PC 80x86 Processors
PC 80x86 ProcessorsThe early 80x86 processors, the 8088, 8086, and 286, are not included, as such processors are not being offered for sale today. There is also no entry for the PowerPC, as the PowerPC cannot run x86 software in its native mode.
In many cases, I have listed only chips from major manufacturers (i.e. Intel, AMD, Cyrix) except in instances where these manufacturers offer no comparable chip (ex. IBM's SLC series). You can use this list as a rough guide as to which processor or PC to buy, although as always, differences between machines ranging from bus architecture to video cards to disk speeds will affect performance. Test and compare systems before you buy.
If you want a rough estimate of comparative processor speeds, look at the iCOMP marks for specific Intel processors.
Speed : internal/external (available clock speeds in Megahertz) Internal cache : size (internal L1 cache, if present) Math coprocessor : internal, external, and chip designation Power : available voltage configurationsDescription.
Chips are usually available in more than one speed, and can be distinguished by the number following a dash after the chip name. For example, a 386DX-33 is a 386DX chip running at 33 MHz.
Clock-doubled and -tripled chips will have two speeds listed. The first, before the slash, is the internal clock, while the second is the external clock.
Speed : 16, 20, 25, 33 Mhz Internal cache : none Math coprocessor : external 387SX Power : 5VThe 386SX was the lowest entry-level 386 processor available. Initially developed to replace the aging 286, the 386SX was usually found on notebooks and diskless network nodes, and was the lowest-level processor able to run 32-bit code. The 386SX offered a 32-bit internal bus on a 16-bit external data path (hence the suffix SX, for single precision), allowing manufacturers to mount them on cheaper motherboards. Early 386SX motherboards lacked a coprocessor socket, in order to lower costs even more.
Speed : 25, 33, 40 Mhz Internal cache : none Math coprocessor : external 387SX Power : 3.3V (25Mhz only), 5VThe Am386SX replaced the Intel 386SX in late 1993 due to its lower cost and faster clock speed. Most 386SXs found today have this chip, although AMD's 386DX chip prices had also rendered it obsolete. Several Asian manufacturers had designed 386 boards that accepted both SX and DX chips, and were upgradeable up to the DLC.
Speed : 16, 20, 25 Mhz Internal cache : none, 4K Math coprocessor : external 387SL Power : 3.3V (16 and 20 Mhz), 5VThe SL was an SX offshoot that offers power management and speed-reduction functions to prolong battery life. Intel produced four distinct models, running at 3.3 and 5V, with cache and without. It was a popular chip for portables, but has been superseded by the 486SL series of processors.
Speed : 16, 20, 25 Mhz Internal cache : 8K Math coprocessor : unknown Power : 3.3VThe 386SLC was the entry-level processor of IBM's SLC family of low-power processors, pin-compatible with Intel's 386SX CPUs. SLC means 386SX, Low power Chip. It improved on the Intel offering by optimizing frequently-used operations, and included the full 486SX instruction set. The chip was unique in that any one chip can run at three different speeds, and could run in "asynchronous" mode, a forerunner of clock-doubling technology. IBM did not sell the chip directly to vendors, rather, it made complete motherboards at its Austin plant.
Speed : 16, 20, 25, 33 Mhz Internal cache : none Math coprocessor : external 387DX Power : 5VThe 386DX was the first 32-bit processor on the market. The DX suffix (double precision) was added to differentiate it from the SX, developed later. Cheaper versions from other manufacturers soon superseded Intel's chips, and falling 486 prices finally rendered it obsolete.
Speed : 25, 33, 40 Mhz Internal cache : none Math coprocessor : external 387DX Power : 3.3V, 5VAMD's 386DXs gave the company its foothold in the 80x86 market, with the 40-MHz version the most popular. The Am386DX was the first chip to run at 40 Mhz, and offered faster performance and full compatibility at a lower price.
Speed : 25, 33 Mhz Internal cache : 1K Math coprocessor : external Cx487SLC Power : 3.3VCyrix's SLC series offered performance slightly faster than a 386DX running at the same clock speed. The 486SLC/e incorporated the full 486 command set, and had a 32-bit internal bus, but a 16-bit external data path, like the 386SX. It did not have an internal floating-point coprocessor, although a hardware multiplier performs integer math about four times as fast as Intel chips. Also, the price for an SLC plus a coprocessor was almost as low as a comparable 386DX, and for a time was sold as a cheaper alternative to the 486. The SLC's 16-bit external bus may make it a socket-compatible replacement for some 386SXs. The 486 number was considered a mere marketing ploy, and the SLC/e was generally derided as a souped-up 386, since the Cx486SLC-25 was slower than a 386DX- 33.
