Clevo W230SD high GPU temp (GTX 960M)

I have finally found the reason for my poor performance in BF3.

The GTX 960M starts to throttle its way down to 405Mhz as it reach 90-92c.

It starts off at 1202Mhz, then throttling 1000, 800, 600, 500 and stops att 400mhz. I assume its because off high temperature, but how hot do this usually get in this machines?

The GTX 960M is the same chip as the GTX 860M, so if anyone has this setup, how hot does yours get?

And att what temp is the GTX 860/960M suppose to start throttling?

 

I have now repasted this thing 6-7 times, using different methods and amounts of paste.

The pipe coming from the gpu is barely hot, while the cpu pipe is really hot to touch, which indicate that the cooler may lack contact on the gpu or its missing the liquid inside. I have tried various amount of paste, but it does not change anything. I'm using Gelid Extreme.

The strange thing is that it seemed to work fine for a couple of days.

Any idea?

 

I have figured out the problem.

Its the copperpipe coming from the GPU. The pipe from the GPU only gets lukewarm, while the CPU pipe is really hot.

If the PC lies flat, the temp is around 60c in BF3 on the GPU. If it sits up toward my legs, the temp rises until 90c. This sounds really weird, but i have played for 30min now and tested, and this is actually the case.

I guess the issue is the liquid inside the copperpipe on the GPU side.

 

Are you gaming with it on your lap? It could be that you are blocking a vent with your leg possibly and causing the higher temps that way.

 

No, thats not the issue. When i have it sitting up its actually a good airgap underneath. When i have it laying flat on my leg there is no airgap, yet the temperature is barely 60c on the gpu in BF3. Sitting up, gpu is hitting 90-93c.

I think this sounds strange myself, but it have to be the liquid inside the gpu copperpipe, i cant see any other explanation.

 

Yes, that makes sense. Holding the laptop at angle, left side higher than the right, ought to drop temperatures significantly. In that position there's a greater gravimetric potential; the water, condensed near the fan, drops more readily to the bottom, where it does most good.

Keeping it on the lap we tilt it toward us, rear side higher than the front. This system's design can't cope with that:

Spoiler: Clevo W230SS - heatpipe design

 

Well it's not water but the idea is correct. Heatpipes are impacted by angle but if there is internal wick damage this can mean this is amplified.

 

I think the heatsink is defective... I would push to have Sager replace it...

 

Ok, the problem is as i said, when having the keyboard facing straight at me, having it raised up against my leg while sitting/laying in bed.

I have comtacted the place where i bought it, so im gonna send it back and get a new one, just to be sure. If the new one behave identical im just gonna keep it and be aware.

However, the CPU pipe handles this angle just fine, so i think the gpu also should.

 

There will be a crash angle for each pipe where performance will drop drastically.

 

You cannot lift the laptop rear edge more than 15 degrees or so before performance plummets on the W230 laptop.

 

Wow I learned a lot reading this topic. So, for best thermal performance, I would need to ideally use my W230SD laptop with it's front tilted up, so that the highest point of the heatpipe is on the cooler? That way the "cold" condensed fluid goes down in the direction of the CPU/GPU, while the hot vapor generated goes up in the direction of the cooler?

This is quite interesting, I had no idea this effect occured here. I'll be sure to play angling my notebook around when it arrives haha.

 

The pipes will be designed to perform best flat or with the rear just slightly raised.

 

Yes, that's what you'd expect they'd aim for when designing the thing ... but oddly enough ... did a quick check (loosely, uncontrolled circumstances) using an XTU 5 min. stress;

Spoiler: Clevo W230SS - temperatures at angles

Of course, bit moot without logging fan speed as well ... but upside-down the fans clearly ran at their lowest rpm's too ... so if you're keen to get top-of-the-list on those benchmark chart thingies . Kinda makes sense though; the die is upside-down on this model (and on most other designs), so it's at the ceiling of the heatpipe. Better fan efficiency too, obviously.

Hmm ... maybe a proper riser would've helped the 'flat-situation' ... in case anyone else wants to check; attached the sheet (XenForo = no more attachments?) uploaded here, it expects a Core Temp log.

