12/09/2014

Intel’s Z37xx CPUs review

iambillbil Calls Out Intel’s Z37xx CPUs

Hoo boy.
Thermal throttling is the dark secret of the newest-generation of CPUs.
When the temperature of the chip hits a set point, its speed is automatically reduced. So a CPU that its maker brags is 2GHz can throttle back to 1.5GHz — or even less — once it gets hot.
This was already revealed with the Tegra K1 in the Xiaomi MiPad.
Now iambillbil is revealing it in the Intel Z37xx series of CPUs being used by Chinese tablet makers (and others).
In a post at the IMP3Net forum, he names four CPUs and illustrates the impact of thermal throttling [Google Translate].
He writes:
Use of a machine 3740D chips, maximum pressure can be below the core temperature of the heater to about 100 degrees, can tolerate 30 minutes, and then a sustained performance decreases until the heat balance and performance.
Use of a machine 3735D chips, maximum pressure can be below the core temperature of the heater to about 100 degrees, can tolerate 10 minutes later, and then a sustained performance decreases until the heat balance and performance.
3735F chip using a machine, can be at the highest pressure below the core temperature of the heating to about 80 degrees, can tolerate one minute later, and then a sustained performance is reduced until the heat balance and performance.
Z3740D: It can sustain its speed for thirty minutes under maximum temperature, then it throttles.
Z3735D: It can sustain its speed for ten minutes under maximum temperature, then it throttles.
Z3735F: It can sustain its speed for one minute under maximum temperature, then it throttles.
As for the 3736F, he’s unsure but surmises it has an even lower maximum temperature and an even lesser tolerance for it.
Our portable digital technology is facing two constraints that have yet to be surmounted:
1) Heat generation that degrades maximum performance.
2) Almost no progress in battery technology to prolong use.
We also have CPU makers banging up against the wall of physical and quantum effects as they continue to shrink circuitry to the nanoscale level. That’s actually a third constraint that could place another ceiling on top of the other two. It’s hardly ever mentioned because so far they’ve managed to continue to shrink the width of electron pathways. But soon they won’t be able to go further without some kind of near-miraculous breakthrough in materials or computing design.
In the meantime, we need to develop a new set of benchmarks to ascertain what the true amount of computing power we’re getting when we buy a device. It’s clear that the days of AnTuTu and other benchmarks are coming to the end of the road. This is why it’s been only in the past year — ironically, with the entry of Intel into tablets — that CPUs are suddenly described as having a “peak frequency” rather than a sustained maximum frequency.
And lest iambillbil has given the wrong impression, these are issues that affect both the Rockchip 3288 and the Allwinner A80 — as well as other new chips. iambillbil is just blowing the whistle on Intel with information that was never before considered.

No comments:

Post a Comment