Alasir Enterprises
Home >  Main page >  x86 CPU Reference (part 1)  
x86 CPU Reference

This document is intended to provide some reference information on x86-compatible processors. I try to keep it as accurate as possible, but I make no claims it's complete or perfect.

Almost no attention has been paid to embedded or mobile processors, as well as Intel OverDrives.

Those CPUs labeled with Pentium Ratings (PRs) were listed under their real frequencies; it might be a little harder to recognise a particular model this way, but overall provides with a more clear picture.

If you have a more detailed or even correct information, please drop me a note.

You're free to use this reference for your needs, just give me a credit!

Special thanks:

 

16 BITS

1.1. Intel i8086

In old good days of 1978 Intel introduced i8086 -- a general-purpose processor capable of being used in so-called "personal computers". It operated with a "brand new" logic, had a 16-bit memory bus and a 20-bit address bus, thus allowing to work with 1Mb of memory mapped into 64Kb segments (16-bit offsets). However, both buses were multiplexed (i.e. only one bus could be active at any moment given).

IBM, a strategic partner of Intel, soon started manufacturing PCs based on this CPU. Several manufacturers (Compaq, Columbia, Kayro, etc.) also built machines with i8086 inside, and claimed them to be "IBM-compatible" (however, their BIOSes weren't totally compatible with IBM's one).

About 14.9 million units were sold (including clones).

1.2. Intel i8088

Later Intel introduced a stripped version of i8086, with a 8-bit memory bus, and called it i8088. Since i8086 was priced somewhat highly, there was a significant demand in a cheaper product, and that was i8088 to stand for.

This CPU was widely and successfully used in IBM PC/XT (Personal Computer/eXtended Technology) and compatibles in early 1980s (in a matter of fact, first commercially-available PC of x86 architecture was built by IBM using i8088, in 1981).

About 20.5 million units were sold (including clones).

1.3. Intel i80186 and i80188

These chips were successors to i8086\i8088, but weren't used widely in PCs because of their all-in-one nature. The core logic was updated with all the real mode instructions of future i80286, for more convenient low-level programming; was added a fault protection algorithm (original i8086\i8088 would simply hang upon executing an unknown instruction).

These chips included a number of peripherals that were usually located onto a motherboard:

  • 2 programmable Direct Memory Access (DMA) controllers;
  • Programmable Interrupt Controller (PIC);
  • interval timers, clock generator, etc.

But, these chips achieved a great success in the world of embedded processors, and were used in various electronic devices (portables, hard disk controllers, modems, etc.). For example, US Robotics used 20MHz and 25MHz versions of i80186 in Courier V.Everything modems, in late 1990s.

1.4. Intel i80286

i80286 expanded possibilities of i8086\i8088 by implementing some new features: it could operate both in the real mode (to keep compatibility with i8086), and the protected mode, addressing up to 16Mb of the physical memory because of the increased address bus width (now 24-bit). New instructions were added into the logic, mostly to drive the protected mode.

The first commercially available machine built with i80286 was from IBM PC/AT (Advanced Technology) line, in August of 1984.

About 36.8 million units were sold (including clones).

1.5. Clones

In early 1980s Intel's manufacturing facilities weren't capable to satisfy rapidly growing market demand on x86-compatible processors. On the other hand, Intel didn't want computer building companies to use CPUs of different architectures than x86 (mostly, Motorola 68K); thereby, Intel licensed manufacturing rights to several semiconductor companies such as AMD, Harris, Hitachi, Siemens, IBM, and probably others. Their CPUs were exact clones to i8086\i8088, i80186\i80188 and i80286:

  • AMD Am8086\Am80186, Am8088\Am80188 and Am80286;
  • Harris HS8086\883, HS8088\883 and HS80286;
  • Siemens SAB8086\SAB80186, SAB8088\SAB80188 and SAB80286;
  • Hitachi H8086\H8088;
  • IBM, Fujitsu, Kruger, etc.

