Gregory's ServerIn 1989, Intel introduced the 486 processor, improving the performance of the 32-bit 386. Unlike the 386, the 486 has an on-chip 8 KB cache and a floating-point unit. The 486 has over a million transistors.
I took this die photo and labeled it with the main functional blocks.
 24 comments
@root42 You have the 486 DX2, an improved chip. One feature on your die that isn't on mine is the clock-doubler PLL (phase-locked loop).  @root42 The microcode is in the same place (accounting for the rotation of your photo). However, there are metal power wires over it in the DX2, so it isn't as visible. @root42 Here's a closeup of the 486 DX2 that makes the microcode ROM more visible. (The layout is identical to your photo.) The vertical yellow metal lines on top of the ROM make it harder to distinguish the ROM, especially with the coloring of your photo. @kenshirriff yes once you know that those structures are metal lines the underlying ROM becomes obvious. Thanks. The chip I photographed is the i486 SX, which is the version of the 486 without a floating point unit. But you can see from the photo that it has a floating point unit on the die. It's unclear if Intel deliberately disabled the floating point unit to segment the market between the low-end SX and the high-end DX, or if chips with an FPU fault were sold as the SX. One of the engineers on the 486 was Pat Gelsinger, who is now CEO of Intel. You can see his initials on the die. The 486 processor uses microcode. Here's a closeup of the microcode ROM; you can see the bits.
[DATA EXPUNGED]
@awoodland The Pentium Pro was the first x86 processor with microcode that could be updated. Intel originally implemented that for testing purposes, but realized after the expensive Pentium FDIV bug that microcode updates were an important feature.  @kenshirriff I remember some beyond useless support person at Intel telling us at VMware that the Pentium did not use microcode when we were trying to find out about a bug that appeared to have a microcode fix that some but not all OEMs had released. OK well we've reversed engineered part of the update for the microcode your processors don't have, so maybe we could talk to somebody more senior?... (eventually got to senior engineering folks who were helpful) A closeup of the edge of the 486 showing bond pads with bond wires attached. Note the complex I/O driver block next to each bond pad. The pads in the center (without drivers) are two of the power and ground supplies. Note the thick vertical lines distributing power/ground to the chip. Here's a Google Drive link to a higher resolution version: https://drive.google.com/file/d/1nQMtDGMQN8W34I486jtm3o4z527TU65V/view?usp=drive_link  @gsuberland It was supposed to be 7155x10870 :( @kenshirriff ah yeah your server will have resized it then. weird final resolution though, I guess it limited based on compressed JPEG size?  @kenshirriff @drwho  @kenshirriff damn this made me find out that CPU cores are a lot smaller than I thought and occupy less than half of the entire thing, both in modern CPUs and I assume in this one if the decode logic part is what would be described as a CPU core (I don't know that much about computer science) @kenshirriff Are more details coming? The part labeled FPU is surprisingly small compared to the parts labeled data path (control). How much was microcoded? @nh I'm not planning to look at the 486 in detail. The FPU is controlled by a "sequencer PLA" for basic operations, with complex operations using the same microcode ROM as the integer operations. I'm not sure why the integer datapath is so much larger than the floating point datapath, maybe the numerous integer registers and the barrel shifter. |
@kenshirriff this hangs on our guest bathroom wall. :)