 

That's not surprising as what you are basically testing is how hard the fan has to work to draw in air, with it upside down there is no restriction on the air flowing in.

 

Yes, though think the pipe-die location may make a (tiny) improvement as well. Can't verify that using a raiser; replaced the 5V fan with a 12V model (though running at 10V). Did, however, make another two 5 min. runs, one with the '12V' flat (really 10V), and 10V upside-down:

Spoiler: Clevo W230SS - temperatures 5V fan and 12V fan (at 10V)

Still a small, 2°C difference, but that's negligible.

That 5V-10V gap though ... crikey . Idle is 35 °C now (used to be 50 °C), both at max performance settings. Of course, this is a ~30 cfm@12V fan, compared to the regular ~10 cfm max found in laptops. New fan table would be in order though; these things do much lower rpm's, so the higher range won't be used and there's too few steps at low temperatures.

 

Do you have some pictures of the setup?

 

Yes, but ordered a 70mm gpu/server fan (6.8W@10V) and a 5.5W Delta to replace the current 12V (4W). Will make a quick guide with the final fan, whichever comes out on top (and a graveyard pic of the fans that didn't survive ... ).

Presume the gpu centrifugal will do best (rated 45 cfm). It may be a tad noisier though, but the stock was no picnic, either. To boot, these things push much greater volumes at lower rpms, so it's a different kind of noise.

 

Well with more CFM comes more noise

 

Be sure what heatsink close good enough and what you use bot so much thermal compound In that way you get bad cooling transfer and overheating in the end

 

And gaming systems with one fan is sucks

 

Not by default, it depends on the design.

 

Finally arrived and got around to this. Still a work-in-progress, mind:

Spoiler: ADDA 12V 0.68A GPU fan (AY07212UB220B00) - test setup

Spoiler: Delta 12V 0.55A desktop fan (AFB0712VHB)

Reason for that was to check which fan does a better job in this specific system (W230ST/SS/SD). Measured distance at which it blew over an object; no contest, the squirrel cage (blower type) wins hands down (also beats the stock fan pushed to 12V).

Minor hiccup; fan was bit taller than the 15mm advertised, so needed some adjustments to make it fit:

Spoiler: Clevo W230SS - Clevo W230SS - stock fan and housing removed, surface sanded

Spoiler: Clevo W230SS - stock fan (2W) replaced with gpu fan (8.2W)

Spoiler: Clevo W230SS - fan spacing enlarged

Spoiler: Clevo W230SS - fan spacing enlarged, side view

Now, there's a few final details that need sorting before this is ready for market:

1. Fan guard (duh)
2. Something to partially re-cover the hole (maybe)
3. Proper fan control

Not sure whether '2' is necessary, but noticed some of the AMD designs have roughly half of the blade area covered. Not all cards use this, however. Anyway, it's a simple thing to check which works better.

The real effort is '3'; RPM/temperature table is way off and without bios-accessible table it'll need a hard-mod. Could go for a trimmer, but that needs user-intervention, not looking forward to that. So will aim for a dual setup: A.) NTC thermistor and B.) a full-on/off bypass (simple switch).

The resistor will be attached flush against the cpu heatsink; it heats up -> resistance will drop -> power to fan increases. Most difficult bit is to find the specific curve necessary; ideally you want ∞ Ω @ 40°C and zero Ω @ 90°C (anyone have experience with these things?). The switch will normally be ' off', but when set to ' on' the fan will run full throttle, regardless of what the thermistor is doing.

A snippet of arcane knowledge; the sense/tachy wire needs to receive a signal of minimum rpm's at a specific temperature. Running without fan or low rpm model, temperature reaches ~45°C and the three leds will start blinking some 20 time before it does a hard shutdown. Not a problem; just need to make sure the replacement fan's tachy wire is hooked to the motherboard fan connector.

 

I believe having it covered partially helps direct all the air through rather than being thrown off at funny angles.

 

Yes, that theory holds (unfortunately). Quick mock-up with some cardboard showed a ~10% difference in favour of covering the blade-area. Simple plastic cover will do, nothing fancy.