1.6. NEC V30 and V20

These chips were first non-licensed clones of Intel i8086 (V30) and i8088 (V20). Intel sued NEC, claiming that it violated the copyright on i8086 and i8088, but lost finally: verdict said that the CPU microcode could be copyrighted, but Intel forfeited all rights because hadn't labeled the processors as such.

There were several modifications of these chips (V20H\V30H, V25\V35, V40\V50, V45\V55). These CPUs included the full i80186 instruction set, and performed about 20% faster i8086\i8088 when running at the same clock speed. Also, they could run in the native Z80 mode (i.e. to execute directly software written for Zilog Z80 CPUs).

1.7. Chips & Technologies F8680

This chip was targeted to the notebook and subnotebook market. It consisted of an i80186-compatible CPU, a CGA-compatible display adapter, a universal asynchronous receiver and transimitter, and a PCMCIA circuitry. Due to the optimised internal architecture, it offered performance comparable to i80286.


Technical datasheet
CPU Speeds Released Mem. bus Addr. bus Trans. Tech. Voltage Instr. set
 i8086  4.77 to 12MHz May,1978 16-bit 20-bit 29.3 th. 2.0µ 5V i86
 i8088  4.77 to 12MHz Feb,1979 8-bit 20-bit 29.3 th. 2.0µ 5V i86
 i80186  6 to 25MHz 1982 16-bit 20-bit ? ? 3V, 5V i186
 i80286  6 to 20MHz Feb,1982 16-bit 24-bit 134.3 th. 1.5µ 5V i286
 V30  8 to 16MHz Mar,1984 16-bit 20-bit 63 th. 1.2µ 5V i186, Z80
 V20  8 to 16MHz Mar,1984 8-bit 20-bit 63 th. 1.2µ 5V i186, Z80
 F8680  8 to 14MHz Oct,1991 16-bit 26-bit 150 th. 0.8µ ? i186

 

32 BITS: NON-SCALAR

2.1. Intel i386DX

i386DX (Double word eXternal, about memory bus) was designed with the 32-bit architecture in mind: a 32-bit non-pipelined IEU (IADD, IMUL, IDIV) with a 64-bit barrel shifter and eight 32-bit general purpose registers, a 32-bit AGU (load\store), and a 32-bit instruction decoder\fetcher. Since the core didn't contain a FPU, a mathprocessor (i387 or compatible) could be installed separately. As of the memory management, in addition to a segmentation unit (like in the previous models) it also incorporated a paging unit, allowing manipulations of the physical address space using 4Kb pages. Was interfaced to two independent 32-bit buses, the memory and the address. When in the protected mode, it could address up to 4Gb of the physical memory. It also supported a protected mode extention called Virtual 8086 (V86), for better multitasking (to set up virtual areas in memory). At the same time, i386DX was completely backwards compatible, being able to execute any 16-bit code.

Interesting, but the first available commercially machine with i386DX inside was manufactured by Compaq, not IBM.

About 15.5 million units were sold.
i386 architecture

2.2. Intel i386SX

i386SX (Single word eXternal, about the memory bus) was marketed as a mainstream CPU. It had the same core compared to i386DX, but the width of both memory and address buses was reduced (to 16-bit and 24-bit, respectively).

About 21.9 million units were sold.

2.3. Intel i386SL

The low-power version of i386SX, with both cache and memory controllers integrated. Designed exclusively for notebooks and other portables. Intel's second attempt to produce an all-in-one chip since i80186.

About 1.8 million units were sold.

2.4. AMD 386 series

In the middle of 1980s AMD launched its own microprocessor department, and began designing a 386-class processor. It took several years to release Am386DX, an alternative to i386DX. However, there were some legal issues, because AMD used Intel's microcode without a license.

2.4.1. Am386DX, Am386DXL and Am386DXLV

Being instruction- and pin-compatible to i386DX, and offering approximately equal performance on equal clock speeds, Am386DX was significantly less priced.

Am386DXL was a low-power, and Am386DXLV was a low-power and low-voltage version of Am386DX.

About 4.3 million units were sold.

