give you info. You just have to talk to their "cluster" people. >Yet he claims 0.5 usec, see how simple it is to refute the nonsense he writes >for those who know technical something in that area? _I_ don't claim that. It is an industry-accepted performance metric for that particular network. Point-to-point, no conflicts. The switch has 8 ports. A can talk to B independent of C talking to D, etc. However, you can continue to flap and wave "it doesn't exist" all you want... > >Coming back again onto simply measurable truths which get rewritten time and >again each time, the hyatt person simply should stick to his word and retire >when losing that match :) Unfortunately for you I didn't lose it, so now what? Any chance _you_ will retire and do the world a _big_ favor? > >>seem to have a problem with your style of presentation and his style of >>presentation can sometimes be a wee bit abrasive at times. [ : ) ] Please do > >replace abrasive by 'outdated knowledge'. It is logical that knowledge gets >outdated when someone doesn't show up at computerchess tournaments. Most >programmers must feel in person reality in such tournaments before realizing >truth. > >Uri Blass is a good example. He started also writing the biggest nonsense here >but i guess he'll learn a lot in 4-14 july 2004. Especially that one should not >open 1.h3 :) > >Hyatt is also in the Uri Blass level. The difference is that he doesn't join >tournaments in 2004 other than some tournament where no commercial programmer is >chatting with you. Please show up at the next CCT and demonstrate how bad I am. :) > >Nothing makes up for chats in a world championship during rounds, between rounds >and in the restaurant in the evening. > >The CCC nothing interesting will be ever posted of course. The data i show here >i already had shown to several persons who joined world champs. > >>not engage in non-technical "personality-bashing" with Bob or anybody else but >>stick to the technical, and please do your very best to remain objective in the >>face of all provocations, if possible. >> >>This is good advice for all of us, IMHO. : ) >> >>Bob D.
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.