Posts by Brandon Clark

1) Questions and Answers : Unix/Linux : Looking for cooling ideas (Message 2009360)
Posted 26 Aug 2019 by Profile Brandon Clark
Post:
Hello group,

I've been trying to get an image of the cluster posted but have been having difficulty getting it to show up. Stay tuned on that . . . .

I agree about starting off with air cooling for the time being. I picked up some USB powered fans and have those loosely extracting warm air from the area of the computers, and it is making a noticeable difference. I could feel the heat buildup between the cases before, but now the air space between computers is back to room temperature. The combination of leaving the case doors off and adding some extra fans is indeed helping.

I'm going to see if I can make some kind of enclosure around the boxes next, and position the fans at each end. That will make the whole assembly a lot less unsightly, as well as facilitate cooling. I also like the idea of keeping some computer components in jars of oil for a while to see how they perform. I may go that route as well.

I'll see if I can get the photos up in the near future as well.
Thanks everyone!
2) Questions and Answers : Unix/Linux : Looking for cooling ideas (Message 2008357)
Posted 19 Aug 2019 by Profile Brandon Clark
Post:
I was wondering about that myself: which components might be affected by the oil. Most oils act as a solvent to some degree, and it's an unknown as to which components might be affected by immersion. I wonder if the board could be sprayed with some kind of sealant before immersion to reduce the effects. Even something simple like a clear krylon spray paint might provide a decent barrier to any solvent action, but should still transfer heat pretty well.
I read somewhere that anything with moving parts needs to be kept out of the oil. I'm guessing that fans and hard drives built for use in air end up burning out when trying to churn through the much heavier oil?

Yeah, on that laptop the graphics are a separate card. It is integrated into the motherboard, but is indeed a separate device. (I had to take that machine apart to replace the CPU and GPU cooling paste, and it was an ugly project to get inside there.) That laptop is another of the salvaged ones I picked up at my prior job.

For the cooling project I'm going to pursue two different routes simultaneously:
1. Open up the computers that are currently running and remount them inside some kind of enclusure where I can use external fans to force air through the whole enclosure. That should help with the "microclimate" around the systems and allow the intrgrated fans to relax a bit.
2. Take a few of the other systems I haven't put into use yet and experiment with running them in oil.
2.1 Remove the fans from a system (CPU and power supply) and see if it will even work in an oil bath.
2.2 If 2.1 is successful for a week or two, take another system and spray paint everything I can get to with a clear coat. Then see if one system dies before the other, or if they both happily crunch along with the air-cooled
systems.

Now I just have to figure out how to get a raspberry pi connected to temperature sensors and in-line ammeters. Then I could do an apples to apples comparison of average temperature, life cycle, and power use to see if two oil-cooled systems do better than air-cooled ones. lol.
3) Questions and Answers : Unix/Linux : Looking for cooling ideas (Message 2007544)
Posted 15 Aug 2019 by Profile Brandon Clark
Post:
Rob,
That's interesting about the built-in Intel GPUs not being all that useful. I've come across posts to that effect before, but haven't pursued it further. I have one salvaged laptop with a built-in Nvidia card that seems to be a good processor. It shows up as "NVIDIA Quadro K2100M (1999MB) driver: 390.11 OpenCL: 1.2" in my computers listing, and is one of the best performing systems. I should probably leave that one turned on?
The only other two machines that are doing GPU computing are two windows laptops. I'll try turning off the GPU computing on those machines and see if it makes much difference in the computing work.
Problem is that in the main cluster (I still need to get a photo up) all of the boxes are running Ubuntu, and so none of those are doing GPU computing. They are all little machines that don't even have a PCI slot on the boards.

Keith,
I may start out with the fan idea. We've recently had an especially "warm" summer here in Anchorage (mid 70's) so all the local stores have been sold out of fans, lol. I should be able to scare up something though.