Also made some progress as to finding the proper thermistor; apparently a Steinhart–Hart equation works best for that. Made a sheet whereby using the desired values (custom fan table) it's possible to enter manufacturer specs. and generate a plot:

Spoiler: 12V, 0.68A fan - NTC thermistor curve

So the NTCLE100E3221***-series is the closest match for this specific fan. Problem now is where to find one ... ' one' as in 'not-10,000'.

 

Yeah, certain parts can be a pain to get if not ordering in bulk. I run into same issues with ICs.

 

I would love to see updates as you get this mod smoothed out. What kind of temps do you get running everything full tilt?

 

Don't know yet; old 12V@10V was about 13°C lower, but that was a simple (and less effective) case fan. Can't measure the new 45 cfm blower; it's hooked up to the regular 5V connector atm. Kinda annoying to run that thing full throttle all the time, want to have fan control sorted first.

Did have some luck with Mouser though; seems ' quantity = 1' is possible.

Scratch that; overlooked the ' R at zero-load' detail (this thing would've gone ' woosh' ...).

Most NTC thermistors, the E3221 included, are intended for low-current applications (measurement only). Setup as a fan controller would be as an in-line resistor, so the current driving the fan goes through them (non-zero load). The resistor needs to be able to handle that (Pmax ≤ Wfan). Fortunately, there's a special type of NTC resistor we can use: an ' inrush current limiter'.

Because there's a current, we have the following effect:
Current goes in -> temperature ICL increases -> resistance drops -> current increases -> temperature ICL increases -> resistance drop ... ad infinitum.

It's not necessarily thermal runaway (since the self-heating effect tapers off due to the non-linear nature of an NTC thermistor), but it does mean a high dissipation factor is very important. This is basically the capability of the ICL to off-load current-generated heat to its surroundings. In a typical AC-DC adapter a higher value would mean it'll be slower to respond to power-on, but also faster to cool down (so it can be used again). Ergo; you can potentially kill any adapter by quickly un- and re-plugging it from the mains (while connected to pc/laptop etc.). The ICL would be heated already and so effectively useless to limit peak inrush current.

Another important factor that determines the DF, rated resistance and Imax is the ambient temperature (standard is 25°C). Since we're attaching the ICL to the heatsink, it'll have the following effects:

1. The dissipation constant will be much higher; we're interfacing solid resistor with copper, instead of normal air (sounds familiar, that ...). This is effect is positive for our intended use.
2. Tamb. is roughly equivalent to Theatsink; most heat will transfer through the heatsink, rather than the surrounding air. This is a negative effect as it limits Imax according to the derating curve.

Couldn't find the dissipation factor air vs. solid, but 'stirred oil' is roughly 8x that of air, so we'll go with that. Think this is everything required to calculate the R-T and, hence, fan table (reversed R, really). Uses 21x iteration to factor in self-heating property:

Spoiler: ICL fan table (beta)

Last values that needs sorting is heatsink's Tmin and Tmax; can't assume it'll be as high as cpu/gpu temp so this needs measuring (cheap thermometer's in the mail). And, of course, the proof's in the pudding; the ICL picked will need a dry-run, think just hook it up to 12V battery, 12V fan and dip it (coated) in a few glasses of water, heated to varying temperatures.

Lastly, reconsidering a manually varying resistor after all (instead of simple on/off switch). Only problem is that small rheostats are prohibitively expensive; the best/smallest option would be a 7.5W Ohmite RCS150 series; quantity 5 again (for $155). Maybe hunt the scrap market ...

 

Slightly off the current topic, but back to the original one :

Does everyone observe this drammatic behavior with their W230 (if tilted, cooling is so bad that it throttle) ?

I had a look a while ago, on how this heatpipes work, and I'm surprised that a tilt can have such an impact ! Normally the liquid that forms on the cold end, can go all the way toward the hot end, whatever the position of the heatpipe thanks to capillary force (action of surface tension), so normally gravity as a negligible impact.


I think that his heatpipe was simply dry and/or defective. Am I wrong?

 

Wicks that aid the return of the liquid can be gravity dependant.

 

My W230SS is very prone to overhearing when tilted also. I think it is normal for these models. Doesn't bother me though. A bigger fan is quite interesting though. Especially if fan speed can be controlled well.

