Linux Adding Leakshield Driver Support For Reporting Liquid Cooling System Leaks

[phoronix]

Phoronix: Linux Adding Leakshield Driver Support For Reporting Liquid Cooling System Leaks

There is pending patches for adding support for the LEAKSHIELD device to the Aquacomputer Linux device driver. German PC component manufacturer Aquacomputer has been enjoying support for their various products on Linux and the latest seeing support in this open-source driver is the Aquacomputer Leakshield for reporting any water cooling leaks and other performance metrics around your cooling setup...

https://www.phoronix.com/news/Aquacomputer-Leakshield-Linux

[aleksamagicka]

The patches are already in after v2

[rob-tech]

More hardware support is good, however, this product's existence is the very reason why my latest top of the line system has high end air cooling. After my friend completely lost his system to an AIO failure, I have vowed to never touch liquid cooling for as long as I'm around. A top of the line air cooler like my Noctua NH-D15s is only about 3 degrees hotter than a good liquid cooler anyway, I always thought it is stupid to have liquids flowing amidst sensitive electronics.

[zexelon]

More hardware support is good, however, this product's existence is the very reason why my latest top of the line system has high end air cooling. After my friend completely lost his system to an AIO failure, I have vowed to never touch liquid cooling for as long as I'm around. A top of the line air cooler like my Noctua NH-D15s is only about 3 degrees hotter than a good liquid cooler anyway, I always thought it is stupid to have liquids flowing amidst sensitive electronics.

While I would agree in general with your statements, there is definitely a place and in some cases a requirement for liquid cooling. Liquid cooling is less about getting "lower temps" so much as getting heat directly outside the box as fast as possible. In 90% of general use cases air cooling is the way to go and liquid cooling offers no advantages as you mentioned.

However if you say have a Threadripper system with dual 3090s in it, used to train ML/AI models (so basically running full blast), I can say from experience that air cooling just cant keep up. In this case liquid cooling is as far as I know the best solution... and allows cramming even more GPUs in.

In that case having this sort of leak mitigation is a huge asset!

Full disclosure... the TR system is currently fully air cooled, but when training ML/AI's we loose a lot of performance to thermal throttling so have been looking very seriously at changing it over to liquid. The Leakshield has been one of our mandatory pieces in the liquid cooling loop design.

[oiaohm]

More hardware support is good, however, this product's existence is the very reason why my latest top of the line system has high end air cooling. After my friend completely lost his system to an AIO failure, I have vowed to never touch liquid cooling for as long as I'm around. A top of the line air cooler like my Noctua NH-D15s is only about 3 degrees hotter than a good liquid cooler anyway, I always thought it is stupid to have liquids flowing amidst sensitive electronics.

This is one of those things. There are setups where the is no choice but to liquid cool. Yes Immersion Cooling or pump based cooling both have there problems.

rob-tech it pays to look at the Leakshield hardware notice a feature.

https://shop.aquacomputer.de/product_info.php?products_id=3488

This cable allows the leakshield straight to your power control button or it can send keyboard commands and so on by USB. Something about server room water cooling systems if there is a problem they power off complete racks.

AIO in desktop computers no connection to power down system in case something is wrong.

Even leak-shield connection is not ideal but it a step in the right direction. Server farms using liquid cooling also have water pressure sensors also maintain level of vacuum and normally have setup to power down to complete racks and drain water if issue is detected.

There is a problem in the consumer water cooling space of not having enough safety measures.
1) negative pressure is a safety measure to air can flow in but water cannot flow out in all bar the major fault.(leakshield does this perfectly) AIO are normally filled the same way leakshield fills with vaccume but they have nothing to maintain this vacuum or detect the loss of vacuum ...
2) system power cut method in case of detected issue.

Same company behind the leak-shield also makes USB flow sensors for water cooling loops.

3m Immersion Cooling liquid for example that following around computer is absolutely no problem its the same chemical used to make lot of PCB water resistant and is not electrically conductive..