Speed : 33, 40 Mhz Internal cache : 1K Math coprocessor : external Cx487DLC Power : 3.3VThe DLC is internally identical to the SLC/e, but has a full 32-bit external bus. Despite the name, the DLC is slightly slower than a 486SX; the 33MHz version roughly on par with the 486SX/25. Like the SLC, it is usually sold as an upgrade for the 386DX. Several software incompatibilities with true Intel processors (most notably timing differences) hampered sales of the DLC, despite an aggressive pricing scheme.
Speed : 25/50 Internal cache : 1K Math coprocessor : external Cx83S87-33 Power : 3.3VThese chips are pin-compatible upgrades that fit over a 386SX processor. They are internally similar to the DLC, although utilizing clock-doubled technology. An aggressive marketing scheme has claimed that these upgrades increase processor performance by over 100%, although tests show that the small L1 cache limits the improvement in speed to 40-50%.
Speed : 25/50, 33/66 Internal cache : 1K Math coprocessor : external Cx83D87-33 Power : 3.3VThe DRx2 is the Cyrix upgrade for the 386DX, replacing the CPU on the motherboard. An upgraded 486DRx2-66 is still slower than a similarly clocked 486SX-33. These upgrades do not have an integrated FPU, instead using Cyrix's FasMath coprocessors.
Speed : 16, 20, 25, 33 Mhz Internal cache : 8K write-through Math coprocessor : external 487SX Power : 3.3V (16, 20, 25 Mhz), 5VInitially disdained as a marketing ploy by Intel, as the first 486SXs were crippled 486DXs without a cache, and the 487SX was grossly overpriced. Intel's OverDrive series, however, made the 486SX a viable alternative to the 386DX. Despite the SX suffix, the 486SX was a full 32-bit processor, and was only differentiated from the 486DX by the lack of a floating-point coprocessor. Performance was twice that of a similarly clocked 386DX.
Speed : 25 Mhz Internal cache : 8K write-through Math coprocessor : external 487SX Power : 3.3VThe 486SL is the low-power APM version of the 486SX, for use in portables.
Speed : 25 Mhz Internal cache : 16K Math coprocessor : unknown Power : 3.3VThe 486SLC is an SX-like design, with a 32-bit internal and 16-bit external bus. Its large internal cache and a more efficient design, however, make it faster than a comparable Intel 486SX.
Speed : 25/50, 33/66 Mhz Internal cache : 8K write-through Math coprocessor : external 487SXThe SX2 is a scaled down implementation of Intel's clock-doubling technology, allowing the processor to run at twice the speed of the motherboard, although lacking the FPU of its more powerful DX2 cousin. Released in early 1994, it runs around 30 percent faster than the 486SX.
Speed : 16/33, 20/40, 25/50 Internal cache : 8K Math coprocessor : external Power : unknownThis relatively rare chip is offered by Evergreen as a 386 replacement upgrade, in the 386DX2+ package. With a 486 core and double-speed clock, it improves 386 performance by some 50-60%.
Speed : 33, 40, 50 Mhz Internal cache : 2K write-back Math coprocessor : external Power : 4VThe 486S is a lesser-known (late 1993) Cyrix design, essentially a low-power version of the 486DX. Cyrix uses its own microcode, allowing it to produce these clones with less legal problems than AMD, although there may be potential incompatibilities with true Intel microcode.
Speed : 25, 33, 40, 50 Internal cache : 8K write-through Math coprocessor : internal Power : 5VThe 486DX was the first processor with an internal floating-point processor; and pushed motherboard speeds up to 50 Mhz. It also boosted integer functions with a redesigned RISC integer core. Cheaper versions made by AMD and Cyrix drove 486 prices down by mid-1994, but by then, Intel had already stopped production in favor of the Pentium.
Speed : 33, 40 Mhz Internal cache : 8K write-through Math coprocessor : internal Power : 3.3V (33 Mhz), 5VReleased in mid-1993, the Am486DX has had its share of legal problems with Intel, which contested the AMD's rights to its microcode. No compatibility problems have been reported, and performance is right in line with Intel's chips. Compaq's decision to use AMD's chips in its machines and Microsoft's endorsement has served to lower fears of incompatibilities with Intel's chips.
