Jump to content
Search In
  • More options...
Find results that contain...
Find results in...

CONSULTATION NEEDED - SUB 0 LIQUID COOLED DAILY DRIVER BUILD

Go to solution Solved by swangotango,
On 1/24/2021 at 9:38 PM, svmlegacy said:

Edit 1: I should also mention that the NAND of NVMe storage is not fond of cold temperatures, and you may see write performance suffer if you choose to put an NVMe drive inside the freezer chamber.

I'll just insulate it. Or worst case put this on it https://www.omega.com/en-us/industrial-heaters/surface-heaters/flexible-heaters/srfra-srfga/p/SRFGA-104-5?gclid=CjwKCAiAu8SABhAxEiwAsodSZHIlhESvfaZlw5u6-fEfef7kUJSWYrsnh8py0R9A9vHt1r0gwzVkgRoCULcQAvD_BwE&gclsrc=aw.ds

 

You guys don't seem very solution oriented. Obviously this all can work and it can work with some diligent napkin math. It's not rocket science. 

CONSULTATION NEEDED

 

I plan on building a complex cooling system in my garage to keep my GPU and CPU around 0* C on a daily basis without condensation concerns, and while minimizing energy consumption concerns. I am hoping this will give me the fastest, stable, convenient and "practical" computer in the world in gaming workloads. I plan on running the I/O through the wall to my office which is backed up against the garage. Before you doubt me, know that a PC with 850W power consumption (higher than most high end single card rigs) emits 3000 BTU/Hour under full load. I have done the math and I believe the solution I've come up with should be sufficient.

 

 

Obviously this is not worth it, and is impractical but it sounds fun. I have a lot of other details worked out as well but don't want to go in to depth on everything at this moment.

  • Does anyone have a sense of how a Ryzen 5950x will preform at -10* C to 0* C on a consistent basis? 10%, 20% even 30% uplift in benchmarks?
  • Does anyone have any ideas as how to make this better?
  • Hit me with every criticism you have! I am curious where the flaws are!
Link to post
Share on other sites
44 minutes ago, swangotango said:

<snip>

People are seeing 4.6 GHz all-core P95 at 57°C PBO enabled, leaving 300 MHz tops for further scaling without manual OCing (I think PBO2 allows extending the limit, but ymmv). You're doing this for much less than 10% performance gain.

Sub-ambient is something that should be reserved for manual overclocking. I doubt you'll see more than 10% gain without comprimising the longevity of the CPU.

 

1/4 HP motor does not mean 1/4 HP of heat transfer. as far as I can tell, they don't list the cooling capacity of the unit. It could be more or less than the motor's power draw.

Also, you'll need to analyze the performance at your desired temperature split. As you mentioned, you're aware, but chillers loose a lot of efficiency as the split increases. There's zero point in taking a blanket number without context.

Your 900W chillers, that's very peak performance.

image.png.61461c95f4d8357ad729a9a0940fc8ea.png

 

Dessicant does virtually nothing in sub-zero temperatures.

 

20 - 30 gallons of coolant for "thermal mass" is completely and uttery excessive. If it's for the freezer itself, ice packs would work much better, as that will take advantage of the latent heat in freezing.

 

One last comment.. just because your incoming coolant is -10°C does not mean that your CPU core will stay at 0°C. Particularly with small lithographies, the heat density is intense, and will easily make a significant temperature split from the cold plate.

 

Edit 1: I should also mention that the NAND of NVMe storage is not fond of cold temperatures, and you may see write performance suffer if you choose to put an NVMe drive inside the freezer chamber.

Main: AMD Ryzen 7 1700X, Nvidia GTX 780, 16 GB 2667 MT/s DDR4 Fedora 33 x86_64

Secondary: Intel Xeon X5670, Nvidia GTX 660, 24 GB 1333 MT/s DDR3 Windows 10 Home x86_64

Server: Intel Xeon X5670, 60 GB 1333 MT/s DDR3-R Ubuntu Server 20.04.1 LTS x86_64

Laptop: Intel Core i5-3320M, 16 GB 1600 MT/s DDR3 Fedora 33 x86_64

Link to post
Share on other sites
2 minutes ago, svmlegacy said:

People are seeing 4.6 GHz all-core P95 at 57°C PBO enabled, leaving 300 MHz tops for further scaling without manual OCing (I think PBO2 allows extending the limit, but ymmv). You're doing this for much less than 10% performance gain.

Sub-ambient is something that should be reserved for manual overclocking. I doubt you'll see more than 10% gain without comprimising the longevity of the CPU.

