Lenovo thinkstation p320 tiny

[Pob255]

I'm currently testing a p320 tiny at work and encountering a couple of odd issues.

First I must say wow this thing is tiny and is packing an i7-7700t, 16gb ddr4, m.2 ssd and a quadro P600 2gb graphics card

First thing I noticed was when I reimaged it, it got very hot.
When doing some testing I've noticed a couple of odd behaviours.
CPU always runs at max speed, max turbo speed so always runs at 3.8ghz even at idle, under full load this will drop 2.8hgz.

There are a couple of issues going on, to start with speed step doesn't seem to work, it is on in the BIOS turning turbo off in the bios it just sits at 2.9ghz at idle.
This unit does run toasty cpu 55-60c range at idle peaking to 80-85c under load, changeing the bios fan setting to performance knocks it down to 50-55c idle, 75-80c load + fan more noisy

Under load with bios fan setting to quiet, the cpu will thermal throttle hard, Aida64 showing spikes of up to 50%
with fan on performance setting, thermal throttleing spikes of 20%
What is odd is spikes in the cloock speed, under load it will drop to 800mhz in spikes, I first thought this is thermal throttling as they occure more often on the lower fan speed setting, however I'm also wondering if it could be power related? could it be the VRM's?
ALso to note, sometimes the thermal throttling is more constant (shown in Aida64) and the clock speed dips don't occure/occure less often.

All told while the specs are great and it's a tiny system I think it's just too much for the small cooler to handle.
The cooler is fins, heat pipes and a blower fan, like a reference graphics card, with a seperate heat pipe connect a large aluminium block on the graphics card (covering gpu+ram+vrm)
If they made the case a bit thicker so the cooler could be a bit bigger this would be a great machine, as it stands the i5 version might be better.

[scaryjim]

... CPU always runs at max speed, max turbo speed so always runs at 3.8ghz even at idle, under full load this will drop 2.8hgz.

There are a couple of issues going on, to start with speed step doesn't seem to work, it is on in the BIOS turning turbo off in the bios it just sits at 2.9ghz at idle....

That sounds likely to be a power profile issue - what OS are you running under? I've heard of a few instances of default power profiles locking CPUs into their lowest power state: locking into the highest would be surprising, but on a workstation the default might be high performance...

It'll also be worth checking the GPU power states too - it sounds like the system uses a single connected cooling system, and even at idle a GPU locked in highest power state will contribute to the thermal load.

I think for any small form factor you're going to hit thermal limits under load, even with a low TDP CPU. This is going to be exacerbated if the OS power profile isn't dropping the CPU (and potentially GPU) into idle states: the platform won't cool down properly when idle, so will have much less scope to utilise the turbo states when it's under load.

EDIT: to give you a quick idea - my laptop has a default power profile which enables speedstep, and at idle my cores sit between ~800MHz and ~1200MHz and ~ 45 C. If I drop into the high performance power profile, the core clock locks at 3.2GHz/1.1v, and the temperatures very quickly jump to mid-50s C - despite the CPU still being idle.

EDIT 2: if I use power saver, it appears to lock in a maximum multiplier of x9, meaning the CPU never runs above 900MHz and the temps actually fall back to just over 40 C.

[Pob255]

Yes I've been thinking that.
As I've mentioned the cooling is combined.
GPU idle runs ~50c (about 5c below the cpu) @139mhz
using aida64 gpu only load goes to ~65c @1620mhz (cpu goes to 65-70c and starts doing the 800mhz spikes & 15ish load)
On the quiet fan setting

OS is windows 10 64bit enterprise
and I forgot about the windows power settings that is why the cpu is running at max turbo all the time.
Still when win is set to max cpu at all times it makes me think that the drops to 800mhz (always 798mhz to be precise) is power related not thermal.
More testing is needed

[scaryjim]

Sounds like the GPU power management is working correctly at least

... Still when win is set to max cpu at all times it makes me think that the drops to 800mhz (always 798mhz to be precise) is power related not thermal. ...

Interestingly I saw that on my laptop too even at idle, with speedfan reporting temps in the mid-50s, so it could be to do with power delivery; perhaps a momentary voltage spike or dip that the CPU reacts to by dropping into the lowest available power state until it registers a stable voltage again. tbh I don't think it's much to worry about; I'd rather the CPU reacted to transient power delivery issues than just tried to barge through them

[Pob255]

Changed windows power setting to ballanced
clock speed now goes between 898mhz and 3.5-3.6ghz so speed step is working, if a bit twitchy
cpu temp is 50-54c constant flux
gpu sits at a slow flux of 46c to 48c and constant 139mhz

Looks like the brefest spike of cpu load causes a clock speed spike, where as the gpu ignores tiny spikes.
Unfortunatly as this belongs to work & is in warranty I cannot pull the cooler off juryrig something better and seperate to test if the original dips are caused by heat

[scaryjim]

... Looks like the brefest spike of cpu load causes a clock speed spike ...

Yeah, that's pretty standard with responsive boost modes. Even my old A10 4600M did basically the same thing. You'll only stay in the bottom power state if the system is literally doing nothing, and that's basically never on a Win 10 install...!

I believe the point is to minimise task energy, so you boost up to as high a level as possible to complete the task as quickly as possible and get back into idle mode.

[DDY]

Hot voltage regulators may also be part of the problem here.

The hotter VRMs get, the lower their maximum current output and sometimes voltage stability is impacted too.

The system may limit CPU boost or under-clock to cope with the reduced amount of power available when the power supply circuitry gets too hot - even if the CPU is cool.

SFFs are characteristically thermally constrained and typically have poor airflow over the motherboard components. Not only that, VRMs don't always have heatsinks and given the lack of airflow rely on thermal conduction to the motherboard for the bulk of its cooling.

It's not just CPUs of course, the same thing happens with GPUs and indeed other devices.