Central Processing Units

Why the processors cannot be interesting? Among typical units used in old PCs like Intel 2 or 386, some AMD and clones, there are some interesting chips with not so much known history. This page shows some of these chips.


386 and 486 clones

In first PC times, the development of CPUs was quite equal. Intel manufactured their 8088 and 86, while some other companies made their replacements which were comparable both in functions and speed. A small exception was Nec with their V20 and V30 processors with a bit better instruction set, but not all mainboards could have it and not all programs could use it. With introduction of 286, the situation changed a little - alternatively to Intel, mainboard could contain AMD, Harris or even Fujitsu chip, and sometimes these chips were able to be clocked at higher frequency (Harris 286 could work even at 25MHz!) or were more energy-efficient. On the right you can see different 286 chips in different packages.
Introduction of 386 processor changed the situation. While Intel decided to stop at 33MHz, forcing customers who wanted a faster processor to choose a 486 series, AMD went to 40MHz, making it one of the most popular processor in Taiwanese mainboards. Even some 33MHz Intel mainboards could run 40MHz AMD with some hardware tweaks. Thus, AMD processors were available both as a chip to solder into mainboard, a separate piece of silicon (photo on the right) or, in cheaper version, a soldering version on a board which converted it to a "socket" processor.

When Intel released 486, they allowed some companies to make CPUs on a license, as well as they started to sell licenses for making older chips. In early 1990s they sold a 386SX (a crippled 386) chip license to IBM, then IBM started to make their own chips exclusively for IBM computers - a 386SLC - a 386SX with power management features and on-chip 8kB of cache. These features made the SLC, which was in fact an accelerates 386SX, sometimes faster than Intel's 386DX of the same clock speed. The next CPU, called Blue Lightning, was more engineered to have a complete 32-bit support and was even sometimes called "the fastest 386".


AMD 386DX 40MHz processor in a protective enclosure
 

Later, around 1993,  when agreement with Intel expired, IBM made another agreement with Cyrix, making a 486-equivalent which was sold as "Blue Lightning" too. They had more in common to Cyrix 486DX2 than to earlier designs made with Intel. Both of these chips, IBM and Cyrix, can be seen in the picture below. The Cyrix chips were not clones of Intel chips, but were re-designed from scratch as functional replacements, so the speed can be a bit different. However, in 1993 Cyrix was accused by Intel for copying the CPU solutions - finally it was dismissed, but caused significant delays for Cyrix.

Different non-Intel CPUs from 386/486 era, as well as math coprocessors for 386.

From top-left to bottom-right:
1 - Intel 387 math co-processor, 20MHz
2 - ULSI 387 math coprocessor, 33MHz
3 - IBM Blue Lightning 50MHz 486-compatible
     (in fact Cyrix)

Bottom row:
4 - Cyrix Cx486DX2 50MHz 486-compatible
5 - UMC GREEN CPU, 40MHz,
     486-compatible
6 - ST486 DX2-66MHz 486-compatible

Remember these Cyrix accusations when we'll discuss another 486 alternative - a small Taiwanese manufacturer re-designed a 486 adding power management features and sold it as Green CPU model SUPER40 (bottom row, middle in the picture). There were versions corresponding to SX and DX CPUs, as well as 3.3V chips for e.g. notebooks or embedded systems, running at 25, 33 or 40MHz. And... Intel accused them of course, also in 1993, for the same thing as Cyrix. However, although the patent violations were exactly the same, it ended a different way - UMCs had to be sold outside USA, and it's clearly marked in the casing. Why? A typical Kafka's illogical case - Cyrix was manufacturing their 486s in companies, which previously paid Intel (by licenses), like ST (the last CPU in the picture) or Texas Instruments, while UMC did it with their own effort.
 


Evolution of Pentium form factor

Different Pentium processors:

Top row:
1 - Pentium 60MHz for Socket 4
2 - Socket 5/7 Pentium 75MHz
     With gold heat spreader
3 - Socket 7 Pentium 150MHz,
      one of the fastest without MMX

Bottom row:
3 200MHz MMX Pentiums in different
    packages:
 - Ceramic package
 - Plastic with metal heat spreader
 - Plastic with built-in heatsink and fan

The initial Pentium processor with "P5" core, used a 5V socket designed especially for it. Intel released 60 and 66MHz versions of Pentium processors for this socket, and switched to Socket 5, in which 3.3V processors could be used and pins were packed more densely. Starting from 75MHz, Pentiums were using Socket 5.
With Pentiums becoming more and more complex, Intel decided to add support for two voltages for one processor. It was introduced around 1995 with Socket 7, which was backwards compatible with Socket 5 CPUs if mainboard's regulation circuits allowed. It became even more useful with introduction of Pentium MMX in 1996.
During this development cycle, Pentiums changed their packages. First ones used only ceramics or had a gold heat spreader (early units). Later CPUs, without MMX, returned to ceramic packages. With introducing of MMX technology heat became a bigger problem, so only some units were made of ceramic - more used plastic packaging with metal heat spreader (in fact most mobile units) or had a special package with heatsink built-in, plastic or ceramic.
Interesting thing is a Pentium 300MHz, probably a fake one (overclocked 266MHz), can be seen in this collection.


