Avalon Water Cooling

[crumbs]

water always wins air hands down in terms of cooling efficiency and capacity. That's why engines are water and not air cooled (except for some low output exceptions, lawn mowers, mopeds, old VWs etc.)
[...]

[1715075127]

1715075127

[1715075127]

1715075127

[1715075127]

1715075127

[1715075127]

1715075127

[1715075127]

1715075127

[PulsedMedia]

water always wins air hands down in terms of cooling efficiency and capacity. That's why engines are water and not air cooled (except for some low output exceptions, lawn mowers, mopeds, old VWs etc.)
[...]

Yes, and that engine is one of those rather low output ones. Yes it is, huge displacement but very low specific output, especially at altitude at thin air.
and as you can see from the image it's still huge effort to cool it, look at all those fins!

I want to see you try cooling an small displacement inline-4 engine which outputs 2megawatts as waste heat with just air. Yes, megawatts. and no, that engine does not melt. Infact, runs very smooth and without serious heat issues.

[DickyWong]

 
http://farm7.static.flickr.com/6237/6273300406_45db451ca2_z.jpg

[crumbs]

water always wins air hands down in terms of cooling efficiency and capacity. That's why engines are water and not air cooled (except for some low output exceptions, lawn mowers, mopeds, old VWs etc.)
[...]

Yes, and that engine is one of those rather low output ones. Yes it is, huge displacement but very low specific output, especially at altitude at thin air.
and as you can see from the image it's still huge effort to cool it, look at all those fins!

I want to see you try cooling an small displacement inline-4 engine which outputs 2megawatts as waste heat with just air. Yes, megawatts. and no, that engine does not melt. Infact, runs very smooth and without serious heat issues.

Jeesh, joking on these boards is dangerous  And yes, i know .58 beats 0.024, Fact Cat agreez.  

[crumbs]

If we's gonna cheat

[Gomeler]

If we's gonna cheat

Space shuttle was technically liquid cooled. If I recall correctly the nozzles were cooled with cryogenic hydrogen that was expanded in tubes surrounding the combustion chambers and nozzles. This expanded hydrogen was then used to drive the oxygen and hydrogen turbopumps that fed the main engine. The wikipedia article has some neat details of how the system worked -> http://en.wikipedia.org/wiki/Space_Shuttle_Main_Engine. For reference, each hydrogen turbo pump was around 70,000 horsepower per engine.

[crumbs]

If we's gonna cheat

Space shuttle was technically liquid cooled. If I recall correctly the nozzles were cooled with cryogenic hydrogen that was expanded in tubes surrounding the combustion chambers and nozzles. This expanded hydrogen was then used to drive the oxygen and hydrogen turbopumps that fed the main engine. The wikipedia article has some neat details of how the system worked -> http://en.wikipedia.org/wiki/Space_Shuttle_Main_Engine. For reference, each hydrogen turbo pump was around 70,000 horsepower per engine.

You're right, now that you mention it   My fail.

[goxed]

No wonder, Avalon sounds like a Jet taking off at full Fan RPMs

[SolarSilver]

No wonder, Avalon sounds like a Jet taking off at full Fan RPMs

If you are referring to a box of fans, you are confusing with another ASIC vendor ;-)

[Felipeo]

I see before, very nice water cooling for Avalon rigs on bitmine.ch, but now I can't find it 

[oxoa]

Nice mod, Im eager to see it mounted... 

[sneef]

I see before, very nice water cooling for Avalon rigs on bitmine.ch, but now I can't find it 

Did you mean this one.
https://devda.ch/content/avalon-qcool-element

It is on a .ch
This is why i wondered if it is the one you mean?

[firefop]

You might have considered this already, but unless you have an incredibly high rate of flow through that block, the inflow side is going to be substantially cooler than the outflow (as water snakes through your block, doing its thing & picking up heat).  Might want to reconsider using the standard "sandwich" approach -- milling out the metal would take minutes, you already have access to an end mill, and if you're worried about leakage, go with 1/8" alum. plate instead of plastic, i've never had problems with leaking o-rings.  What's good for a radiator is pretty lousy for a waterblock

Yes, it's a very basic design. Ideally you want at least triple the surface area inside a waterblock as the area of your hot side.

