AVALON ASIC - QCool element watercooling thread

[dogie]

....
3D printing does NOT create perfect surfaces, and every method has trade offs. The most widely available method (material extrusion) wouldn't print in acetate (not many things will) and would have a major problem being sealed even with a gasket and post processing.
....

Another advantage with an o-ring over a gasket is they can take up higher tolerances, a 4mm cross section o-ring would allow a max surface irregularity of about 1mm at low pressures, that should be plenty for a decent diy 3d printer.

Materials would be a big issue though, I don't do a whole lot with plastics or moldings so generally just use off the shelf stuff like acetal for machined parts. Are there any plastics that would be suitable for 3d printing with the temperatures these things could reach?

If you say they can do 1mm then that would be fine. The typical DIY printers use material extrusion as a process. Typical materials include ABS, polycarbonates, acrylics. Nothing you can extrude using FDS (fused deposition modelling) will be suitable, as the whole point is they turn into mouldable semi-liquids at 70-90C. If your coolant temp is at 50-60C, its too close to maintain rigidity and mechanical properties indefinitely.

Acetate blocks like this honestly wouldn't cost much to have made - or even machine yourselves rather than print. You don't need them to be intricate at all, literally just a cube.

[dogie]

For Avalon heat level, maybe 3 to 4 parallel water channel is OK. With a DDC pump and 240/360 radiator.

I'm actually a bit worried about this. In my computer building days (1k gaming machines) it was always recommended that 240mm was a minimum for 200W+. A 360mm double thickness radiator would only have a max output using 40CFM fans of 450W. a 480mm 600W.

If we're trying to remove ~500W internal heat, its really pushing it to use a 360mm - even with all other things like flow working well. Remember when watercooling systems overheat - shit goes down. Your pump can overheat as it uses the coolant to cool itself, your seals, glues, gaskets etc can leak like hell, your coolant looses some of its properties and strange things can happen. Your blocks themselves can start warping, especially acetate ones. And then your Avalon will [hopefully] power down to due overheating.

Also as a final dampener, remember that a pump and fans is going to be adding another 50-60W to the power bill.

[dogie]

OP,

The simplest and most efficient design would be parallel channels.

The block can be Alu, dont listen to anyone and think cooper is a must. Use Alu with brass radiator. Its so simple you can make that block by hand.

Keep the cost down.

Having parallel channels is worse than the current design (assuming he changed them parallel to the longer axis). You'll only be running one loop as it is unless you plan on multiple pumps and radiators, which you won't be.

So you have to split and rejoin the flows. This incurs quite significant minor losses, decreasing the heat transfer capacity of the system.

[Icoin]

fail.....
you need to maximize the surface area and create turbulence, this gives you 'mixing' events inside the liquid and increases your cooling

What might work is something like a miniature heatsink fins fixed to the bottom plate with the top plate fitting over them so that there are fins inside your flow channel, that would maximize your surface area.

Heat is absorbed by the bottom plate, into the fins, which then have the water swirling about them.

Thats pretty much what dogie already brought up and icoin responded to, something like fining on the plate is needed to increase surface area. The form of the channels could also be modified to promote better flow conditions in the fined areas.

Just noticed something in the photo's with the port positions, they would hit either the case or the electrical connectors. Guessing the ends of the plate are the only safe option for them.

EDIT: Was forgetting the heatsinks go on the back of the boards so there's room for the ports at the top (behind the power supply sections), that means the plate would have to be machined to only contact the underside of the chips too though :/

The idea is to hook the cooler in direct contact with the chips, but you bring me back to my original single unit type configuration A Avalon module is made out of 8 single units mount on a cooler. Each unit can have its own QCool element with in and outlet that would mean 8 times in and 8 times out.  24 or 32 connectors for just outlets needed.
Can you please make a drawing, of how you envision the connection to the water reservoirtank and radiator - a forsided bar drilled, with 8 1/4" inlets and one 1/2" outlet ?

