The God of Computer Coolers

[AlexTheGreatish]

It is nearly impossible to cool Intel’s 13th gen CPUs.. so we got a 5000W industrial chiller.

 

Check out the Thermo Scientific NESLab ThermoFlex 5000 Chiller: https://lmg.gg/7IIR9

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

[Ender Wiggin]

I have a S&A CW-5200 that I am using to cool my new rig. Wish I had gotten it setup before this video dropped. Here is my PC part picker list with the parts I am using.

There is a dedicated loop for the PC and using a heat exchanger for the water chiller. That way I don't have to worry about the issues of debris or other things being introduced into the loop. 
I got the water chiller just to move the heat of the computer out of my room. I would ether place it right outside or in the attic. I had also planned to plumb a radiator to the chiller to chill the air in the server closet. I wish I had gotten the dual loop version of the water chiller to do the server room but just did not think about it. The server closet gets quite toasty so it is much needed improvement.

 

[manikyath]

so.. mostly 'to lead by example' for people aimlessly yammering about video quality.. i'm noting the details where i felt the video was lacking, hoping that it turns into something constructive for the community as a whole.

 

- a lot of the 'down low' shots are quite shaky. and in fact.. a lot of the shots just arent very "steady" as a whole. i understand the freedom of a camera operator holding the camera, but it just feels sloppy.

- the "substantial improvement" at 12:44 is just not displayed at all. did the score improve? or was it just the suggested 67°c?

- for the on-screen images.. please get a screen recording of some variety going and overlay it on the video, it would be much less jarring, especially when the camera operator is shaking.

 

other than that - "cool" video  - i hope to see this chiller appear more regularly.

 

bonus content idea - perhaps give this boring grey thing a makeover? dBrand sponsorship and everyone wins?

 

also - the "alex special extension cord" is a great meme.. but please just stick a propper plug on this thing.

[papajo]

1 hour ago, AlexTheGreatish said:

It is nearly impossible to cool Intel’s 13th gen CPUs.. so we got a 5000W industrial chiller.

 

Check out the Thermo Scientific NESLab ThermoFlex 5000 Chiller: https://lmg.gg/7IIR9

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

This bad girl just BEGS for



+hydra  (dont forget tune the vrm voltage and co values etc in bios before running hydra, there is a guide if you are not sure what to do) + liquid metal extreme 

oh an pair it with the lowest latency (not just cl) 6000 memory or higher 

oh and obviously a low noise/interference  mobo with top vrm + a beefy  psu that can connect its 12v rails for oc

[Luscious]

These would work best in a temperature controlled room, avoiding condensation issues. That's actually why servers have hot+cold aisles, since pushing targeted cold air from the floor below is simpler than plumbing tubes and chilling equipment, never mind cost/maintenance of blocks when dealing with hundreds of processors. TrippLite actually sells a 7000 BTU 8U rack mount AC unit that sits on the bottom of the rack and forces cold air up in front of the rack. It doesn't need 208V either since runs on regular 120V 15A, important for residential/office use. It also only costs US$1000 as opposed to the $12K price tag of this chiller - major savings!!! Combined with a stacked sandwich of chunky 360 rads/fans you could probably achieve very similar results, never have to worry about condensation, run dual D5 pumps for redundancy, and enjoy the added benefit of more coolant volume to absorb the thermal load and run your system QUIET. I have roughly a gallon of EK coolant inside my loop and it's enough to keep a 5960X and four 3080Ti's happy.

[swimtome]

 @AlexTheGreatish  @Aprime augh what WAS the "substantial improvement"?

[papajo]

34 minutes ago, Luscious said:

These would work best in a temperature controlled room, avoiding condensation issues. That's actually why servers have hot+cold aisles, since pushing targeted cold air from the floor below is simpler than plumbing tubes and chilling equipment, never mind cost/maintenance of blocks when dealing with hundreds of processors. TrippLite actually sells a 7000 BTU 8U rack mount AC unit that sits on the bottom of the rack and forces cold air up in front of the rack. It doesn't need 208V either since runs on regular 120V 15A, important for residential/office use. It also only costs US$1000 as opposed to the $12K price tag of this chiller - major savings!!! Combined with a stacked sandwich of chunky 360 rads/fans you could probably achieve very similar results, never have to worry about condensation, run dual D5 pumps for redundancy, and enjoy the added benefit of more coolant volume to absorb the thermal load and run your system QUIET. I have roughly a gallon of EK coolant inside my loop and it's enough to keep a 5960X and four 3080Ti's happy.

