From: walk@mrcnext.cso.uiuc.edu (Todd Walk) Subject: Re: 486SXs as Unix Iron? Date: Mon, 2 Nov 1992 20:49:39 GMT
gary@sci34hub.sci.com (Gary Heston) writes:
>No. They come off the same line in the same fab; they are laser modified
>(I think it's with a laser; probably not a mechanical device) to disable
>the math coprocessor. The die for the 486DX, 486SX, and 487SX are
>identical other than the pre-package mods. (That's right; the 487SX is
>a 486DX chip hacked to disable the CPU and cache....) While the yield
>for 486 dice is certainly not as high as for, say, 1Mb DRAMs, they've
>had a few years to get the bugs out of their process, so there's
>probably a reasonably high yield. (Intel, like most semi manufacturers,
>won't tell what their yield on anything is; that's closely guarded
>competitive information.) The wafers are scored, the individual dice
>are broken out, and they are then probe-tested. If it fails any part
>of the test for the lowest speed grading, it's recycled. If it passes,
>it's screened at various increasing speeds, until it fails; then, it's
>allocated to whatever speed class it passed reliably. (This is a fairly
>generic process; DRAMs go through the same thing.) A lot of chips are
>only tested high enough to meet production requirements; if they've
>already got as many 50MHz chips as they want to ship, the remainder
>will only be tested to 33 (or whatever).
>The only place a 486SX has any use is if you'll *never* want the FPU,
>and need to save a few $$. Otherwise, it's a pointless buy. Considering
>that it has an identical processing effort, and is picked out at a lower
>cost rate, all it does is tell you how much you're being overcharged for
>a 486DX.
>(No, I don't have access to any Intel confidential data. The above
>can be proven by simply getting one of each chip, preferrably already
>failed, and popping off the metal cap on the underside, and has been
>done by a few magazines.)
From what I have heard, this is what intel USED to do. Now they
have a seperate line for 486SX's in order to get a higher yield rate
(a yield rate/cost per chip almost as good as the 386's). Also testing
is done on the highest MHz speed 1st and if it fails the test it's
tested at lower levels. If there are left-over 50MHz chips, and
more 33MHz chips are needed, then all you have to do is label the
50MHz chips 33MHz.
Todd Walk
walk@mrcnext.cso.uiuc.edu