 

I recently bought a clevo w230sd and I'm really astonished by the temperature difference in this graph. I've seen you did heavy hardware modifications to get 12V fans running. But I'm not exactly sure about this post:
1) did you simply put a 12V fan in the pin port of the clevo mb and you found out it uses up to 10V?
2) I know about your problems in getting the bigger fans control to work. Does this apply to the ~10V fan too?
3) if yes, do you think prema would be able to make the necessary BIOS mods necessary to support fan controls of the new model?

 

Sorry to double post but I am seriously intentioned to improve the average temperature of my new clevo w230sd. I have already quoted you where you explain your experiments with the *ss variant but maybe it's passed unobserved. It seems you are the man when talking about fans and hardware experiments. So, can I switch the default fan of the w230 with a 10V one successfully?

Sent from my A0001 using Tapatalk

 

No, that is limited to 5V, like most devices in a laptop. I cheated and soldered wires to the battery connector. Since there's 3x 3.7V 18650 cells in series, that gives 11.1-12.3V, depending on charge level (at 100% the cells' voltages are higher). And, obviously, plenty of amperes; there's 2 series and each can output 11V@5A.

Actually, that graph is nothing like the final product; that was not from a battery-run, but a re-loop over 2x5V, but with very little amperes (~50-60% of max. only). The real thing is ... far out . Hadn't much use for that kind of cooling power; my 4900MQ only TDP-throttled. There was no way around that (barring an MX), but seems the new ThrottleStop has a solution, so ... proper motivation.

The 5V fan connector is controlled, but the battery, of course, is not. The EC has a temp->rpm lookup table, so you want a fan that has the same or lower maximum rpm than that of the original. Think Fn+1 bypasses table entirely, simply releasing the floodgates. This would disregard whatever the fan's maximum rpm is, so that'd be a last resort for any 5V fan.

Yes, with a custom EC, but only for the standard 5V-port. For a high-end 12V fan, however, the 5V-port is all but useless as that can only provide ~2.5W. It'll run, but performance will be worse than the stock fan; it's a lot heavier, so more mass to throw around and no juice to get it going.

Promised a guide and and will do so soon(-ish). Received two leviathan ICLs (dimensions!!) for the auto-fan control, but also need something to switch off the circuit when shutting down the laptop. Could use physical switch, but not looking forward to manual intervention. Now, have some ideas for automation, but first order of business is to replace the battery pack; it's ageing and only has about ~25% capacity left. With ' replacing', that means soldering, of course.

Spoiler: confession

Not terribly good at ' finish one thing first, then start something new'. Get bored rather soon and when something more interesting happens along ... Did finish one small project, though; the humming slimline adapter is dead silent now. Deserves it, too; world's smallest 120W ( pretty expensive, as well), so the noisy capacitor rather spoiled the effect. Kinda hard-to-find; most are three-prong and larger, this is the tiny two-prong. That also means no earth (it's double-shielded), which is a very good thing; no distorted audio when using low-impedance iems or headphones.

Spoiler: silenced and nearly indistinguishable from new ^^

Only took four stages ... :

1. 1 hour for dismantling
2. 5 minutes ordering new film capacitor
3. 4+ months 'other stuff' (<- distracted)
4. 1 hour fix + re-assembly (tricky; cramped space)

Won't be four months this time, expect 2-weeks/1-month for the fan mod. Keen to complete this; have something much better lined up. Will think of that as a 'reward' .

 

Just please be careful if soldering connections that can deliver a LOT of power

 

Was fixing dead image links, so thought to upload some additional photos as well:

Spoiler: Clevo W230SS - 11.1V power source - 1

Spoiler: Clevo W230SS - 11.1V power source - 2

Spoiler: Clevo W230SS - new fan cover (aluminium)

Spoiler: Clevo W230SS - rubber feet replaced

The last adjustment was necessary to accommodate the aluminium cover. Added a little extra height to get more airflow and cut the front feet a little lower in order to tilt the keyboard in a nice angle. Used a dremel to sand off the old rims and get a larger, flat surface. The feet were cut from an old rubber mallet and fixed to the bottom with super glue. Has been this way for over a year, transported weekly and they've held together perfectly.