Back in the early 90’s Intel had a problem with the then new Pentium microprocessors that had a built-in math co-processor that would give inaccurate answers in certain circumstances.
Ever since then manufacturers and high end customers have been extra diligent in confirming any new systems and their math capabilities.
BRAVE AI:
This refers to the Pentium FDIV bug, a hardware flaw in the floating-point unit (FPU) of early Intel Pentium processors released in 1993 that caused incorrect results during specific floating-point division operations. Discovered in 1994 by mathematics professor Thomas Nicely, the error stemmed from five missing entries in the chip’s radix-4 SRT division lookup table. Although the flaw was statistically rare for average users, Intel’s initial refusal to replace affected chips sparked massive public outcry, leading the company to issue a full recall in December 1994 that cost Intel approximately $475 million.
More info:
https://people.willamette.edu/~mjaneba/pentprob.html
I know a great deal about the floating point flaw as I was in Intel Validation at the time.
The big problem is when it made an error, the exception flag never got set.
The coverup was done by Vinod Dahm, head of the marketing team as his stock options were linked to the number of Pentium sold.
Vinod was quietly fired and he moved to an x86 clone company, then to AMD.
Last time I heard he was hiding out in FL running a H1b scam company.
It was discovered by the design team working on Pentium II and the simulations were getting different results.
That was Intel’s “Ford Pinto” moment.