The issue here it a lot more expensive to do water cooling loops correctly with
1)flow sensors to make sure pumps are working.
2) your negative pressure to detect and prevent leaks causing major problems.
3) auto power down systems..
4) auto drain systems

All that even at scale you are looking at 1000 dollars just doing those bits absolutely correctly.
Liquid cooling does not have to be dangerous. Problem cutting corners to save on cost.

Notice number 4 auto drain system something you don't find in consumer hardware only server farm setups. Auto drain is good leak detected shut power down and drain before the negative pressure preventing water from going where it should not fails.

[Teggs]

While I would agree in general with your statements, there is definitely a place and in some cases a requirement for liquid cooling. Liquid cooling is less about getting "lower temps" so much as getting heat directly outside the box as fast as possible. In 90% of general use cases air cooling is the way to go and liquid cooling offers no advantages as you mentioned.

However if you say have a Threadripper system with dual 3090s in it, used to train ML/AI models (so basically running full blast), I can say from experience that air cooling just cant keep up. In this case liquid cooling is as far as I know the best solution... and allows cramming even more GPUs in.

In that case having this sort of leak mitigation is a huge asset!

Full disclosure... the TR system is currently fully air cooled, but when training ML/AI's we loose a lot of performance to thermal throttling so have been looking very seriously at changing it over to liquid. The Leakshield has been one of our mandatory pieces in the liquid cooling loop design.

I have seen a few demonstrations of ducting air to the desired part, and it worked well in all cases (pun intended, I guess). I suppose at least one of your GPUs would be difficult to feed, wedged up against the other, but it could work well for your CPU and second card, if you aren't doing that now. It would take some effort and parts cost to set up, but maybe the gain would be enough that you didn't feel the need for water.

System 76's larger Thelio cases come to mind as a 'professional' implementation, as the second picture on this page demonstrates. If I recall, they use blower cards for that reason.

[polarathene]

After my friend completely lost his system to an AIO failure, I have vowed to never touch liquid cooling for as long as I'm around. A top of the line air cooler like my Noctua NH-D15s is only about 3 degrees hotter than a good liquid cooler anyway, I always thought it is stupid to have liquids flowing amidst sensitive electronics.

I ended up getting a small form factor case and ITX motherboard, and both had clearance issues for decent air cooling compatibility. I think Noctua only had 1 or 2 options that would work but they were very poor performing products (NH-L9x65 I think is one). So I could get one of those which would still be quite cramp in that small space, and may affect performance with the temps it can handle for a AMD 7700X (7900X/7950X is probably pushing my luck), or I go with an AIO which would make it possible to access other components without having to detach the cooling solution.

I've never used AIO though, and potential for leaks sound terrifying. I still haven't bought the parts beyond the case, so I could accept the loss and find a more compatible case, although I really wanted to go with a small ITX system this time around. I didn't consider the clearance issues when buying the case + MB, and it's usually just a few mm difference preventing some nicer air cooling options.

[Jonjolt]

I have seen a few demonstrations of ducting air to the desired part, and it worked well in all cases (pun intended, I guess). I suppose at least one of your GPUs would be difficult to feed, wedged up against the other, but it could work well for your CPU and second card, if you aren't doing that now. It would take some effort and parts cost to set up, but maybe the gain would be enough that you didn't feel the need for water.

This is what they do in a server system, https://www.wiredzone.com/shop/produ...6-216-213-4809

[zexelon]

I have seen a few demonstrations of ducting air to the desired part, and it worked well in all cases (pun intended, I guess). I suppose at least one of your GPUs would be difficult to feed, wedged up against the other, but it could work well for your CPU and second card, if you aren't doing that now. It would take some effort and parts cost to set up, but maybe the gain would be enough that you didn't feel the need for water.

System 76's larger Thelio cases come to mind as a 'professional' implementation, as the second picture on this page demonstrates. If I recall, they use blower cards for that reason.

Yes, air cooling can do some impressive things... with enough mass flow. 1U rack mount servers these days sound like turbine engines under load (and the fans spin pretty close to the same RPM!).