2.4.2. AMD Am386SX, Am386SXL and Am386SXLV

Stripped variants of Am386DX series, like i386SX.

About 5.7 million units were sold.

2.5. IBM 386 series

IBM manufactured x86-family CPUs due to a joint-development agreement with Intel, which allowed IBM to produce proprietary cores based upon Intel's ones by optimising some frequently-used instructions, adding new, etc. So, all these CPUs were based on the i386 core. IBM also manufactured Cyrix CPUs, since September of 1993.

2.5.1. IBM 386SLC

i386SX pin-compatible. The low-power CPU core contained 8Kb cache, and there was only one model capable of running at any of three different speeds, or even in asynchronous mode. It was offered as an upgrade option for IBM PS/2 Model 56 and 57 systems.

2.5.2. IBM 486SLC and 486SLC2

i386SX bus interface, low voltage (3.3V). 486SLC2 ran twice the bus speed (clock-doubled). Both included 16Kb cache, largest among all 386-class CPUs. Upgrade options for IBM PS/2 Model 56 and 57 systems.

2.6. Cyrix 386 series

In early 1990s Cyrix Corp. (from Richardson, TX) entered the microprocessor market with both proprietarly-designed 386-class CPU core and microcode. Cyrix's 386 CPUs were marketed as "486", but in fact they were just enhanced with the 486 command set and the little built-in cache. Later appeared that the microcode wasn't completely compatible with Intel's one, and that caused numerous software incompatibilities.

Since Cyrix had no manufacturing facilities of its own, so all processors were produced on semiconductor factories of Texas Instruments and SGS-Thomson. Instead of fees, certain amounts of CPUs were sold directly by the manufacturers, and under their brands. IBM started producing Cyrix's chips since September of 1993.

2.6.1. Cyrix Cx486DLC and Cx486SLC

i386DX and i386SX pin-compatible, respectively.

About 200 thousand 486DLCs were sold, and about 75 thousand 486SLCs.

2.6.2. Cyrix Cx486SLC/e and Cx486SLC/e-V

Low-power versions of Cx486SLC. Cx486SLC/e-V also featured low voltage (3.3V).

2.6.3. Cyrix Cx486DLC2 and Cx486SLC2

Clock-doubled versions of Cx486DLC and Cx486SLC, respectively.

2.6.4. Cyrix Cx486DRx and Cx486SRx

For upgrade of surface-mounted 386DX or SX CPUs, respectively. Should be mounted over an existing chip, which will be disabled by using the FLOAT pin.

2.6.5. Cyrix Cx486DRx2 and Cx486SRx2

Clock-doubled versions of Cx486DRx and Cx486SRx, respectively.

2.6.6. Cyrix Cx486DRu, Cx486SRu, Cx486DRu2 and Cx486SRu2

Direct replacement upgrades. Provide the same that Cx486DRx (SRx, Drx2, Srx2, respectively), but for the regular PGA socket.

2.7. Chips & Technologies 386 series

Several 386-class CPUs were manufactured by Chips & Technologies, using the proprietary microcode:

  • 38600DX -- i386DX pin-compatible;
  • 38605DX -- i386DX pin-incompatible, with 512-byte instruction cache;
  • 38600SX -- i386SX pin-compatible, never released.

2.8. Texas Instruments 386 series

Texas Instruments manufactured proprietary Potomac series 386-class CPUs based upon the core from Cyrix but with some enhancements, as well as true Cyrix ones. Regardless of their "486" nomenclature, they had a non-pipelined 386 core.

  • TI486SXL-S;
  • TI486SXL-VS -- low-voltage (3.3V);
  • TI486SXL2-S -- clock-doubled;
  • TI486SXL2-VS -- low-voltage clock-doubled;
  • TI486SXLC;
  • TI486SXLC-V -- low-voltage (3.3V);
  • TI486SXLC2 -- clock-doubled ;
  • TI486SXLC2-V -- low-voltage clock-doubled.