Pentium alternatives

Although Intel kept their Pentium architecture in secret, there were alternatives. AMD released their K5 chips in 1996, which was a re-design closer to Pentium Pro. The idea was to use some fast RISC core and command translation logic, which was different than Pentium's approach. So because Pentium and K5 of the same clock had significantly different computational speed, AMD introduced a measure called PR - Performance Rating or "Pentium Rating". The chip with PR166 was working more-or-less as quick as Pentium 166MHz, regardless of its "real" frequency.

IBM Pentium replacements:
Top row:
1 - IBM 6x86 120MHz (PR150)
2 - Cyrix 6x86 133MHz (PR166)

Bottom row:
3 - IBM 6x86 133MHz (PR166)
4 - IBM 6x86 200MHz (PR233)

 

As it was noted in part 1, we left IBM with contract with Cyrix for making their CPUs. As Pentiums became more and more popular, Cyrix had their own Pentium-equivalent called 6x86 and they manufactured it also using IBM's plants. The 6x86 was in fact a deeply re-designed Pentium, so there was no code translation like in AMD's K5. Additionally, early units had compatibility and heat problems, so Cyrix put a "Heatsink/Fan required" text on their processors.

The speed in integer benchmarks was indeed slightly outperforming Pentiums. However, the engineers focused more on the branch prediction and I/O optimizations than on floating point unit which was much slower than Pentium's, comparable to Cyrix 486s (maybe they used just the same circuits?). They somewhat partially got out of problem marketing 6x86 CPUs as... processors for office applications, where math co-processor is used more rarely than in e.g. multimedia or CAD.
A tip of the moment: Quake shows reasonably well and smooth on 100MHz Pentiums. In 133MHz Cyrix frames are lots all time.


The Geode

In the picture on the right we can see an interesting processor: The Geode, made by National Semiconductor in mid-1990s. Although this chip can physically fit into Socket 7, it is not compatible on a software level. Why? Because there is almost a complete computer inside.
There is a Cyrix MediaGX-like core with MMX, 12kB of cache, memory interface, PCI controller, and a CS5530 module which is a video and sound chip. Most typical chipset functions are implemented in this CPU, so it needs a special mainboard and chipset. Being designed for embedded systems, Geode is also a low-power chip.
The only problem is that many of these features directly decrease the computational speed. 300MHz Geode is comparable to Pentium running at 200-233MHz. This chip is a 266MHz.
The Geodes found their place in embedded/control systems, thin clients or point-of-sale terminals. They were also used in some notebooks being frequently reported as "CyrixInstead" CPU (its ID).
The Geode line was made by Cyrix, then National Semiconductor, finally being taken by AMD in 2002.

Pentium II and III, and Slot 1

Slot 1 was an attempt made by Intel to make own CPU connector exclusively for own CPUs and introduce new features. Slot 1 is a slot for circuit board, in which CPU is soldered in, as well as cache chips. Simulatneously AMD introduced similar slot - Slot A. Slot CPUs could be used in SMP, so it was possible to use 2 CPUs in special mainboards. For a really high-end server purposes, a Slot 2 was used, an extended Slot which was targeted towards multi-processor systems.

Unfortunately, there were significant problems with these Slots. First of all, production was problematic - if a typical single single CPU failt factory tests, a CPU is discarded. If a Slot 1 CPU fails, a whole module is discarded - with cache chips which are good and cannot be re-used. Similarly if cache chips fail, a good CPU is discarded.
Another problem was related to mainboard connectors, mounting points and cooling - they were really different. One series of CPUs could be sold with different heatsink mount and cooling mounting systems. Heatsinks were mounted with screws, spring rings or spring levers, as well as with special clips. Because CPU stressed the connector, it was needed to secure the module in mainboard - and in many mainboards this mounting was too loose, causing the CPU module to rock in slot and cause instabilities.

A Slot 2 Pentium II Xeon (upper) and Slot 1 Pentium III (lower)
One of the highest frequency CPU for Slot 1 was Pentium III at 866MHz. In many mainboards uncertain connections caused it to be unstable even at 800MHz.

Pentium III Katmai with external cache (upper) and Coppermine (lower), with cooling (left) removed.
The first manufacturing problem has been solved with Pentium III and Coppermine core which contained cache memory in it, so installation of addiitonal chips was not required. However, increase or speed still caused stability problems in cheaper mainboards.

The last mainboards with Slot 1 were usually upgraded using "SlotKet" type adapters, giving Socket 370 on a Slot 1 board. It was then possible to use a Socket 370 CPU in Slot 1 mainboard. This CPU was usually a Celeron.
Not all SlotKets allowed to install all S370 CPUs, so different modificaiton had to be made (e.g. for Coppermine or Tualatin) depentent on adapter used. With these mods it still could fail with higher frequencies.

 

 

Links:

http://www.cpu-world.com/ - CPU World - the biggest source of information about old and new processors in the Internet.
http://www.cpu-collection.de/ - Collection of various CPUs