News

30 years later, they find a problem that cost Intel 475 million

The first processors Intel Pentium They were launched back in 1993, and were a milestone in the industry but for the worse, since a problem classified as an “FDIV bug” was found that forced Intel to carry out the first recall of a product in its history, a fact that cost the company a few 475 million of dollars. Now, more than 30 years later, an engineer has managed to isolate and locate the problem that gave the blue giant so many headaches.

The well-known hardware engineer and historian Ken Shirriff has published his new discovery on his Mastodon account: he has located the exact transistors of the original Intel Pentium that caused the “FDIV bug”, which caused the processor to be withdrawn from the market in 1994 and which It cost Intel a million. This is very interesting, because it was Intel’s first major failure 31 years ago.

The first failure in Intel’s history, finally exposed

With the help of a microscope, Shirriff has managed to trace the root cause of the problem. The image we show you below is a photo of the matrix of the original Intel Pentium processor, the first with Intel’s P5 architecture and which greatly helped the company make a name for itself in the processor industry (although as you might already guess, it backfired on them).

This processor was manufactured using an 800nm ​​process, and the image above has been taken logically using a microscope. The matrix contains 3.1 million transistors, and thanks to the magnification of the microscope, the grids of these can be clearly seen, allowing the operations of its blocks to be identified. This has been possible precisely because of the simplicity of the 800 nm process… imagine a modern 5 nm processor with the tens of millions of transistors it has… it would be impossible.

The error that caused the “FDIV bug” was due to calculation errors in the programmable logic array (the part labeled PLA in the image). The Pentium’s floating-point unit was much faster than in current chips thanks to the SRT division algorithm, which computes division at two bits per clock cycle versus a single bit per processor cycle.

For this to work, you need a table of 2,048 cells in the die arranged in 112 extremely compact rows. The values ​​are indicated by the presence or absence of transistors along the grid points that make up the table, and would have been a really good form of execution if it weren’t for the manufacturing flaw that led to this problem, and it is What has been discovered is that 5 entries in this table are missing their crucial transistors.

These mislabeled entries caused an error in the floating point calculations, and although it was already discovered by Professor Thomas Nicely, Shirriff has now physically demonstrated it on the processor itself; And not only that, in 1994 it was thought that there were only 5 entries, but now it is known that there were a total of 16, 11 more than was believed.

Intel downplayed the problem

It’s curious because Intel called the FDIV bug unimportant, claiming that it only caused errors once every 27,000 years. IBM, for its part and as it always does, came to the fore stating that this problem occurred once every 24 days, and decided to withdraw the Pentium processors from the market. In the end, Intel bowed to monetary and media pressure and pulled the chips from the market, losing about $475 million in the process.

Post New York Times Intel problem

What is true is that after that Intel launched the Pentium processors again but with this problem solved, and as we have said before, it helped greatly to make a name for itself in the industry.

If you are interested in knowing all the details of this story and the research related to it, we recommend you read Shirriff’s complete thread that we have linked at the bottom of this article.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *