FW13 HX 370 & 2x48GB RAM - Thermal throttled by RAM

[Crazyblox]

Hi folks, recently bought a FW 13, made it my first full linux _{I use arch btw} machine and it’s been awesome - it’s not been without issues though and so I thought I’d make this thread since I haven’t yet found any other owner of this spec talk about it.

The Radeon 890M feels FAST playing games like Doom Eternal on high settings at 1440x960 (Half of native display resolution), and though I know the FW 16 is more gaming oriented, gaming is not my primary use-case and I highly value the portability aspect, so I’m happy with some compromises here and there.

I installed 2x48GB of DDR5 5600Mhz CL46 RAM into my system and as I gave playing games a whirl, a few minutes in the performance would get cut nearly in half - “Odd” I thought, as battery power was at its maximum, plugged into a power source, and ventilation was clear.

With me thinking the underside air intake being a potential issue, I bought some small adhesive kickstands that sits between, but not on, the airflow grille and the expansion card ports, standing the unit up by a few cm. My theory was to prevent immediate degradation of performance when in use with improved airflow, angle my keyboard for typing, and raise the display of the laptop by a few cm when used.

Frustratingly, performance still suffered in the exact same way! I had to look into why this was happening.

_{Worth mentioning I did not know amdgpu_top existed at this point, though this would not have identified my issue.}
radeontop would show Clip Rectangle at 100% when it normally sits at ~70% before performance degrades, but nothing else - My GPU can’t still be throttling with all this air!

I ran sensors and immediately found the culprit whilst the issue was happening - my RAM was hitting 80C! This is near the provided ‘critical’ alarm level of 85C!
As I’m writing this post in Firefox with no demanding workload, the RAM sensors sit at 61C - 6 degrees over the provided ‘high’ alarm level of 55C - My RAM is running hot!.. but why?

After doing some more thinking and more digging, I realized 2 things:

• The RAM modules are not directly cooled by the fan which is connected to the CPU/GPU
• My RAM modules are dual-sided, which places chips on both sides of the module to achieve high capacity, but makes it relatively more difficult to dissipate the heat it generates.

This was pretty frustrating to learn - I had assumed my CPU/GPU would have been the elephant in the room for heat generation, and didn’t even consider the RAM - I’ll take it as a learning experience.

So now I am awaiting my delivery of non-conductive thermal pads of varying thicknesses which I’ll be able to cut to size, with the goal being to improve heat dissipation altogether - Downgrading to single-sided RAM for reduced thermals is not desirable for my use-case - especially if I want to experiment with locally hosting AI, so I hope that the combination of both the kickstand and the thermal pads will allow me to realize this machine’s full potential.

Of course, I’ll update here once I’m able to try the thermal pad solution out.

[Charlie_6]

There are black covers above the RAM slots on recent generations(not on Intel 11th etc)

↑present
↓absent

What are the usage of these covers? can these be removed to improve heat dissipation, are there negative consequences of removing them?

[Jonathan_Haas]

These covers are shielding the RAM from the Wifi signal and should probably not be removed.

[Second_Coming]

You’re probably right…“should probably not be removed”…but then I question, how’s the 11th gen able to function?

[Jonathan_Haas]

I assume they noticed malfunctioning RAM with some Framework 11th’s and then implemented the protection for later versions. It would probably work without it, but could be a tiny bit more unstable. Maybe it’s worse for the later/faster CPUs.

[Crazyblox]

Alright, I got them installed today! Here’s what I did.

As I believe the RF flaps aren’t thermally conductive, I chose to apply only 1 thermal pad for each module between the motherboard and the RAM itself, since it probably wouldn’t serve any benefit to put them on the top unless I got rid of the RF flaps, which I’m not exactly fond of.

I measured the cut of each thermal pad to the extent of the chips on the module itself:

One of my RAM modules when inserted had its sticker facing down which would have got in the way of allowing the thermal pad to be effective, so I removed it with the spudger-side of the tool that came with the laptop to improve effectiveness.

I initially tried using a 1mm thickness, but encountered strong resistance that prevented me from fully inserting the module, so I backed off and tried with 0.5mm instead. Success!

Closed everything back up, turned the laptop back on and got straight back into Doom Eternal.

Where I would have previously thermal throttled within the span of 5 minutes, I was able to play one of the longest levels without the same performance hitches I was getting before - awesome!

So from what I can tell, these changes that I have made has given the laptop enough thermal headroom for me to fully enjoy it.

