Intel Forced to Remove "Cripple AMD" Function from Compiler?
in [Chips]
|
Prev: No multicore grahics from Intel, Intel withdraws Larrabee!
Next: Several Ntoskrnl BSOD crashes over the last week, but different stopcodes
From: Robert Myers on 13 Jan 2010 19:28 On Jan 13, 6:32 pm, Bill Davidsen <david...(a)tmr.com> wrote: > Robert Myers wrote: > > Moral thinking is culturally-bound. An astonishing fraction of > > Americans pay lip service to the (false) notion of universal moral > > absolutes, but, by and large, the moral landscape of the US is > > pragmatic, not absolute. Introducing absolute moral considerations > > into a discussion about commerce as if they belonged there isn't > > unheard of in this culture, but it is largely foreign to it. The > > question is not: would God approve, but, was there a contract, real or > > implied, and was it fulfilled or not. In this case, Intel created an > > implied contract by the promises it made for its compiler and it > > failed to fulfill the terms of the contract. Happens every day in > > commercial transactions large and small. End of story. > > You mean like it's immoral for a home owner to walk away from an underwater > mortgage, but not immoral for the bank to sell the mortgage knowing it will > enter foreclosure? Just so. People will claim that whatever is to their advantage, or, in Yousuf's case, what suits his personal preferences, is what's moral. Robert.
From: chrisv on 20 Jan 2010 10:06 Robert Myers wrote: >People who wanted to use AMD products because they were clearly >superior for some applications didn't use icc because it wasn't the >best compiler for those purposes. Err. Most people don't use compilers. They buy their software already compiled.
From: Bill Davidsen on 20 Jan 2010 13:31 chrisv wrote: > Robert Myers wrote: > >> People who wanted to use AMD products because they were clearly >> superior for some applications didn't use icc because it wasn't the >> best compiler for those purposes. > > Err. Most people don't use compilers. They buy their software > already compiled. > With what, if not a compiler? The vendor buys the compiler, the application runs slow on AMD.
From: chrisv on 21 Jan 2010 08:55 Bill Davidsen wrote: >chrisv wrote: >> Robert Myers wrote: >> >>> People who wanted to use AMD products because they were clearly >>> superior for some applications didn't use icc because it wasn't the >>> best compiler for those purposes. >> >> Err. Most people don't use compilers. They buy their software >> already compiled. >> >With what, if not a compiler? The vendor buys the compiler, the application runs >slow on AMD. Really?
From: Sebastian Kaliszewski on 26 Jan 2010 07:32
Robert Myers wrote: > The general problem here is that you fall into the same trap about > probability and statistics that nearly everyone does, Nope. It;s just your illogical projection of your own thinking. > and it's why the > idea of economic "science" is so appalling. The idea of computer > "science" is nearly as appalling. Nonsense. > > Science relies on the ability to reproduce results by way of > experimentation. If an experiment fails to reproduce a result, it > might be because one or the other of the experiments was flawed in > some way. You can examine the apparatus and the data. You can try > again. Others can try again. Software can be reproducibly experimemted upon. Probablities of various failures are based on real data. > > In economics, you can do no such thing. *Maybe* your clients did well > because they have better methods, or maybe they were just lucky. > There is absolutely no way to tell, and no experiment you can perform > to prove or disprove either hypothesis, because conditions in > economics are never going to be the same way twice on a macro scale. That only shows you what you don't understand. The problem with you is that you'd like to base everything on scientifically rigorous experimentaiton, and what is not rigorously scientifically proven you disregard as nonexistant. But the fact that a thing is not rigorously proven does not mean it does not exist. So *there is* a way to tell. One can not be 100% sure the explanation is right, but one can be sure, the explanation is more probable to be right than wrong. Besides, typically *scientifically proven* means just statistical confidence level 0.95. Thus not insignificant portion of "scientific facts" is false, being just a result of (un)lucky experiment. Poor Robert, even science doesn't bring a 100% sure proof... > If you had said that you understood probability and were confident of > your own methods of estimating risk, I've have had to admit that it's > you're field, you're closer to the action than I am, and you have to > live with the consequences of your own misjudgments. > > Instead, you referred to the judgments of "experts"--your customers in > finance. The entire world of financial engineering and economic > "science" is on the defensive right now because the entire methodology > of estimating risks and placing financial bets failed on a grand > scale. Yet their risk estimation of "ordinary failures" does work. And that's what I've sated. Nothing more. What didn't work was risk extrapolation based on systematically flawed data (data presented only short, and in significant portions of