Doug Siebert (foo@bar.bar) on 5/13/07 wrote:
---------------------------
>
>Somehow I can forsee the discussions in lkml in a few years when the first person
>suggests a mass rename of the x86 arch to x86-32, followed immediately with a discussion
>about whether after that the x86-64 arch should be renamed as x86 :)

As it seems that far too many see this as a dogmatic issue, I would like to pursue it using numbers, as best we can determine them - all references to morons and fools aside...

I don't know all of the numbers, and don't really have time to look them all up so anyone who wishes to provide actual information would be greatly appreciated.

1) The group I was referring to were desktop users, not commercial servers, etc. We can include developers and other 'power users' in the 'desktop users' group, of course.

2) A64 was released around the end of Sept. 2003, IIRC. That is about 3.5 years.

3) Intel released a desktop 64-bit processor in early 2005, IIRC. That is about 2 years.

4) I would guess that the average lifespan of a computer is 5 years, with some people waiting longer to upgrade. I think it would be fair to presume that about 50% to 60% of all computers in use have been upgraded in this time period.

5) AMD has had between 15% and 25% market share during the past 4 years, with a fair percent of that in servers. Desktops have been about 15%, I believe. In addition, not all processors manufactured/sold by them in the past 3.5 years have been 64-bit.

6) I don't know what Intel's mix of 64-bit desktop processors is, but for argument's sake I am presuming no more than 15% to 20% up until fairly recently. If someone has better figures, please present them.

7) If we therefore presume that less than 50% of all systems sold (probably more like 25% or 30%) during the past 3.5 years were 64-bit, and 50% to 60% of all desktops in use were upgraded during this period - that means that 64-bit systems would likely be somewhere between 15% and 20% of the entire market today. Which corresponds neatly with the number I indicated I estimated using our server logs.

8) Most systems under $600 or $700 come with less than 1GB of memory standard. Linus' argument is that 'most' systems are sold with at least 1GB of memory, but I am skeptical of that. I believe that the greatest volume of systems sold will be under $700. These likely don't all come with 64-bit processors, and very likely don't come with a 64-bit OS. Even those with 1GB may not come with one or both of those.

9) The argument has always been that 64-bits is 'free'. Well, I would suggest that if cheap systems have 32-bit processors and OS', while more expensive systems have 64-bit processors and OS' - then it is not free. It doesn't matter *why* it is more, it is a simple fact that it *is*. This, in turn, discourages a certain percentage of people from buying a 64-bit capable system. That will obviously change over time, and possibly a short time - but it is true *today*

10) Since my original skepticism was regarding the assertion that within 'a few years' (perhaps 4 or less?), 64-bits would be so common as to be the default. I don't think you can consider it such until it has *more* than 50% of the *installed* market, and more reasonably 70% or more. If my guesses are close, I think that we won't see such widespread use of 64-bit systems for at least 5 years or more.

11) The vast majority of users only use such applications as browsers, email clients, picture viewers, music players, word processors, tax software, and spreadsheets. Very few are editing pictures (other than doing simple cropping and resizing), and probably won't ever. Even most games that *average* people use are not memory intensive. There is a chance that within 5 years many, if not most, of those users will be doing most of their tasks on a handheld device, and will see little need to upgrade their computer to anything 'fancy' (again - likely to buy low-end systems). Many may just keep their old systems as long as they continue to function.

12) It makes sense that manufacturers of hardware and software would *want* everyone to be using 64-bit systems - because they don't *want* to have two versions of things. It is expensive! So I can fully understand why people in such positions would wave their hands around and say it is *required*.

13) 64-bit drivers are *still* not widely available for many devices, or available at all. If 64-bits was *that* widespread, these devices woult *have* to have 64-bit drivers available. But they don't.

Therefore, the facts seem to indicate that 64-bit systems are still not widely being used, at least not to their full potential (meaning, even if the hardware is capable, the software being used is not).

The fact that there are still *many* systems being sold with 32-bit processors and OS' means that the 'crossover' likely won't occur for more than 3 or 4 years unless something occurs that forces existing users to upgrade to a 64-bit system. My guess is that most people who are hanging on to their systems as long as possible will probably buy the cheapest one they can find to replace it - after all, an old high-end system probably performs worse than even the cheapest low-end systems available (which are likely 32-bit).

I would appreciate it if someone could provide actual numbers (or better ones, at least) that would lead to a different conclusion. I'm a reasonable person, so reason usually works well when trying to convince me.

Now, let me address a couple of nice fallacies that Linus has engaged in (and does so regularly).

1) Prejudicial Language - this is a fallacy whereby one is 'encouraged' to accept the argument because, otherwise they are 'stupid' or 'idiots'. That, of course, does not prove anything except that the speaker doesn't want a discussion, but just wants to silence everyone else. After all, who wants to be considered a fool or an idiot, so who would disagree after such a statement? That is a tactic that I would think should be beneath even Linus. (see how easy it is to use such fallacious arguments).

2) Equivocation - This is where the same word is used in two different ways. For example, I might say "...end users don't require 64-bits". Linus will say "...developers and commercial apps *need* 64-bit addressing, so it *is* required". In the first case, it means that users don't need it to do their typical computing tasks. The second case means that manufacturers must offer a 64-bit model to satisfy *some* users... but it has nothing to do with the first argument.

I'm sure I could find others, but those were the ones that stood out in my mind...

Regards,
Dean