Elmer, Intel P2 has 31 software, 7 hardware UNFIXED bugs. K6? 3,3
You sound like Fuchi demanding proof that I am an AMD shareholder.
No, I did not ask you whether you own or ever owned AMD stock. I don't care, that has nothing to do with whether you can post accurate information. If Intel is paying you, THAT has everything to do with it. Since you refuse to answer, WE WILL ALL ASSUME THAT YOU ARE PAID BY INTEL.
AMD says the system behaviour will be unpredictable
Read the document amd.com. This phrase occurs 4 times in paragraphs 2.3.1, 2.4.1, 2.6.1 and 2.6.2. All of these bugs were corrected in the C stepping or earlier. My statement is correct and yours is incorrect.
The K6 FPU bug, and your apology for it, sounds just like Intel's response to the FDIV bug that caused a $500 million recal.
1. Only occurs in extended precision (21 digit) floating point mode, only used in a handful of applications
2. The probability of the Intel bug occuring randomly was 1 in 65000 or 1 in 4,200,000 (not sure which). Pretty small, but apt to happen in a scientific computing environment.
3. The probability of the AMD bug occuring randomly is
1 in 48,000,000,000,000,000,000,000 (1 in 4.8x10^22), at least a million billion times less likely to happen.
4. The K6 causes an interrupt if this 1 in 48 million million billion event ever occurs. This makes it easy to 'catch' the bug in software with no speed penalty. In Intel's case, no interrupt, so software was not a viable option.
In short if all 2 million K6's produced so far did nothing but do extended precision multiplies, it would take 76 years before *one* of them made an error. (assumes 10M extended precision multiplies per second).
Petz |