time unrepresentative). > > Even in catastrophe, there will be some winners. There will always be > winners and losers in finance, but the fact that one group happened to > be a winner in some particular situation proves nothing. That the > entire system nearly collapsed and *didn't* collapse only because of > extraordinary and morally questionable intervention is another > matter. In fact such intervention is quite often assumed for similar situtaion. I.e. management of such big istitutions knows they are too large to fall. I.e. they assume they will be not allowed to fall. > In retrospect, as with a collapsed building, it's easy to see > the structural flaws, which is a different matter from bad luck. > > The point is that the methods of risk estimation failed to identify > the structural flaws and that that failure itself appears to be > structural. > > The general problem is that you can never know if you left something > out. That's true even in laboratory science, but if a completely > different group sets up another experiment and gets nearly the same > result, you have some confidence that you have science and not mere > coincidence. > > Climate "science" has the same problem. We can't perform experiments > with weather and climate (at least not yet). Yet people are forever > looking at irreproducible subsets of data and claiming to draw > conclusions from them, just as you are looking at an irreproducible > subset of data and claiming to be able to draw conclusions from it. You can look at historical data, and check if your prediction based on (for example) 20 year old data predicts situation from just the last year. Thus short term (climate) models can be created (and are). Mid and long term are different story (mid term in case of climate is centuries) as there is no precise data (but still there is some indirect data but of much worse qiality, noisy and suspectible to systematic error). [...] >> You have little grasp of reality. >> >> The wordld of software is as is because what you advocate is simply not >> effective. It's plain simply better instead of spending money to bring >> software error levels to your liking to spend them on something >> productive. It only pays for itself in places where cost of failure is >> extreme. >> >> For example theft is at a significat levels in supermarts, yet >> increasing security is simply more expensive than dealing with the fact >> that some percentage of goods will get stolen. Similarily bank robberies >> are pretty frequent in Europe (in major cities you have one bank robbery >> every 5 days) yet as average amount stolen is less that one year wage of >> a security person, it simply does not pay to increase security. >> > The analogy is not a good one. If someone takes a loaf of bread and > doesn't pay for it, your entire loss is the cost of putting that loaf > of bread on the shelves. If someone cracks your credit card > processor's security, the potential loss is simply incalculable. Of course it is calculable. First you can't loose more that you have, then that processor can't process more that it's bandwidth allows. Then any self respecting bank will have alarm bells setting off if some credit card processor usage rate went up, etc... > > The problem with software reliability, just as with placing financial > bets, is that there is no way of foreseeing or even bounding the > possible costs of a mistake. Nonsense. If your transaction system is inoperable for one day because of failure your losses can be well calculated. If it was designed right then suddent failure taking longer than a few days and not caused by "act of God" (disaster, war, etc) is so improbable, thet there are simply other, more serious risks, like too many bank clients who want order some operation and institution simply unable to handle all the traffic. But even in case of total failure which can't get recovered for a month, which would cause practically all the clients flee, the maximum loss is still bounded. One bank of a hundreds would end it's existence. But pleas show me such an example: a bank killed by it's software becoming inoperational due to bug. What you advocate is that city of London had (on a payroll) a snow cleaning brigade with all the needed equipment (which must be maitained) just because once every few years there is snow there (like this year). > That you continue to insist you can > place bounds on the possible losses exposes the shallowness of your > thinking. Nope. It's your insisstence on declaring losses unbounded what shows you simply don't know what you're talking about. > > People in the business whom I respect say things similar to what > you're saying, but they say it with regret and humility: we get crappy > software because no one is willing to pay for good software. And no one is willing to pay for yopur "good" software because current "bad" software is good enough, and it works just now and is cheaper that that "good software" would ever be if it existed. And that's all because losses are bounded, and not having software at all would mean that lost gains would far surpass losses caused by that "bad" software we have today. \SK -- "Never underestimate the power of human stupidity" -- L. Lang -- http://www.tajga.org -- (some photos from my travels) |