 

1/4 HP motor does not mean 1/4 HP of heat transfer. as far as I can tell, they don't list the cooling capacity of the unit. It could be more or less than the motor's power draw.

Also, you'll need to analyze the performance at your desired temperature split. As you mentioned, you're aware, but chillers loose a lot of efficiency as the split increases. There's zero point in taking a blanket number without context.

 

Dessicant does virtually nothing in sub-zero temperatures.

 

20 - 30 gallons of coolant for "thermal mass" is completely and uttery excessive. If it's for the freezer itself, ice packs would work much better, as that will take advantage of the latent heat in freezing.

 

One last comment.. just because your incoming coolant is -10°C does not mean that your CPU core will stay at 0°C. Particularly with small lithographies, the heat density is intense, and will easily make a significant temperature split from the cold plate.

 

The 20 - 30 gallons of coolant is to mitigate the other issues you spoke of. It seems ridiculous, but if you know you have 20-30 gallons of coolant stored at -25* before you turn on the computer, then you know you have the ability to run with liquid at that temperature for at least some time.

 

As far as dessicants go, I was worried about this. I found that some dessicants work better at lower temps than others. Do you have any other ideas for eliminating freezer frost? That's one reason I didn't want to use Ice, I wanted to limit the moisture that can dissipate into the air in the freezer. I was considering buying a large dehumidifier for the garage that it will be in. Have any sources to back up your claim?

 

1/4 HP does have some bearing on the power of the unit. We know it must be around 1000BTUs an hour. This is a well reviewed and high end freezer for lab use. So we know that it must have decent cooling power. It takes a surprisingly large amount of BTU to cool down food products when they are put in the freezer. About 1000BTUs/hour. So we can trust this number somewhat. I'm guessing the total system will have about 7000 BTUs/Hour cooling capacity at 25*C with no temperature delta. Maybe closer to 1000 BTUs/Hour at 0*C. This will still be enough to slow down the coolant temp rising from -25*C to 0*C. The sweet spot for benchmarks will be when the system has been off for a day, and has just been turned on. 

Link to post
Share on other sites
6 minutes ago, svmlegacy said:

 

 

The best thing you can do to mitigate frost is to not open that freezer. Dehumidification in the garage will help for when you must open it.

 

If you're relying on thermal mass for cooling, you're really putting the point of the chillers in question. Talk to Koolance and get some real data on the machine to design by. This is pure speculation on your part regarding its performance at nearly minimum set point. If you're going to do that, at least run a trial on a single unit to validate its performance at the split you're trying to hit.

 

Good to know about the dehumidification.

 

I want to go with these chillers because I know they have decent capacity and minimum set points, but also they have pretty trick software to allow me to control them with the computer. It seems like this is the be approach, also considering how affordable they are, to go sub zero for any amount of time and run very cold the rest the time. If you can point me to a chiller that is more powerful with the ability to have it turn off when the computer turns off, but still under $4000 I would be happy to see it.

Link to post
Share on other sites
9 minutes ago, swangotango said:

As far as dessicants go, I was worried about this. I found that some dessicants work better at lower temps than others. Do you have any other ideas for eliminating freezer frost? That's one reason I didn't want to use Ice, I wanted to limit the moisture that can dissipate into the air in the freezer. I was considering buying a large dehumidifier for the garage that it will be in. Have any sources to back up your claim?

I work in an industry that cares very much about the humidity in air, and we consistently target a dewpoint of -10°C. The water content of the air below zero is very very small. Check a psychometric chart for verification of this. Frost builds up from the warm wet air entering the freezer when it's opened, which dews out when it goes below zero.

 

The best thing you can do to mitigate frost is to not open that freezer. Dehumidification in the garage will help for when you must open it.

 

9 minutes ago, swangotango said:

1/4 HP does have some bearing on the power of the unit. We know it must be around 1000BTUs an hour. This is a well reviewed and high end freezer for lab use. So we know that it must have decent cooling power. It takes a surprisingly large amount of BTU to cool down food products when they are put in the freezer. About 1000BTUs/hour. So we can trust this number somewhat. I'm guessing the total system will have about 7000 BTUs/Hour cooling capacity at 25*C with no temperature delta. Maybe closer to 1000 BTUs/Hour at 0*C. This will still be enough to slow down the coolant temp rising from -25*C to 0*C. The sweet spot for benchmarks will be when the system has been off for a day, and has just been turned on. 

If you're relying on thermal mass for cooling, you're really putting the point of the chillers in question. Talk to Koolance and get some real data on the machine to design by. This is pure speculation on your part regarding its performance at nearly minimum set point. If you're going to do that, at least run a trial on a single unit to validate its performance at the split you're trying to hit.

