Advanced Micro Devices (AMD) is one of the most passionate companies in the personal computer industry. The web is full of comparisons between the Athlon, PIII, and PIV examining their performance in almost every possible scenario, and hardware forums feature battles between Intel and AMD fans.
According to AbbreviationFinder, AMD is founded in 1969 by Jerry Sanders and seven other people. In its first five years, AMD is dedicated to redesigning and improving other companies’ products offering more warranty on its final product. By the fifth year, it already has 1,500 employees and a turnover of 26.5 million dollars. In 1976 AMD and Intel sign an intellectual property license exchange agreement. In 1985 AMD entered the list of the 500 most valuable companies in the world.
In the days when the 386 and 486 microprocessors reigned, AMD had a fairly simple purpose: “to copy Intel products, optimize the designs somewhat and offer them at more Mhz as their own products at a lower price”. These times were characterized by numerous lawsuits between Intel and AMD.
The Am386 and Am486 were the result of this little legal policy on the part of AMD, which had an enormous popularity reaching with them the largest market share it has ever had, around 30%. The stability of AMD products at the time was identical to that of Intel products.
The Pentium, when debuting, broke many schemes within AMD. This micro brought many new features to the x86 (PC) market: superscalar architecture, separate level 1 (L1) cache (8KB for data and 8KB for instructions), a pipelined FPU (mathematical coprocessor) that would soon become something mandatory, 64-bit data bus, dynamic branch prediction, and multithreaded capabilities.
Due to this situation, AMD worked hard with the K5, adding 24KB of L1, 8KB for data and 16KB for instructions, which was an improvement over the Pentium. The K5s were introduced relatively late (October 1996) and their performance at the same Mhz was almost comparable to the Pentium.
The K5 did not go over 133 Mhz when the Pentium reached 200 Mhz, so AMD used a trick by naming its microprocessors by the “PR” or “Perfomance Rating” index. This means that an AMD K5 PR166 performs like a Pentium 166 Mhz, according to AMD. In this way, users who were unaware of this fact, when buying a PR166 they thought they were buying a CPU that ran at 166 Mhz when it was actually 116 Mhz. The K5 era can be defined as one of AMD’s worst due to the loss of market share and performance competitiveness with Intel.
After the failure of the K5, AMD needed a good design to return to the scene capable of competing with Intel, but this would take many years, so AMD decides to buy NeXGen in early 1996 for about 800 million dollars and optimize the Nx686. AMD saw the Nx686 as a quick and inexpensive salvation, improving it and becoming the K6.
The K6 was a better and more advanced chip than the classic Pentium and the Pentium MMX looking much more like the Pentium II.
AMD had the performance crown with the K6 with the K6-233Mhz for about a month, just until Intel introduced the Pentium II in the summer of 1997, turning the situation around.
It is then that AMD presents the K6-2 in May 1998 with a minimum frequency of 300Mhz, introducing multiple improvements such as: Super 7 platform with 100 Mhz FSB (micro-board connection bus) and AGP x2, 21 3D NOW instructions !, double superscalar unit for the execution of MMX instructions.
The Pentium II could not compete with the K6-2, so Intel introduced the Celeron, a processor destined for the low-cost market where the K6-2 reigned. The level 2 cache of the K6-2 worked at the speed of the FSB, that is to say 100Mhz, while the L2 of a Pentium II located in the cartridge itself worked at half the Mhz of the micro. This made the K6-2 up to 350Mhz comparable to a PII in general purpose applications.
AMD that was very aware of this problem in February 1999 introduced the K6-3, which incorporated 256KB of L2 integrated on the chip itself. Its performance in general purpose applications experienced a notable improvement while in FPU this improvement was not so great due to the nature of the FPU operations, less susceptible to being cached.
The K6-3 would have been a huge success for AMD if it weren’t for the numerous manufacturing problems it presented. It was AMD’s first micro with L2 on-die and its inexperience in the matter cost a lot. The K6-3’s L2 cache, like all memory, is organized by row and column like a spreadsheet. If a certain area of memory is defective, the entire memory is rendered useless. This is the case unless you have sophisticated systems that re-map faulty areas of memory using lasers to another error-free area. AMD did not have it and the quantity of K6-3 that did not exceed the quality levels were sold as K6-2 with the L2 deactivated, there being no difference between a K6-2 and a K6-3 with the L2 deactivated.
Despite the commercial success of the K6 series, AMD has been losing money for years, bringing the most anticipated AMD processor, the K7, to the market. According to many analysts the K7 or Athlon meant live or die.
AMD had no right to use any Intel platform after Socket 7 so it had to design its own infrastructure. This is something very difficult so AMD decided to acquire the rights to the EV6, the FSB used by the last generation of Alphas, the Alpha 21264.
The Athlon debuted in June 1999 and it more than met all the expectations raised. For the first time, an AMD processor was able to beat Intel’s competitor chip in every respect. The Athlon debuted in Slot A format running at 500, 550, 600 and 650 Mhz, 50 Mhz more than the faster Pentium III running at 600Mhz.
The Athlon’s performance improvements were the result of advanced design of the new AMD chip. Notable features include: 128KB of L1, the largest level 1 cache seen on dual-port x86; 512KB of L2 on cartridge-mounted SRAM chips; Very advanced EV6 Bus (FSB) running at 200 Mhz DDR (100×2); super-segmented pipeline (superpipelined) to achieve high clock frequencies; the first 3-drive segmented x86 superscalar FPU (Sum, Product and Load / Storage), AMD’s first segmented FPU; an extension of 3DNOW.
The combination of an L1 cache 4 times larger than the L1 of the PIII, a bus much more advanced than the GTL of the PIII running at twice the frequency and a much more elaborate FPU made it seem that the Pentium III was a clearly inferior product.
Intel was not going to sit idly by and responded with a 0.18micron version of its PIII that carried 256KB of L2 on-die 3 times faster than the Athlon cache. In addition, this new PIII cache was connected to L1 through a 256-bit bus, allowing much larger transfers.
As time went by, both AMD and Intel were introducing higher Mhz versions of their flagship chips. It was a crazy Mhz race that AMD always won by releasing a version 50 Mhz more than Intel. This race destined to reach the Gigaherzio was won by AMD.