I dont think that they mean to use this as a cooling solution in the sense of having a machine that will keep operation critical (or at least like every day use) equipment at a reasonable temperature.

I think they got it just to like test/OC stuff by using this instead of e.g liquid nitrogen or what not. 


e.g the machine you are talking about (7000 BTU ) in theory can absorb up to 7385391 joules per hour or 2051.5 whathours of thermal energy..

The machine tested in the video can do double that and then some or in other words it could cool the same thermal load with the same properties (e.g IHS thermal conductivity thickness volume etc) to the same temperature delta as the 7000 btu method you describe, but significantly faster. 

[tkitch]

28 minutes ago, swimtome said:

 @AlexTheGreatish  @Aprime augh what WAS the "substantial improvement"?

They did show Cinebench scores, and said they made it past 40,000 which was an improvement over the base watercooled performance.

 

Edit:  36,600 (something) was the base score:  See 12:25 in the video, getting past 40K is a 10%+ improvement which is very significant.

[idmb]

I've been waiting for Alex to do this for YEARS. Nice to see safety glasses when playing around with it.

 

But when are you modding the CPU to improve thermal contact? 

[Shortyman]

I have a background in HVAC and would recommend trying to slow down the water flow through the block in the computer by closing the ball valve on the supply line Never slow it down by the return line. Sometimes to much water flow does not allow enough transfer of heat.

 

Just a thought

[Bitter]

1 hour ago, Shortyman said:

I have a background in HVAC and would recommend trying to slow down the water flow through the block in the computer by closing the ball valve on the supply line Never slow it down by the return line. Sometimes to much water flow does not allow enough transfer of heat.

 

Just a thought

Normally that could be a worry but in this instance I think the water is moving so fast that boundary layer isn't a concern.

 

@AlexTheGreatish Next stop de-lid and direct die cooling, all aboard wooo wooo on the overclock train!

[swimtome]

19 hours ago, tkitch said:

They did show Cinebench scores, and said they made it past 40,000 which was an improvement over the base watercooled performance.

 

Edit:  36,600 (something) was the base score:  See 12:25 in the video, getting past 40K is a 10%+ improvement which is very significant.

My quote is from 20 seconds later from just chiller before overclocking. No score shown. 

[ImorallySourcedElectrons]

7 hours ago, Bitter said:

@AlexTheGreatish Next stop de-lid and direct die cooling, all aboard wooo wooo on the overclock train!

They should get a lapping and polishing machine next and thin those dies down to <80 ųm You can pick up the logitech ones for less than 4k EUR on eBay if you look around a bit, with the added benefit that you can confuse the hell out of everyone.

[tkitch]

6 hours ago, swimtome said:

My quote is from 20 seconds later from just chiller before overlooking. No score shown. 

They state various scores they got in the video.  

From a baseline of what it got watercooled with a pair of 360mm Rads, they increased by over 10%.  WTF are you looking for exactly? 

[ToboRobot]

Standardized water temp testing sounds most useful... but using that chiller to cool a 1u gaming server in a rack is probably the most fun way to deploy this.

[Needfuldoer]

So, when is this thing going to become the heart of Whole Room Watercooling v2?

[Bitter]

They'll have a cold computer but a hot room lol. I wonder how hard it would be to tinker with it and have it run sub zero (Celsius) chilled fluids? Not that hard I assume to make a lil Arduino or something to offset the temp sensor to fool it into doing as you wish. How cold depends on the refrigerant used of course.

 

@AlexTheGreatishnew video series about hacking the chiller, I'll watch.

