[Ultra] In what Situation should it boost up to 2.56GHz?

Hey Folks,

trying out my Ultra board over here (received it yesterday, sadly they still missed the proper screws for me) and playing around with it.

I can't seem to boost my CPU to 2.56GHz in any Situation, wether it's all 4 threads on full load or one thread on full throttle. It doesn't go any higher than 2.08GHz.

Debian 8, Kernel 4.9, Ultra Board
Linux chip 4.9.0-0.bpo.2-amd64 #1 SMP Debian 4.9.18-1~bpo8+1 (2017-04-10) x86_64 GNU/Linux

Greetings
David

 

I'm testing this with Debian 9 (Stretch) and problem seems to be that default policy doesn't allow max speed.

After installing package linux-cpupower (seems to be available for Debian 8 also from jessie-backports) I run "cpupower frequency-info" which reports:

• hardware limits: 480 MHz - 2.56 GHz
• current policy: frequency should be within 480 MHz and 2.05 GHz

Unfortunately I've been unable to change the default so far.

 

More testing:

• "cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq" returns "2048000" (i.e. 2.048 GHz which according to intel-pstate documentation is then rounded to nearest allowed value, which seems to be 2.08 GHz)
  • that value should be possible to change by running as root "echo -n 2560000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq" but that doesn't work
• "cat /sys/devices/system/cpu/intel_pstate/max_perf_pct" returns "80" (i.e. maximum allowed speed is 80% of maximum available speed - 2.048 GHz is 80% of 2.56 GHz)
  • this values should also be changeable with echo, but it isn't

 

I can alter /sys/devices/system/cpu/intel_pstate/max_perf_pct to any value under 81. 81 and higher will be reverted to 80.

Greetings
David

 

Can someone verify on windows - Ultra or Advanced Board - if the CPU can boost to their respective maximums?

Greetings
David

 

Looking around on the internet for Intel Burst but I only get vague results (like this) that Burst is not the same as Turbo Boost and cannot be set manual and also is difficult to measure if it is active. It will only be switched on for a couple of seconds by an internal CPU algorithm

 

wow, I thought that burst and turbo boost meant about the same. I start feeling that UDOO devs advertising burst speed instead of base speed is almost a fraud.

 

I guess what is not made clear this that this range of Intel CPUs are from the Mobile range and therefore are optimized for battery operation ie laptops. The same CPUs are re-badged for the embedded market with extended availability. Unfortunately these processors will inherently aim to lower the clock speeds to reduce power consumption. Burst mode is partially an attempt to temporary speed up processing with the net effect of lowering the clock afterwards in a shorter time frame. To gain an insight in the how Intel processors are optimised for mobile usage you can read section 13.2 MOBILE USAGE SCENARIOS of the Intel Architectures Optimization Reference Manual . Note its not all applicable to the Silvermont/Airmont microarchitecture for Braswell Mobile range.

 

My advanced plus(all four cores) settled on 1759.960MHz for 15+ minutes when I did kernel compilation using '-j 6'. Temperature for all cores were in range 70C to 75C. This was without any cooling modification on debian stretch.
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1792000
$ cat /sys/devices/system/cpu/intel_pstate/max_perf_pct
80

 

From what I've been able to find out so far, we can not modify the cpu frequency above 2.00GHz. This is handled by the CPU automatically and only for when it's needed. I've been able to successfully change the governor from powersave to performance, but I've not been able to change the cpu frequency. However, even when I max out all 4 cores, I have not seen the frequency go above 2.0GHz.

This looks to be more of a hardware limitation. Now I'm curious to see if I can find a way around this. And my to-do list keeps getting longer. :/

 

You will need to factor in thermal loads for Burst Mode. Burst was introduced a decade ago in the Lincroft chips - this is a very good article on how it worked back then. It is quite different from Turbo Boost.

Basically, burst allows a core to sprint at a speed 100% governed by the current temperature of the CPU. Once the CPU thermal limit has been reached, the CPU will immediately throttle back to normal frequencies. As the name implies, it is meant for very short sprints.

If you are running without a fan, it might be possible that you can't reach the maximum burst frequency, or at least you might not be in burst long enough for telemetry to even pick it up.

 

Key info other than marketing terms such as Burst freq and ATOM/Celeron/Pentium etc. is:
TDP and SDP.

In the case of N3160, TDP is just 6W and SDP even lower: 4W. Latter is the design/operations goal.
With 4W as its limit, the microcode in the Airmont CPU will make sure it's a low-power "Mobile class" CPU as
oppose to a performance oriented desktop-class CPU.

Key:
old classification/tier:
Server > Desktop > Embedded > Tablet > Smartphone "class"

Recent classification/tier:
Desktop > Mobile > Tablet "class"

 

Ok, something is definitely up then...

I just got my Up2 board which uses a 1 generation newer CPU, 1.1Ghz base, with 2.5Ghz burst mode.
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq shows 2500000 (2.5Ghz)
sudo cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq (under IDLE) shows 1019836 (about 1.1 Ghz, as it should)

Then running sysinfo CPU test in another window and rechecking cpuinfo_cur_freq it reports back 2389901 - a burst up to nearly 2.4Ghz...

The same test on my Udoo x86 Advanced reports:

scaling_max_freq = 1792000
cpuinfo_max_freq = 2240000 (2.24 Ghz, which is the proper burst speed for this chip)

cpuinfo_cur_freq (on the cure that has sysbench running on it of course) = 1759960... It never gets up to the full speed it should even with a CPU fan AND a big desk fan blowing on it.


Checking scaling_governor reports back "powersave". I used cpufreq-set to change the governor to "performance" for each of the cores.
No effect on the tests though.

In the UP2 EFI/BIOS there's some options for settings for the Burst mode on the chip.

Has me thinking that the Udoo gang needs to do something they didn't realize in the BIOS to get the burst mode to actually work on these boards.

 

It's certainly possible.

 

UDOO x86 Project Update #36 should provide a hint...

 

I don't see any connection to boost speed with that update about voltage regulator problems.

 

Are you implying that the re-designed board still has power regulation issues that would prevent the cores from reaching max burst freq? That's a bit of an accusation.

 

Hey I didn't say it. Here in the design revision/prototype, copper heat sink is used.
In production, 2x RT8175A w/ 2x SO8 packaged external MOSFETs are used. Controllers are WDFN 4x4 package and using thermally bonded PCB as the heatsink. SO8 is the plain SO8.
No heatsink and no active cooling is used. SECO smart engineers must have done enough calculation.
You decide and make your own judgement call.

 

Did you all miss this post of the Udoo team? It IS boosting to max frequency!