You can run floating point calculations on any CPU. You don't need acceleration, but you have to then actually manually do those in microcode. This applies on VAX.
Thoughts about Moore's law
The chart below is in logarithmic scale:
https://qph.cf2.quoracdn.net/main-qimg-c4d853586c9387861f0f219700476c85-lq
It charts the growth of cpu speeds in comparison with the VAX 11/780. It suggests that Moore’s law stopped applying around 2003. I decided to continue the chart using gigaflops as a metric for cpu speed.
I don’t have data on how many floating point operations the VAX 11/780 could do but I suspect it is “none” because I think the first cpu advertised as being capable of floating point operations was the Intel 8087 in 1980.
According to stack exchange it could do 50 kiloflops or .05 megaflops or .00005 gigaflops. It was in 1975 that Moore made his prediction that computers would double in speed every eighteen months, which is the same as saying it would quadruple every three years. If that had been what happened, these would be the numbers:
1980: .00005 gigaflops
1983: 0.0002 gigaflops
1986: 0.0008 gigaflops
1989: 0.0032 gigaflops
1992: 0.0128 gigaflops
1995: 0.0512 gigaflops
1998: 0.2048 gigaflops
2001: 0.8192 gigaflops
2004: 3.2768 gigaflops
2007: 13.1072 gigaflops [note that in 2007 the Intel Core 2 Quad Q6600 was a high end cpu, and it could do 25 gigaflops so we were a bit ahead of Moore’s law]
2010: 52.4288 gigaflops [note that in 2010 the Intel Core i7-980X was a high end cpu, and it could do 98.4 gigaflops so we were a bit ahead of Moore’s law]
2013: 209.7152 gigaflops [note that in 2013 the Intel Core i7-4770K was a high end cpu, and it could only do 177 gigaflops so we fell behind Moore’s law at about this time]
2016: 838.8608 gigaflops
2019: 3,355.4432 gigaflops
2022: 13,421.7728 gigaflops
2025: 53,687.0912 gigaflops
But in 2025 we are actually at 1,696 gigaflops in 2025 with the Intel Core i9-10900K. We are now significantly underperforming Moore’s law. It slowed down around 2013, by my numbers.
Discussion
oh cool
I hope measuring from 1980 instead of 1978 won't make too much of a difference
When I was younger I had to design different floating point operations using microcode.
And that was fun. That's the layer where you decide where the actual "physical" 1s and 0s go and its just one layer above talking voltages...
The fun part is that there are all sorts of tricks to optimize the number of tics it takes to do some more complicated assembly instruction.