By: David Hess (davidwhess.delete@this.gmail.com), August 24, 2018 9:26 pm
Room: Moderated Discussions
Maynard Handley (name99.delete@this.name99.org) on August 24, 2018 12:03 pm wrote:
>
> Is there a real (and realistic) fear that these instructions can generate so much heat so fast
> that the existing thermal tracking is too slow? And something that could be fast enough (eg
> as has been suggested, limiting instruction throughput at some point --- maybe issue, maybe
> decode) is not realistic why? Because the entire thermal modeling system runs at say 100th
> CPU frequency and it was too hard at the time to bolt on a new, better targeted, system?
That is absolutely the case and for a reason not often discussed.
The temperature is not even across the semiconductor die; it is not even close. (1) Function units with high transistor density create local hot spots and it is the peak junction temperature which matters so a thermally based control circuit requires a thermal sensor at each hot spot.
Feedforward control of the thermal dissipation by counting function unit cycles actually works better and takes fewer resources than measuring actual temperatures as long as the overall temperature is known.
There are also dI/dV limits but that is a different subject.
(1) For small dies, the thermal time constant is 100s of microseconds to milliseconds across the die. You can actually measure and see it on an oscilloscope when testing a monolithic transistor array.
>
> Is there a real (and realistic) fear that these instructions can generate so much heat so fast
> that the existing thermal tracking is too slow? And something that could be fast enough (eg
> as has been suggested, limiting instruction throughput at some point --- maybe issue, maybe
> decode) is not realistic why? Because the entire thermal modeling system runs at say 100th
> CPU frequency and it was too hard at the time to bolt on a new, better targeted, system?
That is absolutely the case and for a reason not often discussed.
The temperature is not even across the semiconductor die; it is not even close. (1) Function units with high transistor density create local hot spots and it is the peak junction temperature which matters so a thermally based control circuit requires a thermal sensor at each hot spot.
Feedforward control of the thermal dissipation by counting function unit cycles actually works better and takes fewer resources than measuring actual temperatures as long as the overall temperature is known.
There are also dI/dV limits but that is a different subject.
(1) For small dies, the thermal time constant is 100s of microseconds to milliseconds across the die. You can actually measure and see it on an oscilloscope when testing a monolithic transistor array.