Author: Robert Hyatt
Date: 21:01:44 02/23/03
Go up one level in this thread
On February 23, 2003 at 23:04:07, Jeremiah Penery wrote: >On February 23, 2003 at 21:05:31, Robert Hyatt wrote: > >>I think the math is simple. If you believe that overclocking _can_ cause a >>failure, then doing it long enough is going to guarantee a failure. The only >>issue is whether it is a detected failure or not. Many are not. > >The odds of having a car crash each time you drive are probably greater than an >overclocking failure (detected or not) that may happen only one of a billion >times a certain instruction runs. That's not going to stop you from driving, >nor does it mean that driving long enough will guarantee a crash. An >overclocking failure isn't going to cause injury or death either. Further, a >car usually costs more than a computer. This is utter nonsense to someone doing the things I am responsible for here. IE can you imagine this scenario: A scientist approaches me and asks to use my cluster to do his genome pattern matching algorithm that is going to need a few months of computation time, even on this cluster. He runs it, and discovers _the_ specific protein that will correct a defect that causes some specific form of cancer. The drug company spends 500 million dollars to fabricate that particular set of proteins, and treats 200 test subjects who promptly die. And then the scientist says "damn, I was running on a machine that produced unreliable results. I have re-run the computations 5 times and the answers did not match the first run." We can't take that kind of chance here. What is a "cheap car crash" to you is potentially billions of dollars of "problem" here. Hence my firm stand on _reliability_. Run your hobby stuff all you want on hardware that _probably_ is reliable. But I'm going to run our software on hardware that _is_ as reliable as it can possibly be. > >So you gamble with your car, your life, and the life and property of others >every time you drive, in an activity where the odds of failure aren't very >favorable in the long run. However, you're afraid to gamble with what could be >a cheap piece of hardware, in an activity where nobody has the potential to get >hurt or die, and the odds of failure are middling at best.k See above. If you think "nobody has a chance to be hurt or die" at a large medical institution such as UAB" then you are not just mistaken, but _terribly_ mistaken. And even if someone doesn't die, tossing 1/2 billion dollars down the toilet is _not_ the way to "win friends and influence people." > >Now, in case my argument is misconstrued _again_, I'll try to clarify. I am not >advocating overclocking. I do not overclock myself. I am _only_ trying to >point out the logical fallacies in the arguments that overclocking is either >necessary for certain systems or that it is an absolutely unsafe activity. What is the probability of an overclocked CPU failing? I don't have a real idea. But it is > 0. Are you willing to flip a coin, knowing that heads you win, tails you win, but "edge" and you lose a billion dollars? I'm not. They once had a video at a Las Vegas casino showing such an "edge" occurrence. The coin went up, came down and hit and started spinning like a top. And on the very slick surface (Don't remember if it was glass or marble or whatever, I'm sure someone else saw it) it slowed down and stopped and remained on the "edge". Once is enough to convince me that I'm not going to take the risk. I'm going to use a coin without a flat edge, so that landing on it is really impossible. And let _me_ be clear too. I don't overclock. I have never overclocked. I will _never_ overclock, unless it is for fun and games, and not for something I am going to _depend_ on. I might do it for chess, unless I were playing Kasparov for a Million dollars.
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.