
2-6X less energy.
So split the difference and call it 4X less energy. Lets assume the opposition are pulling 500 Watts, so 4X that is 2000 Watts. So 500 Watts minus 2000 watts must mean these ships GENERATE 1500 Watts when in use.
Google on Wednesday revealed more details of its fourth-generation Tensor Processing Unit chip (TPU v4), claiming that its silicon is faster and uses less power than Nvidia's A100 Tensor Core GPU. TPU v4 "is 1.2x–1.7x faster and uses 1.3x–1.9x less power than the Nvidia A100," said researchers from Google and UC Berkeley in a …
.. a story about Turbo Pascal Units, as introduced with TP4 back in 1987.
As WP says, these were "tightly linked to the internal structures of the compiler, rather than standard .OBJ linkable files. This improved compilation and linkage times, but meant that .TPU files could not be linked with the output of other languages or even used with different releases of Turbo Pascal unless recompiled from source."
> And you had to press the Turbo button on the front of your PC before compiling.
I had to wait for a PC with one of those.
I recall a friend at University bringing code for an assignment up to the labs and telling me the compilation of it "can't have worked, it normally takes around a minute". The Unix system* in front of him was so much faster than he was used to and it had given him the prompt back almost instantaneously - and done so silently since there were no errors.
(* and those were just the regular undergrad boxen; one or two beefier systems were ready to be discovered tucked away elsewhere)