FWIW, the thermal pads I ordered had a thermal conductivity rating of 12.8W/mk, which from basic research lands at the higher end of effectiveness. If you choose to do this yourself, ensure that you pick a non-electrically conductive solution and that you have the choice of experimenting with 0.5mm and 1mm thicknesses - my RAM had chips on both side of the module which factored into me using 0.5mm.

[Charlie_6]

Is there a way of stress testing the RAM? I tried $ stress -m 100 -t 60 -v that heats up the CPU while $ stress -m 1 --vm-bytes 20000000000 --vm-keep -t 60 -v occupies the RAM space but only heats up the RAM a little bit

[Second_Coming]

memtest, prime95, HPL-MxP

[Charlie_6]

I’m using 32GB single channel installed on the right side (away from the CPU) and after 11 minutes testing it’s already 59.2C

Additional cooling is definitely needed for dual channel.

Update: I already got an alarm when running $ sudo memtester 1024 10

spd5118-i2c-20-51
Adapter: SMBus PIIX4 adapter port 0 at 0b00
temp1:        +64.5°C  (low  =  +0.0°C, high = +55.0°C)  ALARM (HIGH)
                       (crit low =  +0.0°C, crit = +85.0°C)

Update: the left slot is worse

[bron]

That’s interesting. What brand of memory? I’m running 2x64GB Crucial and haven’t seen any RAM-related temperature issues.

[bron]

Oh, from the picture it looks like it’s Framework’s memory?

[Second_Coming]

@bron

I’m running 2x64GB Crucial and haven’t seen any RAM-related temperature issues.

Great to know. I’m going to order some from Amazon.ca (Currently on sale for $383.59 CAD)

[lbkNhubert]

Can you link to what you used?

[Second_Coming]

My guess is something like this:

OwlTree 3 Pack Thermal Pad,100x100mm 0.5mm 1mm 1.5mm Highly Efficient Thermal...

Or go with pre-cut, overpriced ones:

Thermalright Extreme Odyssey Thermal Pad 12.8 W/mK Non Conductive Heat...

[lbkNhubert]

Great, thank you!

[Crazyblox]

Hi, sorry for the delay!

@Second_Coming has provided some good examples, though these are the pads I bought.

Aukuoy 3 pack 12.8 W/mK Thermal Pads: 100 x 100 x 0.5 mm/1 mm/1.5 mm high...

They all more-or-less do the same thing - just ensure you get acquainted with what specs are used in this space aside from abiding by the general rule that the higher W/mK the better.

[Crazyblox]

Just want to provide an update here from my initial thermal pad installation…

It has definitely helped and provided that extra bit of overhead, but in workloads where RAM access is incredibly optimized there is still evidence of thermal throttling.

In this case, I have been giving the latest DOOM entry ‘The Dark Ages’ a go; completed it on my desktop PC and is incredibly fun, but out of morbid curiosity I wanted to see how far I could push this system still…

The game’s rendering pipeline 100% relying on raytracing means that RAM access gets absolutely hammered, and I could happily say that it’s engine idTech8 is one of the most optimized raytracing-based rendering pipelines for gaming.

Playing on lowest/off settings w/ XeSS Ultra Performance & lsfg-vk at 1440x900 16:10 (Near 1440x960 3:2, but not advertised as a mode by the display despite being x0.5 native resolution) allows me to play the game at acceptable levels of detail under the context of 30W.

‘lsfg-vk’ is a Linux port of ‘Lossless Scaling’ which performs much faster due to switching to vulkan, and when not thermally throttled I am able to play with a base FPS of 40, with an output FPS of 120.

The way thermal throttling manifests in Doom TDA is that the framepacing kicks the bucket. The base FPS flip-flops every split second between 40FPS and 20/30FPS - I assume that this is in sync with the system polling the RAM thermals, and that it’s consistently teetering between the throttling threshold.

I’m going to explore further solutions - my initial thermal pad install was quite conservative given that there’s another side of the RAM still yet to have thermal pads applied to - the RF flaps are the current concern for if they would hinder heat transfer (Not removing them to sacrifice stability for more performance), but sandwiching it between 2 thermal pads to spread heat to the upper section of the laptop might be viable.

Will post back if I make any improvements in the situation.

[Charlie_6]

I have a silly idea. Drill a hole or grill on the chassis where the WiFi card sits on, to make a secondary intake to allow air to flow from the WiFi card past the RAMs to the fan

[Bernd_Steinzimmer]

Are you sure the RAM is throtteling, how do you know that? Normally the CPU will throttle because of high CPU temps.

[Grandpa_engineer]

because the bandwidth on the ram gets cut by half or more