fir1 in R2021b vs R2017

I noticed fir1 produces different results in R2021b vs R2017. In 2017 coe17=fir1(30, 3000/48000) results in

coe17 =

-0.0006 -0.0014 -0.0027 -0.0044 -0.0062 -0.0070 -0.0054 0.0000

0.0102 0.0255 0.0452 0.0673 0.0892 0.1078 0.1203 0.1247

0.1203 0.1078 0.0892 0.0673 0.0452 0.0255 0.0102 0.0000

-0.0054 -0.0070 -0.0062 -0.0044 -0.0027 -0.0014 -0.000

coe21 = fir1(30,3000/48000)

coe21 =

0.0004 0.0010 0.0020 0.0039 0.0069 0.0112 0.0170 0.0240

0.0322 0.0411 0.0501 0.0586 0.0660 0.0718 0.0755 0.0768

0.0755 0.0718 0.0660 0.0586 0.0501 0.0411 0.0322 0.0240

0.0170 0.0112 0.0069 0.0039 0.0020 0.0010 0.0004

Why is there a difference?

[Star Strider]

I have not noticed any difference (I no longer have access to either of the R2017 releases) however it does not appear to have changed in any way over those years according to the Release Notes.

.

Answer 1

[Walter Roberson]

I just tested with R2017a and R2021b on my system. The results were bit-for-bit identical between the two systems.

Question: do you happen to have an AMD cpu ?

Answer 2

[gim hom]

Thanks for such fast replies. Turns out to be a flaky CPU. R2017 was on an old machine.