Extreme throttling introduced on A733

Hello,

With this commit: configs: cubie_a7*: adjust temperature control parameters · radxa/allwinner-device@7716c3a · GitHub Radxa basically introduced throttling the board to the absolute minimum speed (~400 Mhz) at the point that 60 degrees is exceeded.

Unless someone uses highly efficient cooling, the board will always do this, even under light load. Which means that this commit makes the boards effectively 5x slower.

Please reconsider this @radxa - can you at least relax this extreme throttling a bit?

@ChenJaly @feng please take a look.

1 Like

Sorry if I may be misunderstanding how this works, but the board started throttling more than before.

Hi @incognito ,
Due to the high heat of A733, it is unavoidable. Even when reduced to the lowest frequency, after approximately half an hour of stress testing, the SOC will still shut down because it reaches 100+ degrees.
We are about to launch a heatsink, or you can use small heatsinks to improve this situation

Thank you. I am observing, however, that the board throttles to 416 Mhz immediately at 60 degrees, when before it was not. I have a proper heat sink installed.

Does this commit have the result I observed?

是的,使用官方的镜像会降频,发挥不出全部的CPU性能。@ChenJaly @feng 建议不要限制CPU的频率,让用户自己考虑散热机制,毕竟选择这款芯片的开发版,主要也是为了使用这款芯片的性能,并能让它全部发挥出来,否则还不如选择其它更低功耗芯片的开发版。

你们只需要在产品手册文档里面加一句,需要配置合适的散热器才能稳定运行就可以了。

我搞了一个6CM x 6CM 的大散热片来被动散热,长时间压力测试全负荷运转,也不会死机。CPU温度可以稳定在95度以下,室温20度左右。

Hi @incognito @qq20739111 ,

By following these steps, the temperature control system can be completely disabled, allowing you to manually control the CPU frequency.
通过一下步骤可以使温控系统彻底失效,并允许你们手动控制cpu频率。

rsetup > Hardware > Thermal governor > user_space

Since we need to consider all user scenarios, we may temporarily be unable to modify the current temperature control settings.
由于我们需要考虑到所有用户的情况,可能暂时无法修改当前的温控配置。

1 Like

When you first power on, you’ll notice that it doesn’t immediately drop to the lowest frequency upon reaching 60 degrees. Instead, it goes through a period of PID algorithm adjustment, and the thermal control model only then drops to the lowest frequency at 60 degrees. So, it should be able to handle instantaneous loads.
当你刚开机的时候你会发现并不是一到达60度就会降至最低频,而是经过了一端时间的pid算法,温控模型才会在60度时自行降至最低频,所以对于瞬时负载应该是可以应对的。

@ChenJaly 其实没必要限制到60度,CPU工作到七八十度,甚至偶尔九十度,也是很安全的。可以适当放宽温度限制比如在80度。

您可以看一下我发布的这个测试,系统如果加了这个限制后,CPU性能下降了一倍以上。

如果限制在60度,就算是在有散热器的情况下,多核性能几乎发挥不出来。因为这款SoC就算只有单核工作,单个大核拉满工作,都有约1.5W,60度很容易就达到了。

1 Like

@ChenJaly I can confirm that changing the thermal policy to user_space disabled throttling.

If not using rsetup, the command is:

echo user_space | sudo tee /sys/class/thermal/thermal_zone*/policy

or (on Armbian by nicka):

create an udev rule under /etc/udev/rules.d/99-thermal-policy.rules

SUBSYSTEM=="thermal", KERNEL=="thermal_zone*", ATTR{policy}="user_space"

this persists the governor over reboots.

With my heat sink it rarely exceeds 70 at full load, which is safe.

Thanks

@incognito :+1: U must have used a fan.

yes, I have a large passive heat sink (the old blue Rock 5B passive cooler) and also use a fan when needed

Hi @ChenJaly, any ETA on this new heatsink ?

Thanks

60度这个数字主要是为了和我们其他的产品行为保持一致,并非一个精调的数据

What is you product model

Using an A7S. Looking for something better than a standard 20cm fan. Something along the lines of what Radxa did for the A7A.