19

AMD really started to come into its own as a competitor to Intel around the time of the Intel 80486 introduction. AMD countered Intel by shipping the popular (at least with PC enthusiasts) Am386-40, which many have claimed performed as well as the Intel 80486-25 at a lower price point.

I don't know if AMD itself marketed the Am386 as comparable to the Intel 80486, but they definitely did market their 80486 "clones" as comparable to Intel's Pentium. They even claimed specific performance comparisons, such as the Am586 (which is a 486 clone) being equal to a "P75" (e.g. 75MHz Pentium).

If these claims are "real", then what technical achievements were specifically done by AMD in order to allow their last generation "clone" CPUs to perform on par with Intel's newest generation? Is it really so simple as just getting to higher clock speeds than Intel did? Or, was there more to AMD's technical achievements even in these early years?

Brian H
  • 60,767
  • 20
  • 200
  • 362
  • 1
    Anecdotally, when we were going to buy a "486DX2/66", the salesman said "You can get Intel, or you can get AMD for 20$ less", so we bought the AMD. So at least in this case, the Am486 was a direct competitor to Intel 486DX2/66 – Jonathan Dec 16 '20 at 13:34
  • The AMD 5x86 (not "586") pulled the same trick against the standard 486 processors as the Intel did with the Pentium MMX (P55C) compared to the Pentium (P54C): They doubled L1 cache size, which resulted in 10 to 20 percent higher IPC for "typical software of that era". It's up to you whether you call "more cache" a "technical achievement" or not. – Michael Karcher Feb 24 '23 at 21:29

2 Answers2

20

The Am386 and Am486 were designed as clock-for-clock equivalents of the corresponding Intel CPUs, based on reverse-engineering and AMD’s previous second-source licenses — at least the Am386 even used the same micro-code as Intel’s 80386. The only speed advantages came from higher clock speeds (40MHz v. 33MHz) and, in some Am486 models, the use of write-back cache instead of write-through. Both were marketed against the corresponding Intel CPUs, not the next one up, although magazine reviewers were quick to point out that, for example, the 40MHz Am386 could keep up with the 25MHz Intel 486DX in many tasks.

The Am386 die even bears a “© Intel ’85” note; however comparing it with the Intel 80386 die shows that the Am386 isn’t a direct clone.

The next CPU in the range, the Am5x86, was mostly still the same Am486 core, but clock-quadrupled to 133MHz and more, and with a larger cache. This allowed it to compete, in some benchmarks, with 75MHz Pentiums; this is when AMD introduced the “Pentium Rating” marketing speed rating, which would be used by a number of x86 CPU manufacturers.

The first AMD CPU which used a significantly different design to Intel’s was the K5, based on AMD’s 29k RISC CPU; its complexity however limited AMD’s ability to scale its clock speed, so the K5 couldn’t compete on performance, only on price, offering similar performance per dollar (the Pentium had initial teething problems, but by the time the K5 was released, the Pentium had quite high clock speeds, and its performance got better with the MMX variants released just after the K5 became available). Its successor in the market, the K6, was based on NexGen’s Nx686 after AMD acquired NexGen, and proved capable enough to compete with three generations of Intel CPUs, from the Pentium to the Pentium III (I’m leaving the Pentium Pro aside since it was positioned quite differently).

Stephen Kitt
  • 121,835
  • 17
  • 505
  • 462
  • 1
    Fine summary - except, I wouldn't call the K5 disappointing in any way. They did deliver well, at least on par, but usually better than their Intel counterpart in the same price segment. Of course, it came rather late and failed against the P55 (Pentium MMX) with it's enhanced architecture. – Raffzahn Dec 15 '20 at 16:32
  • 1
    True, I’ve reworded that section to hopefully be more accurate. – Stephen Kitt Dec 15 '20 at 16:50
  • 5
    I joined Nexgen shortly before it was acquired by AMD. As I recall, AMD's motivation for acquiring Nexgen was the relatively poor showing of the K5. AMD's initial target for the K5 had been to ship 100 MHz parts by the middle of 1995. But first silicon fell far short of the frequency target. After additional efforts to fix the frequency issues, volume production commenced late and initial parts for public consumption ran at only 75 MHz. Too little too late to become a truly successful product. FPU performance was a bit anemic compared to the competition as well (by design, as I recall). – njuffa Dec 16 '20 at 03:08
10

Is it really so simple as just getting to higher clock speeds than Intel did?

Yes, it is that simple. Up and including the 486 AMD's CPUs were developed close to Intels devices, supported by in detail information provided by Intel as well as reverse engineering. AMD adapted the design to their production process. This included low level changes in how certain circuits got implemented. Higher speeds are usually a result of that adaption.

Or, was there more to AMD's technical achievements even in these early years?

No, but...

No in sense of functional structure. The CPUs were in all logic parts designed exactly like Intel's template. Same blocks, same path design, same registers, same micro code.

Yes as implementation of elements like transistors, resistors, capacitors did vary due different manufacturing technology. And that's an area substantial for gains in clock frequency and thus performance.

'Technical' (whatever that may be) advancement does not only come from processor design, but at least as much from chip design. And the later is a major component in clock frequencies a certain logical design can reach when made into a chip. And that has been an area were AMD was obvious ahead of Intel.

A Bit of AMD/Intel Licencing History

It's important to keep in mind that AMD and Intel agreed in 1976 to a cross licence and technology transfer lasting for 20 years (until 1995) which covers all their circuitry and micro code. It is that contract that made AMD a supplier of fully compatible CPUs from 8080 until and including the 80486 - Intel in turn got the 9511 FPU and other chip designs.

In 1982 they added another contract specifying technology transfer as primary means. So AMD had not only the rights to reverse engineer all Intel CPUs but as well the right to get all information needed direct without the need to reverse engineer. The 1982 contract got amended in 1984 were AMD agreed to not receive information about Intels new embedded measures for in-circuit emulation (ICE).

This was in turn the base for Intel's 1994 law suit against AMD and its AM486 product line, as AMD copied as well the micro code handling the ICE operation, which Intel claimed being a breach of the 1984 amendment. AMD's position was that the microcode was an integral hardware part of the 486 and thus covered by the base agreement.

Dan D.
  • 103
  • 3
Raffzahn
  • 222,541
  • 22
  • 631
  • 918
  • Interesting to know that the AMD and Intel cross-licensing agreement started in 1976. What does ICE give that single-step mode selected by the trap flag does not? Was there a winner (do not know the legal terminology) in the 1994 law suit? – Single Malt Dec 15 '20 at 18:07
  • ICE is about real time monitoring without interference in the monitored system, single stepping requires cooperation of the host/system software. The case was, as usual, settled out of court in Jan 1995 (see page 11 of this report) With AMD keeping the rights to manufacture everything up to 486es including all Intel micro code (with some restrictions). The general licence agreement was continued, but exempt all Pentium and later CPU. – Raffzahn Dec 15 '20 at 19:00
  • @SingleMalt: At least for CPUs without complicated caching arrangements, an ICE can record information about what instructions are being executed and what data is being fetched, in real time, without affecting the behavior of the system being monitored. This can be very useful when trying to troubleshoot device drivers which need to must data within a short time of its become available to avoid data loss. – supercat Dec 16 '20 at 17:58
  • @supercat so it is more of a real-time functionality and the recording of the instructions would mean the microcode would be a good place to copy these and the data fetches as many (most?) instructions would go through the microcode. Makes sense that would be useful for things like drivers. – Single Malt Dec 16 '20 at 21:29