Technical datasheet
CPU Speeds Released Cache Mem. bus Addr. bus Trans. Tech. Voltage Instr. set
i386DX 16 to 33MHz 1 Oct,1985 none 32-bit 32-bit 275.3 th. 1.2µ or 0.8µ 5V i386
i386SX 16 to 33MHz Jun,1988 none 16-bit 24-bit 275.3 th. 1.2µ or 0.8µ 5V i386
i386SL 16 to 33MHz Oct,1990 none 16-bit 24-bit 855 th. 0.8µ 3.3V, 5V 2 i386
Am386DX Am386DXL Am386DXLV 16 to 40MHz (DX) 20 to 40MHz (DXL) 25 to 33MHz (DXLV) Mar,1991 (DX,DXL); Oct,1991 (DXLV) none 32-bit 32-bit 161 th. 0.8µ 5V; 3.3V (DXLV) i386
Am386SX Am386SXL Am386SXLV 16 to 40MHz (SX) 20 to 40MHz (SXL) 20 to 33MHz (SXLV) Jul,1991 (SX,SXL); Oct,1991 (SXLV) none 16-bit 24-bit 161 th. 0.8µ 5V; 3.3V (SXLV) i386
IBM 386SLC 16 to 25MHz Oct,1991 8K WT 16-bit 24-bit 0.8 mil. 1.2µ 3.3V i386
IBM 486SLC 16 to 25MHz ? 16K WT 4-way 32-bit 32-bit 1.35 mln. ? 3.3V i486
Cx486SLC 20 to 33MHz Apr,1992 1K WT 2-way 16-bit 24-bit 0.6 mil. 0.8µ 5V i486
Cx486DLC 25 to 40MHz Jun,1992 1K WT 2-way 32-bit 32-bit 0.6 mil. 0.8µ 5V i486
IBM 486SLC2 32(16) to 60(30)MHz Sep,1992 16K WT 4-way 16-bit 24-bit 1.35 mln. 0.7µ 3.3V i486
Cx486SLC/e 25 and 33MHz Dec,1992 1K WT 2-way 16-bit 24-bit 0.6 mil. 0.8µ 5V i486
Cx486SLC/e-V 20 and 25MHz Dec,1992 1K WT 2-way 16-bit 24-bit 0.6 mil. 0.8µ 3.3V i486
C&T 38600DX 33 and 40MHz ? none 32-bit 32-bit ? ? 5V i386
TI486SXL* ? Feb,1994 8K WT 2-way 16-bit 24-bit ? ? 5V, 3.3V i486

1. First 16MHz i386DX had a clock trouble and were sold as 12MHz.

2. 16 and 20MHz i386SL required 3.3V while 25 and 33MHz -- 5V.

 

32 BITS: SCALAR

3.1. Intel i486DX (P4)

i486DX was the first x86-family processor with both FPU and cache (8Kb) integrated into the core. Also it was the first scalar processor, able to execute most integer instructions in the single-clock mode (1 command per 1 clock tick), because of the 5-stage integer pipeline driven by certain RISC techniques. All these improvements resulted in about 100% integer performance increase compared to equally-clocked i386DX. However, the built-in FPU remained non-pipelined.

3.2. Intel i486SX (P23)

A low-priced version of i486DX, without FPU. Some initial models had no cache as well. Some models had the built-in FPU, but disabled; most models had no FPU circuitry at all. For hardware floating-point calculations it required a mathprocessor (i487), which was nothing but a fully-featured i486DX with a slightly different pinout; once installed it turned i486SX into indefinite sleep, and performed well for both.

3.3. Intel i486SL

Intel's third attempt to produce an all-in-one chip; the core of i486DX + three controllers: memory, ISA and PI (Peripheral Interconnect bus). Like before, exclusively for notebooks and other portables. Of course, i486DX pin-incompatible.

3.4. Intel i486DXL and i486SXL

Nothing but power-saving modifications of i486DX and i486SX, respectively.

3.5. Intel i486DX2 (P24)

A clock-doubled version of i486DX: the first x86-family CPU to run a FSB at half a core speed.

3.6. Intel i486SX2

A clock-doubled version of i486SX.