 

Main: AMD Ryzen 7 1700X, Nvidia GTX 780, 16 GB 2667 MT/s DDR4 Fedora 33 x86_64

Secondary: Intel Xeon X5670, Nvidia GTX 660, 24 GB 1333 MT/s DDR3 Windows 10 Home x86_64

Server: Intel Xeon X5670, 60 GB 1333 MT/s DDR3-R Ubuntu Server 20.04.1 LTS x86_64

Laptop: Intel Core i5-3320M, 16 GB 1600 MT/s DDR3 Fedora 33 x86_64

Link to post
Share on other sites
19 minutes ago, svmlegacy said:

I work in an industry that cares very much about the humidity in air, and we consistently target a dewpoint of -10°C. The water content of the air below zero is very very small. Check a psychometric chart for verification of this. Frost builds up from the warm wet air entering the freezer when it's opened, which dews out when it goes below zero.

 

The best thing you can do to mitigate frost is to not open that freezer. Dehumidification in the garage will help for when you must open it.

 

If you're relying on thermal mass for cooling, you're really putting the point of the chillers in question. Talk to Koolance and get some real data on the machine to design by. This is pure speculation on your part regarding its performance at nearly minimum set point. If you're going to do that, at least run a trial on a single unit to validate its performance at the split you're trying to hit.

 

Would this be better: https://www.usalab.com/usa-lab-40-c-20l-recirculating-chiller-uc-20-40-30l-min/

Link to post
Share on other sites
1 minute ago, swangotango said:

As I mentioned before, you need to get proper datasheets for performance at the temperature split you're designing at. I can't tell you how either performs.

Main: AMD Ryzen 7 1700X, Nvidia GTX 780, 16 GB 2667 MT/s DDR4 Fedora 33 x86_64

Secondary: Intel Xeon X5670, Nvidia GTX 660, 24 GB 1333 MT/s DDR3 Windows 10 Home x86_64

Server: Intel Xeon X5670, 60 GB 1333 MT/s DDR3-R Ubuntu Server 20.04.1 LTS x86_64

Laptop: Intel Core i5-3320M, 16 GB 1600 MT/s DDR3 Fedora 33 x86_64

Link to post
Share on other sites
7 hours ago, swangotango said:

Yes it would work better but now you are not so power efficient.  You are looking at a 164lb cooler running off of 220v and 20 amps.    If you are going to cool it to -40C here is what you need.  (and you will because a chip running and the cold loss through the motherboard, to get 0C at the junction, the coolant temp will need to be much colder)  Keep in mind it says it can do 12,400 BTU which is over 3K watts, the colder you go the less cooling capacity it has.  As chillers drive colder they loose the amount of watts they can drive because it gets harder for the compressor.  Industrial chillers will supply a datasheet with a temp versus watts curve so you can find the load you need at the temp you need and see if it compatible with the application.  Notice that system says that the -40C was taken at 0 load.  What is the duty cycle on the system?  I know of systems right now that will say they can do -40C and while they can, they can't hold it for very long or they have such low cooling capacity at that temperature they are practically worthless. 

 

CDA (Clean Dry Air) or synthetic air or pure N2, something that doesn't have water in it.  Running parts cold for any length of time will cause icing if they have any exposure to moisture.  This is not something desiccant can do.  Your system needs to be in a box with positive pressure of the CDA and if you want to ever open the box you need to let the system warm up.  Don't forget all of the insulation you will need around the tubing outside of the box.  You wouldn't believe how big and fast the ice balls can form on an exposed -40C pipe or panel.

 

What kind of fluid does this system use?  If you run industrial temp fluid, which this system might need, they can be over $1000 USD a gallon.  Galdon and Florinert are two examples of such fluids running in an industrial cooling system.  Even getting them can be hard if you are not tied in with industrial suppliers.

 

Real industrial coolers that keep chips at -40c and will for days at a time cost around $35K USD on up and you need a constant supply of CDA rated at -70C or better.

 

At my work we do this very thing for days at a time on various chips.  Icing and getting chillers that won't overheat after 6 hours is a big deal. 

 

Link to post
Share on other sites
15 hours ago, vsteel said:

What kind of fluid does this system use?  If you run industrial temp fluid, which this system might need, they can be over $1000 USD a gallon.  Galdon and Florinert are two examples of such fluids running in an industrial cooling system.  Even getting them can be hard if you are not tied in with industrial suppliers.

 

Real industrial coolers that keep chips at -40c and will for days at a time cost around $35K USD on up and you need a constant supply of CDA rated at -70C or better.