[tkitch]

47 minutes ago, Bitter said:

They'll have a cold computer but a hot room lol. I wonder how hard it would be to tinker with it and have it run sub zero (Celsius) chilled fluids? Not that hard I assume to make a lil Arduino or something to offset the temp sensor to fool it into doing as you wish. How cold depends on the refrigerant used of course.

 

@AlexTheGreatishnew video series about hacking the chiller, I'll watch.

not worth the risk of the condensation on tubes and the like.  

[Bitter]

1 hour ago, tkitch said:

not worth the risk of the condensation on tubes and the like.  

Absolutely worth the risk. The machine doesn't stop at dew point, there's already the strong possiblity of condensation. 

[GodAtum]

On 3/8/2023 at 7:33 PM, Ender Wiggin said:

I have a S&A CW-5200 that I am using to cool my new rig. Wish I had gotten it setup before this video dropped. Here is my PC part picker list with the parts I am using.

There is a dedicated loop for the PC and using a heat exchanger for the water chiller. That way I don't have to worry about the issues of debris or other things being introduced into the loop. 
I got the water chiller just to move the heat of the computer out of my room. I would ether place it right outside or in the attic. I had also planned to plumb a radiator to the chiller to chill the air in the server closet. I wish I had gotten the dual loop version of the water chiller to do the server room but just did not think about it. The server closet gets quite toasty so it is much needed improvement.

 

Thats really interesting. Why do you need a pump when you have the water chiller already? How many watts does the system pull on average?

[Ender Wiggin]

3 hours ago, GodAtum said:

Thats really interesting. Why do you need a pump when you have the water chiller already? How many watts does the system pull on average?

Was not sure if I was going to have enough flow if I were to place the water chiller in the attic. Also just helps me test. I could also possibly use it to add a second loop for the server room. If I were to do it again I would just buy the upgraded version of the water chiller pump (cheaper from china). 

At idle I don't expect more than 200 or so watts (this would be high). Under load, I plan to be able to deal with at least 1000 watts. I don't think I will ever hit this number but it is a good safety margin. In the end, this decision was made for me by the water chiller. The 5000 series water chillers are the smallest and cheapest that used a phase change cycle. The 5200 happen to be the same price as the 5000 so I went with that and thought the extra headroom would just have it run less often. I was also thinking of having an external tank inside to increase the volume of water so that it would not need to run as often. I am going to manage the temperature setpoint so as to not hit the dew point inside where the computer is placed. As I plan to run this setup whenever I am on the PC I don't want to run the risk of killing my components due to condensation.

 

Aquarium water chiller talk: 

Spoiler

I looked at a number of aquarium water chillers and they did not inspire confidence. Not only were they overpriced as hell they tend not to be rated for continuous loads. Not to mention the cooling capacity was also way worse. In the end, water chillers designed to operate continuously for something like a laser cutter made a lot more sense.

The new crop of battery-operated coolers is promising but not rated for continuous loads. If you ever made a phase change cooler back in the day, you would know, if you used minifridge components it would die in a couple of months. You had to find a cheap window unit AC and those could run for hours on end.

Misread question response: Why two loops?

Spoiler

The water chiller is technically an open loop that could let debris in more easily. I also don't want to have to manage all the materials in both the PC cooling loop and the water chiller loop. All the material in the PC loop will be stainless or copper. It is rather easy to find stainless hardware due to food safety stuff made for homebrewing. I get the secondary advantage of being able to run classic antifreeze coolants so that if the temps drop outside (where the water chiller is) it won't affect the operation of the loop. I can then add whatever coolant I want for the PC loop.

 

[Commodore256]

I've talked to somebody that complained about this video and said the CPU was overvolted beyond need and by throwing that many volts at the CPU, you lowered the life span of the CPU for no reason and you would have gotten better temps and better turboing at a lower voltage while still being over volted.

[Spindel]

The thermal cycling from that cooler would kill the cpu much faster than having it running at 100 degC constantly. 

[tkitch]

50 minutes ago, Spindel said:

The thermal cycling from that cooler would kill the cpu much faster than having it running at 100 degC constantly. 

Why?

[ImorallySourcedElectrons]

41 minutes ago, tkitch said:

Why?