We actualy have no solid data to work with, thats why it would be great if someone of the existing AVALON users could post some measurements.
Thank you!

My questions to BitSyncom still are:

A. Are we gonna be able to get just the PCB parts from AVALON without the aluminium stuff, PSU and fans?

B. Will the design of 28nm be similar to avalon so the cooler can be reused?

[Tamerz]

The block can be Alu, dont listen to anyone and think cooper is a must. Use Alu with brass radiator. Its so simple you can make that block by hand.
Keep the cost down.

Except why would you want to lock out using PC cooling hardware which is already abundant and easy to buy? It will give a lot more flexibility in the long run.

[Icoin]

The block can be Alu, dont listen to anyone and think cooper is a must. Use Alu with brass radiator. Its so simple you can make that block by hand.
Keep the cost down.

Except why would you want to lock out using PC cooling hardware which is already abundant and easy to buy? It will give a lot more flexibility in the long run.

Yes indeed:


This is allready in use of GMP.

Maybe you find something that would fit to the modules, but i guess a made for Avalon is the bether option. When there are professional cooler producers why dont they come up with a commercial Avalon watercooling solution, but the point here is to stay open source.

[Nemesis]

OP,

The simplest and most efficient design would be parallel channels.

The block can be Alu, dont listen to anyone and think cooper is a must. Use Alu with brass radiator. Its so simple you can make that block by hand.

Keep the cost down.

Having parallel channels is worse than the current design (assuming he changed them parallel to the longer axis). You'll only be running one loop as it is unless you plan on multiple pumps and radiators, which you won't be.

So you have to split and rejoin the flows. This incurs quite significant minor losses, decreasing the heat transfer capacity of the system.

Wow ,.... you're dumber than i thought. You got reading comprehension problem?

This is what i said, MEng eh?

[Nemesis]

This is the raw study of the QCool watercooling element for AVALON ASIC. Since we dont have the device present yet, and we had to estimate the distances it is only possible to present a raw drawing. As soon we can take measurements on the module, the QCool element will be completed and published.
The price is estimate to be 3 BTC for a 3 modules Avalon Asic.

I have a masters in Mechanical Engineer and 7 years experience in cooling design and that plate is a bit weird. You're going to be losing SO much head pressure and slowing your flow down massively - which is something designed to happen in the radiator, NOT the block.

You want as much fluid ramming through the block as possible as opposed to it slowing down and heating up too much. In the radiator you want it slowed down so it gets as close to ambient as possible.

Take a look at typical GPU flow paths. They have minimal pathing and interference:

http://sphotos-a.xx.fbcdn.net/hphotos-ash4/402876_474673442586692_1610541966_n.jpg
http://www.techpowerup.com/img/06-10-12/GPUblock_Side.jpg
http://www.google.co.uk/url?sa=i&source=images&cd=&docid=yIMMU6W0vyjfTM&tbnid=DhAugiqOkm2ZcM:&ved=0CAUQjBwwADhz&url=http%3A%2F%2Fekwaterblocks.com%2Fuploads%2Fimages%2FVGA-FC.jpg&ei=U7ZaUdOMEqjF0QWkhYH4Aw&psig=AFQjCNHre9qLYOiZ1y0-iV9gJ2tJFWpsBA&ust=1364985811358312
http://www.bestgraphicscard.net/wp-content/uploads/koolance-470-water-block.jpg

You can use riffles or pins like in CPU blocks to increase surface area if your simulations (...I assume you are using CFD...) need more heat transfer, but please dont meander the flow channel.

http://www.overclock3d.net/gfx/articles/2008/03/11094211657l.jpg

Its a shame you claimed to be a MECHANICAL ENGINEER.

Your post is a joke right?

In a closed loop, IT DOES NOT MATTER WHERE THE PRESSURE LOST IS.

I would start asking your university for a refund

xD Of course it does! If you have massive minor losses [the frictional losses due to flow geometry] then you lose head pressure.

Lower head pressure -> lower volume flow rate
Lower volume flow rate -> lower heat transfer rates at block and radiator
Lower heat transfer rates at block and radiator -> lower system heat transfer at given temp
Lower system heat transfer at given temp -> worse cooling system
Worse cooling system -> money printer not as cool

You want all of the pressure loss to be in head pressure and major frictional losses [frictional losses due to the liquid flow].

Should i tip you for your college fund (read your sig)? mr. MECHANICAL ENGINEER? I dont think so because you're being ripped off .....

I never said pressure loss doesnt matter. I said i have a problem with this statement (especially from someone claimed to be MEng, while still asking for college fund...)

Code:

You want as much fluid ramming through the block as possible as opposed to it slowing down and heating up too much. In the radiator you want it slowed down so it gets as close to ambient as possible.

In a closed loop, you care only the pressure loss of the whole system. Your flowrate b4 and after block is the same. The statement above is the most BS i've read. Not surprising tho, i got that alot on forums when their first sentence is " I'm an..... Engineer"

[Nemesis]

The block can be Alu, dont listen to anyone and think cooper is a must. Use Alu with brass radiator. Its so simple you can make that block by hand.
Keep the cost down.

Except why would you want to lock out using PC cooling hardware which is already abundant and easy to buy? It will give a lot more flexibility in the long run.

MOST PC watercooling rads are BRASS radiators.

What make you think i meant something else?

[dogie]

Any thoughts on bonded plastics at high temperatures? Lamination's would be an easy option to get even distribution without putting a lot of kinks in the flow and gluing them together would make it very simple. @dogie but any response appreciated.

For a gaming machine, its possible as you're hoping to only be at max load for a few hours a day - but in a mining rig with 24hr operation its really pushing it. You'd be better off overspecing and using more expensive solutions than risk warping/leaking and ruining your $8k machine.

Code:

You want as much fluid ramming through the block as possible as opposed to it slowing down and heating up too much. In the radiator you want it slowed down so it gets as close to ambient as possible.

In a closed loop, you care only the pressure loss of the whole system. Your flowrate b4 and after block is the same. The statement above is the most BS i've read. Not surprising tho, i got that alot on forums when their first sentence is " I'm an..... Engineer"

Volume flow rate is constant, flow rate is NOT as it is dependent on cross sectional area. In a radiator you do split up the channel quite a bit in order to slow down the flow rate to maximise contact time.

Wow ,.... you're dumber than i thought. You got reading comprehension problem?

This is what i said, MEng eh?

You said parallel channels in a way which suggested multiple loops - but never mind. Your attitude really stinks and I don't want to interact with you any more.

[Matthew N. Wright]

I've seen that design someplace before. 

[iammagicmike]

....
From CPU/GPU cooling that would be the path (a bad one) etched into the copper. Generally copper is cheap enough (esp for $300+ retail) to make a sandwich of copper plates, although for cost efficiency a single copper plate with acetate cover.

Unequal cooling is countered by high flow rates which tends to keep the coolant temp consistent -> equal cooling. His channel needs a total redesign.

It does but the biggest issue with an Avalon block atm is volume. If there was high enough volume for injection molding and complex CNC operations then a heavily optimized design would be cost effective but for low volume it has to be simple. Even that design has hidden costs, both the fixation and sealing methods for the plate and the mountings for the block aren't included.

The inlet and outlet port sizes are a limiting factor on that layout, smaller ports and more of them would allow for a thinner plate and better heat distribution and maybe allow copper strips to be used for heat transfer instead of a full sized plate.

This is 2013, not the 15th century. Advanced CNC machining is everywhere, just find a workshop near you, give them $20 a unit to to watch over your design being printed and they'll do it. Why exactly are you trying to injection mould copper again?