The Nvidia server GPUs are an example of this. Air cooled PCIe, but the blowers are built into the chassis. The key is they can be very loud, this is not a problem when its in a data center... more of an issue if its in a dev machine in an office.

That said I know of a large number of servers in one of the data centers here that moved over to liquid cooling the rack mount units because of the GPU heat load!

[oiaohm]

I have seen a few demonstrations of ducting air to the desired part, and it worked well in all cases (pun intended, I guess). I suppose at least one of your GPUs would be difficult to feed, wedged up against the other, but it could work well for your CPU and second card, if you aren't doing that now. It would take some effort and parts cost to set up, but maybe the gain would be enough that you didn't feel the need for water.

System 76's larger Thelio cases come to mind as a 'professional' implementation, as the second picture on this page demonstrates. If I recall, they use blower cards for that reason.

Ducting with fans does work up to a point as long as you have the airflow space and you are not trying to melt the earth..

https://www.asrockrack.com/general/productdetail.asp?Model=ROMED4ID-2T#Specifications

Yes people talk about the thread ripper. Server side ITX EPIC does exist.

Laptops, high density servers and people attempting to stuff the most powerful computer in to the smallest possible case all end in melt the earth problem where there is too much heat for fans, fins and air ducting to deal with.

https://www.froresystems.com/

The airjet tech by this company can transfer more heat into the air than old school cooling fans can but they are not scaled up enough for custom build or server systems yet.

Air cooling using fan and fins run into the same problem this has hard limit limited efficiency. Water/liquid cooling allows larger cooling area but don't solve the basic problem. The basic problem is the thermal boundary layer this limits how much heat can be transferred into air by fan and fins this is 30% cap on how much heat from the surface area can be transferred out by the fins. Airjet is interesting it tech breaks the boundary layer allowing 100 percent to be transferred to the air.

Of course Airjet as it scales up there are now other problems. CPU/GPU running at 85C with a AIrjet tech does result in the air out of it being 85C so ducting to mix the outflow with other air so it remains safe.

That said I know of a large number of servers in one of the data centers here that moved over to liquid cooling the rack mount units because of the GPU heat load!

https://lenovopress.lenovo.com/lp0636-thinksystem-sd650-direct-water-cooled-server-xeon-sp-gen-1

It not just GPU heat load. Quad CPU systems in 1U also come next to impossible to cool.by air these days.

Airjet tech long term might be able to best water cooling in consumer market. Server market where you want to get the heat out the building water cooling for different uses will remain.

Also do note with fan based air cooling GPU are getting insanely heavy. Think GPU can be 1/3 of the weight to get the same cooling by having increased cooling efficiency but then you can have very hot air to deal with or you water cool to have that mass off the motherboard.

Something that is not considered is the heavier the heat-sink solution inside a PC not mounted to the case the more likely it to be damaged in transport if the packing is not exactly right.

AIO do in theory and practice have a serous advantage when it comes loss mass inside case. Issue is lack of safety features two should be there.
1) flow sensor to detect failed pump/blocked pipes.
2) pressure sensor to detect if critical leak has happened. AIO are filled with negative pressure think tamper seal on jar(you know the one that pops up when you open jar) those are also packed under negative pressure.

Without the pump running new AIO that is not massive damaged will not leak its internal water at any major speed. Yes rainy weather more risk to computer than the AIO liquid when the negative pressure is there. Problem current AIOs you cannot tell when they have lost their negative pressure.

Another trap about current AIO is they all leak air in. So all current AIO have X number of years until they lose their negative pressure and come risk to leak without pump running with user having no way to know this has happened.. So all AIO after X years of use should be reporting critical leak has happened.

This is the problem water cooling does not have to be that dangerous. Early PC water cooling was very much use random parts with no safety features AIO are still lacking safety features. Most people doing videos online how to build a water cooling loop don't have the safety features.

When it comes to water cooling for most of consumer PC we are still in the first car made level of safety features. Yes the first car had breaks you could no trust and steering that was questionable. Air cooling like it nor that is brute forcing the problem.