As a word of warning, I'm grossly simplifying and skipping over a few important details here to give a general idea of what it looks like. For a modern CPU the case is quite a bit more complicated especially if things like EMIB are considered where silicon dies in multiple technologies are stacked and bonded together in combination with through silicon vias (TSVs) and other craziness like that.

 

Let's take a simpler chip than a CPU (source: https://electroiq.com/ti-ships-40-m-fine-pitch-copper-pillar-flip-chip-packages/

Let's go over the parts:

• The grey square at the top is the silicon die, that's the actual chip.
• Bumps are little metal structures grown on chips to make the contact pads come above the surface, looking at it from the top before the chip is mounted, they look a little like this:
  
  (source: https://electroiq.com/ti-ships-40-m-fine-pitch-copper-pillar-flip-chip-packages/)
  The top of this bump (called a pillar in this instance) is covered with a little bit of solder.
• The interposer is the large square thing that you insert into the socket, it's usually made from a cheaper material and has multiple layers of metal to make electrical interconnects. You can see this as a really advanced circuit board. The interposer is usually made out of something like alumina ceramic, FR-4 circuit board material, or polyimide circuit board material. 
• The interposer has metal pads at the top and bottom to make electrical contact. The above image has solder balls at the bottom, but this could just as well be pins or just large metal pads like you see with intel CPUs. 

To assemble this thing, the chip is basically pressed into the interposer and heated very rapidly to the melting temperature of the solder. Creating an electrical and mechanical connection between the chip and the interposer. Now as to why dramatic temperature changes are really bad for packages like this: Consider that a 1 m long piece of silicon expands by about 3 µm per degree celsius. This means that when a 20 mm CPU that suddenly goes from 20°C to 100°C expands by about 5 µm. But those little metal bumps are very bad at transferring heat, it's a bit like dipping a spoon into hot coffee: the tip of the spoon gets hot but it takes several minutes for the part you hold to warm up. So the interposer doesn't really expand, and this causes very large forces on those tiny little solder joins you can see:

(source: https://electroiq.com/ti-ships-40-m-fine-pitch-copper-pillar-flip-chip-packages/)

 

To solve this issue, we add a material between the chip and the interposer to fill up these gaps and to take the mechanical load off the little bumps and the solder joints. Basically, we're gluing the chip to the interposer. (That's the gooey stuff you see around the sides of the CPU die in pictures of decapped CPUs.) This "underfill" as we often call it, absorbs the mechanical stress and ensures better heat transfer between the chip and the interposer. 

 

Sadly, this doesn't solve the problem, on average alumina expands 3.5ish times more than silicon for a similar temperature increase (depends on the allotropes if you want to go down that rabbit hole), high quality FR-4 expands 4 times more than silicon for a similar temperature increase, particular grades of polyimide expand around 6 times more than silicon for similar temperature increase. So while your CPU die might expand by only 4 µm, the interposer is now expanding anywhere from 14 µm to 24 µm. So you still have a significant amount of stress, it's just spread out a lot better.

 

If the temperature change is more gradual everything has time to heat up and cool down, reducing the total difference in thermal expansion at any point in time. As a result, the damage caused is generally less with slower ramping and cooling, and you run less risk of things like the glue losing adhesion to the CPU or interposer. This also has direct implications for thermal management through die-thinning (reducing the thickness of the chips), because as you thin a die down you lose the ability to do certain of these things without harming your yield significantly or causing a lot of crib deaths. There are a lot more things to consider here, like the fact that the thermal expansion of many interposer materials can be quite uneven, or that the interposer might increase a lot in thickness - which means you have to avoid hot spots, does the die have any cracks in the side (aka, which dicing method was used), etc. These sort of things are the dark underbelly of electronics that folks outside of the industry rarely get to see.

 

 @AlexTheGreatish: Has anyone at LTT ever considered trying to get in touch with Amkor to see if you can get a factory tour there? They developed a lot of the crazy packaging you see in things like CPUs and SoCs, and they aren't restricted by ITAR like a lot of the competing manufacturers.