Speed : 25/50, 33/66 Mhz Internal cache : 8K write-through Math coprocessor : internal Power : 5VUntil recently the fastest processors available, DX2s use a "clock-doubling" technology that allows the chip to run at twice the speed of the motherboard. This allows for faster performance on cheaper, easier-to-manufacture boards, although designs for a 50/100 DX2 have been reportedly shelved, due to heat problems. Intel offers the Overdrive series of upgrade processors, which allows clock doubling on existing 486SX and DX systems. DX2s offer a 70 to 80 percent performance increase over a similar DX. A DX2-66 is slightly faster than a DX-50 in on-chip operations, but falls behind in I/O.
Speed : 20-40, 25-50, 33-66, 25-75, 33-99 Internal cache : 8K write-through Math coprocessor : internal Power : 5VThe Overdrive series of processors are strictly not stand-alone processors; they are upgrades for existing systems. There are two types of Overdrive-capable systems, SX systems, which fit the ODP into the Vacancy socket originally intended for the 487SX, running concurrently with the SX, and DX systems, which directly replaces the existing processor with the ODP. An Overdrive offers increases up to 50%-70% in performance, with a compatible system. The clock-tripled 486DX4-75 and 486DX4-100 ODPs were released in late 1994, and the long-awaited Pentium P24T upgrades are scheduled by year's end.
Speed : 25/50, 33/66, 40/80 Mhz Internal cache : 8K write-back Math coprocessor : internal Power : 4VThe clock-doubled version of the 486S, it is currently one of the cheapest 486s available, although its non-standard 4V power requirement has caused more than one problem with some motherboards. Like the DLC, sales are hampered by rumors of software incompatibilities.
Speed : 25/50, 33/66 Mhz Internal cache : 16K Math coprocessor : external Power : 3.3VThe top-of-the-line processors of the IBM SLC family, these processors offer a low-power alternative to Intel's DX2s. Using 0.7 micron technology and a clock doubler, its larger cache and higher speed promise faster performance than Intel's low power designs.
Speed : 20/40, 25/50, 40/80 Mhz Internal cache : 8K write-through or write-back Math coprocessor : internal Power : 3.3V, 5VAMD's clock-doubled chips, released in late 1993, are reigning as the top 486 seller, like its sibling the Am486DX. Performance is said to be comparable to the Intel chip. Compaq uses the Am486DX2-40 extensively in their low-end notebooks. The DX2-80 is now the fastest available clock-doubled chip, offering up to a 21% improvement over Intel's DX2-66.
All AMD DX2 and DX3 chips produced after Nov. 1994 are now low-voltage models with write-back L1 cache.
Speed : 25/75, 33/99 Mhz Internal cache : 16K Math coprocessor : internal Power : 3.3VIBM's clock-tripling chips, code-named "Blue Lightning", entered the market late 1993. These offer performance beyond the DX2, and for a time were the sole clock-tripled chips extant. These chips are also being marketed upgrades to 386 systems, the 386DX3+, mounted on small upgrade boards to match the pinouts to the 386. Such setups more than double the performance of the upgraded system. They can also be used to upgrade existing 486 systems, with lesser improvement.
Speed : 25/75, 33/99 Mhz Internal cache : 16K Math coprocessor : internal Power : 3.3VReleased in mid-1994, Intel called these chips the DX4, although it uses the same clock-tripling technology found in the Blue Lightning, supposedly to show the speed advantage over the plain 486DX. Utilizing Intel's new 0.6 micron technology and a larger 16K cache, these chips are designed to fit in the niche between DX2s and Pentiums.
Speed : 33/99, 40/120 Mhz Internal cache : 8K write-through Math coprocessor : internal Power : 3.3VAMD's own clock-tripled answer to the DX4 entered the market in mid-1995, with lower prices, and is being marketed as a low-cost alternative to the Pentium-60 and -75. The DX4-120 entered full production in August 1995, making it currently the fastest 486 device available.
Speed : 60, 66, 75, 90, 100, 120, 133 Mhz Internal cache : 16K write-back Math coprocessor : internal Power : 3.3V (90 and 100 Mhz), 5V (60 and 66 Mhz)Intel's much-vaunted Pentium chips entered the market in early 1993, although the earliest models offered slightly less performance than a DX2-66. Code-named the P5, it was then renamed the Pentium, rather than the expected 586, to allow Intel to copyright the processor name. A 64-bit internal bus and superscalar design offers 75 percent more performance than a DX2-66, and the enhanced floating-point unit runs at five times the internal FPU of the 486. It is the first chip capable of running 64-bit software, and later versions might have a 64-bit external bus as well. The Pentium clock speed, which require some of the fastest boards ever designed for an x86 system, as well as the scarcity of the chip itself, has driven Pentium prices way up, twice or even three times that of a DX2-66.
