Bill Gates Six Forty Kbytes Quote

Question: Did BillGates ever really say that "640k is all anyone will ever need"? Does anyone have the time and place for this quote? And what was the context?

That's three questions :) But no, he didn't. See http://web.archive.org/web/19961106040603/http://htimes.com/htimes/today/access/oldfiles/gates23.html for an example of Bill denying he ever said it. (Although, later down the page, he says "Probably the fastest conventional telephone dial-up modem you'll ever have is 28.8."! Tell that to all the people using 56k modems....)

[on modems...]'

In Bill's defense here; much of the information theory of the time (ShannonsTheorem?, etc.) suggested that around 30k was indeed a theoretical limit, given known properties of the telephone network's signal integrity. What allowed the end run around Shannon (and my memory was a bit hazy here) was a) the use of data compression in modems, and b) the use of adaptive modulation/demodulation technologies that can detect the quality of the phone line, set the data rate appropriately, and exhibit GracefulDegradation. Many locales with crappy phone lines cannot get 56k data rates (uncompressed) no matter what they do; 56k is an upper limit, not a guarantee.

And some of the design decisions are questionable--the combination of adaptive signal equalizing techniques with data compression makes the latency of 56k modems just terrible. (See ItsTheLatencyStupid). Web access (and virtually any small-packet TCP/IP with lots of round trips) suffers tremendously as a result.

Be that as it may, 33.6 is still the basic standard for modem speed minus all the smoke and mirrors involved with 56k. But he wasn't far off the mark, unlike the 640K quote attributed to him. I suspect it was probably an IBM sales engineer who said it, thus turning understandable cost-savings into a crummy design principle.

The way the "56k" standard works is to move one of the modems into the phone network itself, the digitizers on the Telco end of your local loop get their bits directly from the "modem" on the other end. It's not analog. Furthermore, the 56k is only in one direction. Give Bill some slack on this one, he was right. (He's still evil, though) -- MikeWarot

The original IBM PC could have avoided the "640K of RAM" limit by avoiding memory-mapped video and device drivers -- a difficult approach, or by swapping memory-mapped video RAM and driver ROMs into and out of the CPU's address space as needed -- an approach that can be complex, somewhat costly, and certainly bug-prone.

Even if they had gone out of their way to avoid imposing a "640K of RAM" limit, they would still have hit the limit at 1024K of RAM -- the total address space of an Intel 8088/8086 CPU. So they made a tradeoff of maximum RAM, to simplify access to video RAM and driver software in ROMs.

Cue obligatory rant of how the world would have been a better place if IBM weren't such cheapskates, and had put a MotorolaSixtyEightKay processor in the PC instead of the horrid 8086/8088.
Don't get me started!!!! SIGH. (or the beautiful pdp-11 workalike by...I forget who...western digital??...or National's 16000...) Intel + IBM did an unimaginable amount of harm. The Pentium is still register-starved, as I said on another page last week.
Then again, Motorola shot themselves in the foot with the bad design of the 68k MMU, and with the lack of recoverable interrupts until the 68020, making paging and fully protected processes problematic-to-impossible until then...that also held back the rest of the industry that was trying to do the right thing.
- While Motorola isn't in the same league as XeroxCorporation when it comes to this sort of thing...they do a damn fine job of shooting themselves in the foot every now and then. (Anyone remember the 96k DSP?)

CategoryQuote