3.7. Intel i486DX4 (P24C)

A clock-tripled variant of i486DX, with the larger cache. Could run with clock-doubling (CLKMUL pin).

3.8. AMD 486 series

Two years after Am386, AMD started manufacturing Am486-family CPUs. All initial models were complete clones of i486 ones, with the same core and microcode. Later, due to numerous legal issues with Intel, AMD invented the proprietary 486-class microcode, and implemented it.

3.8.1. Am486DX

A clone of i486DX.

3.8.2. AMD Am486SX

A clone of i486SX, with the proprietary microcode.

3.8.3. AMD Am486DXL, Am486DXLV and Am486SXLV

Low-power modifications. V-versions also operated at a low voltage (3.0V). SXLV-version used the proprietary microcode.

3.8.4. AMD Am486DX2 and Am486DXL2

Clock-doubled versions of Am486DX and Am486DXL, respectively.

Some 3.3V 66MHz and 80MHz CPUs were actually DX4 100MHz chips that failed factory tests and were sold as DX2, though most of them worked well at 100MHz, with a proper cooling.

3.8.5. AMD Am486DX4

A clock-tripled version of Am486DX. Could run with clock-doubling (CLKMUL pin). 100MHz model had either a write-through or write-back cache, while 120MHz -- a write-back only.

3.8.6. AMD Am5x86 (X5)

Regardless of the nomenclature, Am5x86 was nothing but a clock-quadrupled version of Am486DX, with a larger cache (16Kb). Known to be easily overclockable to 160(40)MHz. On integer operations performed like Intel Pentium-75, with overclocking -- like Pentium-90.

Note: since motherboards had no jumper to enable 4x multiplier, Am5x86 interpreted 2x mode as 4x.

3.9. IBM 486BLX, 486BLX2 and 486BLX3 (Blue Lightning)

These CPUs were manufactured by IBM under the same joint-development agreement with Intel as 386 series. Blue Lightning CPUs were designed as i486SX with a 16Kb cache, and with i386DX-compatible pinout.

486BLX2 and 486BLX3 ran with clock-doubling and clock-tripling, respectively. In a matter of fact, they were the fastest CPUs to fit the 386 PGA socket.

3.10. Cyrix 486 series

About a year after releasing the proprietary 386 core, Cyrix announced a new one, and released a number of true 486-class CPUs.

3.10.1. Cyrix FasCache Cx486S (M5)

i486SX pin-compatible, with a little 2Kb cache and without a FPU.

3.10.2. Cyrix FasCache Cx486S/e and Cx486S-V

Low-power versions of Cx486S. Cx486S-V also operated with a low voltage (3.3V).

3.10.3. Cyrix FasCache Cx486DX and Cx486DX-V33 (M6)

Fully-featured versions of i486DX, and pin-compatible. Cx486DX-V33 was a low-voltage (3.3V) version of Cx486DX.

3.10.4. Cyrix FasCache Cx486DX2 and Cx486DX2-V33 (M7)

Clock-doubled versions of Cx486DX and Cx486DX-V33, respectively.

3.10.5. Cyrix FasCache Cx486S2, Cx486S2/e and Cx486S2-V

Clock-doubled versions of Cx486S, Cx486S/e and Cx486S-V, respectively.

3.10.6. Cyrix FasCache Cx486DX2-V

A low-voltage (4V) version of Cx486DX2.

3.10.7. Cyrix FasCache Cx486DX4 (M9)

A clock-tripled version of Cx486DX, i486DX4 pin-compatible.

3.10.8. Cyrix Cx5x86 (M1sc)

A simplified version of future Cx6x86, with only one 7-stage integer pipeline. i486DX4 pin-compatible. Could operate with clock-doubling. Most advanced x86-family scalar CPU:

  • two AGUs (load and store), and each unit had a proprietary FIFO buffer and a dedicated 32-bit bus to the on-chip cache. Thereby, read\write operations could be done simultaneously;
  • a branch prediction algorithm with a branch target buffer (256 positions), to minimise pipeline failures while executing wrong branches;
  • a return stack, for a faster RET handling;
  • a data forwarding (aka memory read bypassing), an instruction priority algorithm, etc.