 

At my work we do this very thing for days at a time on various chips.  Icing and getting chillers that won't overheat after 6 hours is a big deal. 

 

Automotive coolant will be fine for -20c. 

I don't need additional air blown into the box, I just need to keep the box below freezing and keep it sealed to outside air. Might need to beef up the seals a bit.

Link to post
Share on other sites

-> Moved to Custom Loop and Exotic Cooling

^^^^ That's my post ^^^^
<-- This is me --- That's your scrollbar -->
vvvv Who's there? vvvv

Link to post
Share on other sites
On 1/24/2021 at 9:38 PM, svmlegacy said:

Edit 1: I should also mention that the NAND of NVMe storage is not fond of cold temperatures, and you may see write performance suffer if you choose to put an NVMe drive inside the freezer chamber.

I'll just insulate it. Or worst case put this on it https://www.omega.com/en-us/industrial-heaters/surface-heaters/flexible-heaters/srfra-srfga/p/SRFGA-104-5?gclid=CjwKCAiAu8SABhAxEiwAsodSZHIlhESvfaZlw5u6-fEfef7kUJSWYrsnh8py0R9A9vHt1r0gwzVkgRoCULcQAvD_BwE&gclsrc=aw.ds

 

You guys don't seem very solution oriented. Obviously this all can work and it can work with some diligent napkin math. It's not rocket science. 

Link to post
Share on other sites
9 minutes ago, swangotango said:

You guys don't seem very solution oriented. Obviously this all can work and it can work with some diligent napkin math. It's not rocket science. 

 

On 1/24/2021 at 11:21 PM, svmlegacy said:

you need to get proper datasheets for performance at the temperature split you're designing at.

 

Yes, it can work. I highly reccomend proving the concept and learning from a small system, as well as trialling the components you're planning to use, before moving into a full-fledged system like this.

 

As far as the NAND flash comment, I had added that simply as a reminder. Something to consider and think about. Obviously it isn't a huge issue.

Main: AMD Ryzen 7 1700X, Nvidia GTX 780, 16 GB 2667 MT/s DDR4 Fedora 33 x86_64

Secondary: Intel Xeon X5670, Nvidia GTX 660, 24 GB 1333 MT/s DDR3 Windows 10 Home x86_64

Server: Intel Xeon X5670, 60 GB 1333 MT/s DDR3-R Ubuntu Server 20.04.1 LTS x86_64

Laptop: Intel Core i5-3320M, 16 GB 1600 MT/s DDR3 Fedora 33 x86_64

Link to post
Share on other sites
On 1/25/2021 at 8:52 PM, swangotango said:

I don't need additional air blown into the box, I just need to keep the box below freezing and keep it sealed to outside air. Might need to beef up the seals a bit.

 

Even with really good seals you are going to ice up the box.  You have to get the moisture out before everything goes cold. 

 

You are greatly underestimating the amount of energy required to dissipate this much heat and how fast ice will grow. 

 

Link to post
Share on other sites
6 hours ago, vsteel said:

 

Even with really good seals you are going to ice up the box.  You have to get the moisture out before everything goes cold. 

 

You are greatly underestimating the amount of energy required to dissipate this much heat and how fast ice will grow. 

 

ok so tons of desiccant for when it is room temp, let it sit locked closed for a couple days, then turn on the freezer. Also maybe some added silicone on the seals.

 

Link to post
Share on other sites
On 1/27/2021 at 5:32 PM, swangotango said:

I'll just insulate it. Or worst case put this on it https://www.omega.com/en-us/industrial-heaters/surface-heaters/flexible-heaters/srfra-srfga/p/SRFGA-104-5?gclid=CjwKCAiAu8SABhAxEiwAsodSZHIlhESvfaZlw5u6-fEfef7kUJSWYrsnh8py0R9A9vHt1r0gwzVkgRoCULcQAvD_BwE&gclsrc=aw.ds

 

You guys don't seem very solution oriented. Obviously this all can work and it can work with some diligent napkin math. It's not rocket science. 

It isn't that it can't be done. It is the upkeep and risk vs reward that make it an issue. 

 

My solution and a project I was doing but decided to give up on was very similar to this one... except I was going to use something like a raspberry pi to constantly monitor water temps and ambient temp/humidity to keep temps below ambient, but also above the dew point.

 

In the end it would need to occupy a closet or other remote space due to the noise level. Then it would be terrible in terms of cost effectiveness and lastly each loop would need to be changed as to not waste the energy used to keep lower temps.

 

My project was going to be for 4-5 computers, but having to pull radiators and losing the ability to easily use the computers elsewhere just didn't feel worth it. 

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Newegg

×