Kitz Forum
Chat => Tech Chat => Topic started by: Weaver on October 08, 2020, 11:30:50 AM
-
A few questions for those of you who have ‘turbo boost’ technology in their processors:
- What’s your experience of this?
- I’m thinking about the case where a workload is thrashing the CPU and RAM flat out, either with one single core in use or all of them, or also including hyperthreading so 2n threads waiting to run. If you run one such experiment, how long does it take for the clock frequency to go down and by how much? (I’m not sure how you would tell, other than by repeated runs of a benchmark job, taking account of cache hot/coldness. To be clear, the main test workload does not involve running multiple diverse apps.)
- What about one single thread running flat out vs all cores or vs all threads ? I’m wondering if lighting up multiple cores causes so much slowdown that it’s largely counter-productive because it reduces the clock too much?
- What about ambient temperature ? Do jobs complete faster in a really cold room?
- Do you have a fan on either the processor or whole motherboard?
- Do you have a dynamic fan speed?
-
Dynamic clocking is pretty much instantaneous.
I have an Intel 5830K, which was built around 5 years, over clocked to 4.5Ghz, and water cooled, clock speed adjusts up and down depending on load, but all six cores with hyperthreading will happily run all day long at 4.5Ghz.
My other system has a 12 core Ryzen 3900X, also water cooled, but an all in one, this will happily run all cores hyper threaded at 4.2Ghz, depending on load one or two cores will hit 4.6Ghz, others will be between that and 4.2, or even sub 1Ghz if no demand, its all done on the fly. You need good cooling on these CPU, they will throttle down if too hot.
Most desktop PCs will have active cooling these days, and the fans will be adjusted on the fly.
Depending on the cooling built into the PC a cold room may make no or very little difference, mine ran perfectly fine when my room was plus 30 degrees in the summer.
-
1 - My experience is generally positive Weaver, it brings automated overclocking to the masses.
2 - This depends on how the hardware is configured (mainly bios) and also on things like power draw and heat. My intel 9900k will stay pegged at max clocks forever, as I have set ridiculously high boost parameters and have a high end cooler on it. Of course this means the cpu runs out of spec in such a scenario. My Ryzen 2600X I have it on stock configuration, it will generally be able to keep at its max single core boost clock for a very long time as it doesnt heat the cpu up much or drag it to its power limits, but a all core overload will after a short time slightly downclock, but noy by much though, still at turbo clocks, just not max turbo clocks.
Bear in mind real world loads will keep clocks higher for longer as not many real life workloads will peg maximum load sustained.
3 - Ambient temperature impact will depend on bios configuration and how effective the cpu cooling is.
4 - I have dynamic fan speed, used to use performance profile but now on balanced, most modern PC's now have dynamic fan speeds using PWM connectors.
Additional Notes.
Many cpu's will default to an over voltage, which is because each cpu is different in terms of the quality of the silicon and boards are configured to make sure the worst cpu's work. If you get a better than a worst cpu, you can decrease the voltage, which in turn decreases heat and power consumption, my 9900k defaulted to 1.36v, I run it at 1.25v and it runs around 20C cooler and uses about 40W less under full load.
Also on ryzen's if you disable the turbo boost (XFR, stands for extended frequency range), you get so much better power efficiency, its diminishing returns as clocks go up which is why turbo clocks are not standard clocks, my ryzen at 3.6ghz uses about 60% power as it does in XFR at 4.15ghz.
I feel AMD have done it better, manual overclocking on the ryzen's give you very little improvement over the automated configuration, whilst bigger gains can be had from intel although in their latest generation chips the overclocking headroom has reduced, so they getting there as well.
On my pfsense unit which has an i5, I looked for the power/performance sweet spot, it will downclock as low as 500mhz if you let it, but I observed the voltage and heat at 1200mhz vs 500mhz isnt that different, and then it starts rising rapidly above that so I set the min clock on that to 1200mhz, and set the p-states to a less aggressive configuration so it only ramps up when under sustained load instead of brief spikes.
-
Wow, thank you for the generous replies. I have never had such a CPU so my own experience is zero.
BTW Is there an easy way to get the current clock frequency? Or do you have to just run a benchmark? (The problem with that is that the execution of the benchmark, unless it is small, will heat up the cpu and reduce the clock frequency anyway. Duh.)
How does one disable turbo boost - through the BIOS settings? I presume there’s some way of doing this dynamically on an x86 whatever CPU - through a machine-specific register (MSR)? or a CR or whatever it is?
-
There's absolutely loads of ways to monitor realtime clock rates, here's just two examples.
https://www.cpuid.com/softwares/cpu-z.html
https://www.hwinfo.com/
The best way to make alterations is through the BIOS, although there are various programs the can do it through the OS, such as Ryzen Master.
It needs a lot of heat before a CPU will start throttling, My systems maxed out with stress testing don't exceed around 70 degree's, C but the amount of heat is dependent on many factors, especially voltage and of course how good the cooling system is.
I recently overclocked the ram in my Ryzen system from 3600Mhz to 3800Mhz with much tighter timings and had to fit a ram cooler to keep it cool otherwise it wasn't stable. I actually designed and 3D printed my own housing.