All,
I did some more looking around online and came across mineral oil-submerged systems. It's probably overkill for what we do, but that would be pretty cool. I'm going to do some more research and then build two proof-of-concept systems. First I'll use a raspberry pi, and then I'll use another one of these salvaged desktops that I have. If it all works I might be able to move everything into a big tank and then use a fish tank bubbler to bubble cool air through the oil, as well as set up basic circulation. I'l post about it when I get further along.
4) Questions and Answers : Unix/Linux : Looking for cooling ideas (Message 2007186)
Posted 13 Aug 2019 by Profile Brandon Clark
Post:
Hello group,

I've been repurposing old hardware and putting it to work for SETI, but now I think I'm running into cooling issues. I need help with some ideas on how to approach the problem.

The setup: I have a cluster of six (soon to be seven) small Dell desktop machines sitting next to one another. They are mini-ATX-ish size, or something like that. They each use their own power supply that is in the case. Each case has a fan on the CPU and another one built into the power supply. Everything is running Ubuntu with a base install and nothing special going on with the hardware (no overclocking, etc.).
I'll try to post a photo this evening when I get home.

The problem: My power bill has been increasing noticeably, and I think that a big part of it is the cooling requirements. All the individual power supply and case fans run 24-7. I've taken to leaving the side panel off of the cases, which seems to help a little bit. I need to come up with some kind of cooling plan for the system as a whole though, rather than relying on each system to cool itself.
My hope is that some kind of overall cooling for the whole cluster will help dramatically, and allow the individual case fans to slow down. Hopefully that will be more efficient that having all the individual fans running continuously. I'm accumulating the hardware to bring online a second identical cluster in a month or two, so I want to solve this issue before that.

The design: not sure yet. Any ideas?
I'm leaning towards building some kind of enclosure that I can mount the desktops into, and then use a few USB fans to blow air over the whole cluster. My hope is that would help remove heat more quickly and allow the case fans to slow down, with a net increase in electrical efficiency.
Eventually I would like to get something like a truck tool box and locate the cluster outdoors. I live in Alaska and we get about 4-5 months where the temperature never gets above freezing. For the fall-through-spring seasons that might be a good way to keep things cool. That doesn't help me for the summer months though.

Other (possibly nuts) ideas:
-Remove everything from the individual cases and build some kind of structure to hold the motherboards and associated hardware out in the open, kind of like a mining rig. (Could really help with cooling, but would be quite an intricate project too. Could be expensive, and definitely would be time-consuming.)
-Do the above, but submerge the whole setup in some kind of fluid cooling. (Not sure what to use for the working fluid though.)
-Run more than one machine off larger, externally-located power supplies. (Would this give me more efficient use of power, rather than having one power supply for each system? How would the wiring work?)
-Buy a chest-high freezer at Costco, bribe my girlfriend with jewelry so I can set it up at her house, and house the cluster inside it. (The jewelry purchase might be more expensive than the current electricity cost though.)

What do you guys think?
Brandon
5) Questions and Answers : Unix/Linux : What GPU to buy? (Message 1999530)
Posted 25 Jun 2019 by Profile Brandon Clark
Post:
Special Sauce application? I'm not familiar with that.
6) Questions and Answers : Unix/Linux : What GPU to buy? (Message 1998980)
Posted 21 Jun 2019 by Profile Brandon Clark
Post:
That's good to hear. I just don't have any experience doing GPU computing with BOINC yet, so it's new to me. Never been a heavy gamer either, so graphics cards are not my strong point.

It was a surprise to me to find that the same chipset gets used across many different video cards. I always figured that whomever made a video card made all the critical components. Thinking about it now though it isn't so surprising. The same thing happens with motherboards: an Asus motherboard might have Sandisk memory and an Intel processor.
7) Questions and Answers : Unix/Linux : What GPU to buy? (Message 1998891)
Posted 20 Jun 2019 by Profile Brandon Clark
Post:
First time posting here . . .

I'm interested in this question too. I am putting into service two older Dell machines that I got hold of, and would like to add a GPU to them. Currently I have 10 machines running BOINC, but none of them have a GPU for processing (just onboard graphics, which don't seem to get used much under Ubuntu).

The question I have is: if the same chipset (GTX1060, for example) is used in multiple different video cards from multiple different manufacturers, how would I decide what card to buy? I cna work through differences such as power requirements and physical size, but what I'm wondering about is compatibility with the BOINC software under Unubtu.

How would I know that a given card will work, even though it has a recommended chipset?
Many thanks,
Brandon Clark





 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.