Unfortunately, the branch prediction and the return stack should not be enabled because of a buggy implementation (possible system crash). With all the features enabled, 120MHz CPU performed on integer tasks like Pentium-90. But, due to the large core size this CPU was very badly overclockable and required a good fan.

3.11. UMC U5SX, U5DX and U5SXL

United Microelectronics Corporation (UMC), a Taiwan semiconductor company, started its own line of 486-class CPUs. But since it violated some Intel's patents, these CPUs were restricted for sale on the US market.

Pin-compatible with i486SX (U5SX) and i486DX (U5DX); U5SXL was a low-voltage (3.3V) version of U5SX. A mathprocessor (U5) could be installed with U5SX and U5SXL.


Technical datasheet
CPU Speeds Released Cache Mem. bus Addr. bus Trans. Tech. Voltage Instr. set
i486DX 20 to 50MHz 3 Apr,1989 8K WT 4-way 32-bit 32-bit 1.2 mln. 1.0µ or 0.8µ 5V i486
i486SX 16 to 33MHz Apr,1991 8K WT 4-way 32-bit 32-bit 0.9 mln. 0.8µ 3.3V, 5V i486
i486DX2 40(20) to 66(33)MHz Mar,1992 8K WB 4-way 32-bit 32-bit 1.2 mln. 0.8µ 5V i486
Am486DX 33 and 40MHz Apr,1993 8K WT 4-way 32-bit 32-bit 1.2 mln. 0.7µ 5V i486
Am486SX 33 and 40MHz Jul,1993 8K WT 4-way 32-bit 32-bit 0.9 mln. 0.7µ 5V i486
Cx486S 33 to 50MHz May,1993 2K WB DM 32-bit 32-bit ? ? 5V i486
Cx486DX 33 to 50MHz Sep,1993 8K WB 4-way 32-bit 32-bit 1.1 mln. ? 3.45V, 5V i486
Cx486DX2 40(20) to 80(40)MHz Sep,1993 8K WB 4-way 32-bit 32-bit 1.1 mln. ? 3.45V, 5V i486
Am486DX2 Am486DXL2 50(25) to 80(40)MHz Sep,1993 4 8K WB 4-way 32-bit 32-bit 1.2 mln. 0.7µ 5V i486
IBM 486BLX 486BLX2 486BLX3 16 to 33MHz (doubled, tripled) late 1993 16K WT 4-way 32-bit 32-bit 1.4 mln. 0.8µ 3.3V i486
i486DX4 75(25) and 100(33)MHz Mar,1994 16K WB 4-way 32-bit 32-bit 1.6 mln. 0.6µ 3.3V i486
i486SX2 50(25) and 66(33)MHz Mar,1994 8K WT 4-way 32-bit 32-bit 0.9 mln. 0.8µ 5V i486
UMC U5SX and U5DX 25 to 40MHz Aug,1994 8K WT 32-bit 32-bit ? 0.6µ 3.3V, 5V i486
Am486DX4 100(33) and 120(40)MHz Sep,1994 8K WB 4-way 32-bit 32-bit 1.2 mln. 0.5µ 3.3V i486
Cx486DX4 75(25) and 100(33)MHz Sep,1994 8K WB 4-way 32-bit 32-bit 1.1 mln. ? 3.45V i486
Am5x86 133(33)MHz Nov,1995 16K WB 4-way 32-bit 32-bit 1.6 mln. 0.35µ 3.45V i486
Cx5x86 100(33) and 120(40)MHz Oct,1995 16K WB 4-way 32-bit 32-bit 2.0 mln. 0.65µ 3.45V, 4.0V i486

3. 50MHz model was introduced in June of 1991, but due to numerous overheating problems and overall system instability was recalled.

4. 80MHz Am486DX2 was released in September of 1994.



Designed and maintained by Alasir Enterprises, 1999-2005
rhett at alasir.com