The 90 and 100 Mhz models, the P54C, entered the market in late-1994, and a 150 Mhz version is reportedly in the works. The P54C incorporates clock division, power management, an APIC (Advanced Programmable Interrupt Controller), and a new dual-processing scheme (SMP) that allows two P54Cs on a single motherboard.
A 75 MHz low-power P54C was released in November 1994, targeted at notebooks and portables. Also in November 1994, a flaw in the microcode of the FPU in some Pentium-60s was discovered, six months after the chip's release. The resulting fiasco failed to reduce Pentium sales, however, and in mid-1995 Intel announced two new models, the P-120 and P-133, using new 0.35 micron, 3.3V technology. The P55C, using 0.35 micron technology, running at a clock speed of 150MHz, is scheduled for late 1995.
Speed : 60, 66 Mhz Internal cache : 16K write-back Math coprocessor : internal Power : 3.3VAn Overdrive version of the Pentium, the P24T, is chip-compatible with most 486 motherboards. To address heat problems, it will include a small fan and piggybacked voltage transformer for compatibility with 5V motherboards. It was scheduled for release by the end of 1994..
Speed: 60, 66 MHz Internal cache : 16K instruction, 16K data, 4-way set associative Math coprocessor : external Nx587 Power : 5VNexGen, a relative newcomer, released the first prototypes of its CISC/RISC hybrid chip in early 1994. Unlike the Pentium, it uses an external FPU, instead integrating larger split instruction and data caches (twice as large as the Pentium), and an on-chip L2 cache controller, which communicates with the external cache via a dedicated bus. It translates 80x86 CISC instructions into RISC86 RISC code, and puts them through two 7-stage integer pipelines (versus 5-stage on the Pentium) and three execution units. Finally released for production in late 1994, the company has so far failed to produce the chip in sufficient volume to make a dent in the market.
Speed : 30/90, 33/100 Mhz Internal cache : 16K unified, with 256 byte instruction-line cache Math coprocessor : internal Power : 3.3VCyrix's newest project, code-named "M1", is its fifth generation chip, splitting from Intel's long-established architecture. Utilizing 32 GPRs with register renaming (to maintain compatibility), 7-stage integer pipes, dynamic branch prediction with speculative execution, and four execution units (as opposed to three in the Pentium), test M1 units are scheduled for release in mid-1995. IBM will be manufacturing the M1, and has bought the right to produce the chip under its own name. Volume production will begin in late 1995.
Speed : 50/100, 33/99 MHz Internal cache: 16K unified write-back 4-way associative Math coprocessor : internal Voltage : 3.3VThe 5x86, formerly the M1sc, is the first chip in the M1 series. Initial tests indicate it to be slightly faster than the Pentium-75, and twice as fast as Intel's 486DX4-100.
Speed : 100-120 MHz, 150 MHz in 1996
Internal cache: 16K instruction, 8K dual-ported data, 4-way set
associative
Math coprocessor: internal
Power : 3.3V
AMD's fifth generation processor is code-named "K5", which originally
stood for Kryptonite-5, the fictional element that could destroy Superman. It
was later changed to Krypton-5, to avoid charges of copyright infringement from
DC Comics.
Implemented in 0.5 micron, three-layer-metal CMOS (the same as for the AMD486DX4), the chip has 4.3 million transistors, utilizing a six-unit, four-way out-of-order issue, superscalar architecture, and an x86 instruction decoder that splits CISC instructions into smaller R-ops (RISC operations). Clock speeds are above 100 MHz, although the external bus can run at one-half to one-third processor speed. The chip is still pin-compatible with the Pentium P54C series, although the company says it achieves a 30% increase in integer performance over comparable Intel chips.
Speed: Unknown, probably 120-150 MHz Internal cache: Unknown Math coprocessor: Unknown, probably internal Power: UnknownIntel's engineers are working yet again on the next generation of processors, reportedly utilizing markedly different microarchitectures from the current x86 and Pentium families. possibly with integrated PCI and/or DSP functions. It is expected to be a two-chip module (in a dual-bay wide housing) that improves on the Pentium core, with integrated 256K L2 cache. The P6 is scheduled for release in late 1995, while the P7, allegedly already on the drawing board, is slated for 1997.
This list is also mirrored on the Szeto Family Home Page.
Most of the data was taken from various magazine articles, Usenet postings,
and the Intel, AMD, and Cyrix Web pages.
The information in this
file was researched in mid-1994 to early 1995, and some or all of the data
presented may be out of date. For any updates, comments, suggestions, mail evc@mudspring.uplb.edu.ph.