Inlet and outlet port sizes should be defined by current 1/4 3/8 1/2 standards of all the other fittings on the market, for internal and external diameters, and thread dimensions.

Sealing method is pretty damn simple and as cheap as you like. Rubber gasket around the edge, 8-16 bolts through clearance holes or threaded into the copper through the acetate.

Is overheating a geniune concern for the Avalon?  I have a batch 2 and batch 3 preordered, I was under the impression that they will function out of the box.  Will the cooler increase performance or something?  I just do not understand the purpose for building this. 

If current avalon owners are frantically scrambling to develop a cooling solution then I'd like to be prepared for when mine are shipped.  I work in a shop with a few friendly techs that would gladly CNC something like this for me. 

[SebastianJu]

Like i said.... OP, KISS ( keep it simple stupid)

ALL these engineers who want complicated design forget that it works more efficient when the heat source is concentrated in one spot.

The Avalon board distributed the heat source relative evenly. Run many parallel channels and you will have the best bang for buck.

You're not gonna be able to push those ASICs too high anyway.

I thought about completely oil submerging the avalon but then... how to overclock that thing? Its not that there are tools out there where you can flash the firmware or something. The only thing is to change the clockrate. Standard is 282MHz and you can go up to 300MHz i believe. And it runs fine with that setting with the normal cooling system. Beyond that level the problems rise or it cant go higher by system (Im not sure what it is. The writer spoke about a "wall" at 300.). So i wonder even when you put down the temperature very much... how will you overclock it? Overvolt it? Or do you think you can go higher than 300MHz then? I asking myself if a new firmware or similar is needed to do that.

[loufiethecat]

overvolting is the only way. Mineral oil cooling is interesting got any links on where to get the materials and how to do it properly? I'm thinking of dropping all my kit in oil...

[SebastianJu]

You should search for oil submerging in google to find infos about... there are different oils with different dis-/advantages.

[el_rlee]

are the measurements published somewhere?

[dogie]

overvolting is the only way. Mineral oil cooling is interesting got any links on where to get the materials and how to do it properly? I'm thinking of dropping all my kit in oil...

Oil dipping does NOT work for 24/7 operation. The basic principle is that it has so much heat storage that you can game for 1-4 hours without temps rising significantly, and then slowly release the heat over the other 23-20 hours.

If you're permanently mining, then heat in > heat out = you'll have to turn it off a day a week.

[el_rlee]

You can cool the oil...

Does anybody know the measurements of the heat sink? I will let myself manufacture some water cooling elements, maybe I also offer them for sale.

[ursa]

Hi,
I'm also thinking of a solution to cool down the avalons ( i purchased 10k Chips for personal use with no intent to sale anything) ... Does it actually worth spending the money in water-cooling? Do you think over-clocking them? Right now, i am mining with 600 Spartan6 and i have them over-clocked and i reach ~230 Mh/s ( between 226 and 248 MAX @ 1,34V) and i have them cooled with standard heatsinks and fans...and i don't have any issue with them. Also, overclocking means bigger power consumption , bigger risk and bigger problems, and a lot of time spent on the project,without talking about the amount of money to invest in extreme cooling...
Do you think it will worth?

[Icoin]

Hi,
I'm also thinking of a solution to cool down the avalons ( i purchased 10k Chips for personal use with no intent to sale anything) ... Does it actually worth spending the money in water-cooling? Do you think over-clocking them? Right now, i am mining with 600 Spartan6 and i have them over-clocked and i reach ~230 Mh/s ( between 226 and 248 MAX @ 1,34V) and i have them cooled with standard heatsinks and fans...and i don't have any issue with them. Also, overclocking means bigger power consumption , bigger risk and bigger problems, and a lot of time spent on the project,without talking about the amount of money to invest in extreme cooling...
Do you think it will worth?

Quote from https://en.bitcoin.it/wiki/Avalon :
About water cooling: yes, do it.

As soon the devices for GMP arrive:)