Re:AMD laptops have a hidden 10-second performance delay
by ajnin on Wednesday November 25, 2020 @02:08AM
(#25206658)
I'd have thought the opposite : run the CPU at full power so you can terminate short tasks quickly and go back to sleep. Then if the task takes longer, throttle down as needed to regulate heat dissipation. Since there are a lot of small tasks when using a computer that would be also better for responsiveness. It's not like a car that takes a lot of energy to accelerate, you can just change the CPU clock speed as you want. What I'd like to see is a benchmark of the actual amount of energy used to perform tasks in various conditions.
Re:AMD laptops have a hidden 10-second performance delay
by vvanders on Wednesday November 25, 2020 @01:16AM
(#25206435)
Yup, if I recall from the article the actual power numbers were NDA'd but pointed towards AMD destroying Intel on a battery usage basis.
Sadly very few companies are integrating the higher end Zen chips.
Re:AMD laptops have a hidden 10-second performance delay
by bayindirh on Wednesday November 25, 2020 @01:20AM
(#25206455)
IIRC lately, Intel's real TDP under serious load is almost twice as the public value. I guess it's also valid to their server parts to a certain degree.
Intel has a habit of hiding some ugly things about their chips under NDA until it has competition or the chip is last-gen. The uglier thing is they're adding these values back to their documents later as a revision and telling look, it was already there! (cough, AVX base frequencies, cough!).
Competition and being a very big company changed them... a lot.
Re:AMD laptops have a hidden 10-second performance delay
by rainingcatndogs on Wednesday November 25, 2020 @01:16AM
(#25206434)
Yes, that seems like a feature.
Frequent throttling to full power wastes too much energy in the long run.
IIRC Linux CPU frequency scaling can achieve the same effect with hysteresis settings (which I used to tame my jumpy CPU back in the day).