A better design is to break each input tube into a dozen or two dozen much smaller tubes inside... you have the added benefit of much faster rate of flow through the block and also get more heat conductive surfaces... this translates into better heat exchange from the block to the liquid.

[crumbs]

You might have considered this already, but unless you have an incredibly high rate of flow through that block, the inflow side is going to be substantially cooler than the outflow (as water snakes through your block, doing its thing & picking up heat).  Might want to reconsider using the standard "sandwich" approach -- milling out the metal would take minutes, you already have access to an end mill, and if you're worried about leakage, go with 1/8" alum. plate instead of plastic, i've never had problems with leaking o-rings.  What's good for a radiator is pretty lousy for a waterblock

Yes, it's a very basic design. Ideally you want at least triple the surface area inside a waterblock as the area of your hot side.

A better design is to break each input tube into a dozen or two dozen much smaller tubes inside... you have the added benefit of much faster rate of flow through the block and also get more heat conductive surfaces... this translates into better heat exchange from the block to the liquid.

I never had to cool setups like this, where the heat is dissipated into the circuit board's ground plane (?) & the package dictates board-side cooling (well, VR chips, i guess).  This is the best pic of the chip's pron side i could find:
Not sure if there's some sort of plate-through from the solder pad at the center to the ground plane, or if the only copper the chip sees is that center pad itself.  If it's the latter. i hope the pc board's skinny, not the best thermal conductor.
Have you considered something like this?
I was clicking through pics on Google & this is exactly like something i've made using nothing but Home Depot plumbing hardware & solder.  The tubing is simply laid into the routed channels & pressed flat (so it forms a D, flat side facing out ) just like in the direct contact heatpipe heat sinks. Dressed with a file for that extra-fancy ground finish.  The manifolds are just plumbing pipe, drilled & endcapped.  Fill all the gaps with cheap heatsink grease & it's ready to serve.  Works killa' & never leaks.   

[firefop]

You might have considered this already, but unless you have an incredibly high rate of flow through that block, the inflow side is going to be substantially cooler than the outflow (as water snakes through your block, doing its thing & picking up heat).  Might want to reconsider using the standard "sandwich" approach -- milling out the metal would take minutes, you already have access to an end mill, and if you're worried about leakage, go with 1/8" alum. plate instead of plastic, i've never had problems with leaking o-rings.  What's good for a radiator is pretty lousy for a waterblock

Yes, it's a very basic design. Ideally you want at least triple the surface area inside a waterblock as the area of your hot side.

A better design is to break each input tube into a dozen or two dozen much smaller tubes inside... you have the added benefit of much faster rate of flow through the block and also get more heat conductive surfaces... this translates into better heat exchange from the block to the liquid.

I never had to cool setups like this, where the heat is dissipated into the circuit board's ground plane (?) & the package dictates board-side cooling (well, VR chips, i guess).  This is the best pic of the chip's pron side i could find:
Not sure if there's some sort of plate-through from the solder pad at the center to the ground plane, or if the only copper the chip sees is that center pad itself.  If it's the latter. i hope the pc board's skinny, not the best thermal conductor.
Have you considered something like this?
I was clicking through pics on Google & this is exactly like something i've made using nothing but Home Depot plumbing hardware & solder.  The tubing is simply laid into the routed channels & pressed flat (so it forms a D, flat side facing out ) just like in the direct contact heatpipe heat sinks. Dressed with a file for that extra-fancy ground finish.  The manifolds are just plumbing pipe, drilled & endcapped.  Fill all the gaps with cheap heatsink grease & it's ready to serve.  Works killa' & never leaks.   

This would actually be much better imo... the more smaller tubes the better (to keep that flow rate high). The other interesting thing about this... if you were to vary the length of the micro pipes by making in non-rectangular with a delivery tube at an angle - you could potentially shape the heat profile of the sink in a beneficial way... if that doesn't really apply to the design...

The other method is use something like your pictures but twice as high... with only having active flow through half of it (the part in contact with the board) Effectively giving you the ability to store more heat away from the chips...

Or you could even combine these ideas using shorter micro tubes on the chip attached section and much longer ones on the parts high out... (which actually might give you slightly better cooling at the cost of needing a bigger pump.

[el_rlee]

Finally found some time...


Removing the PCB from the original heatsink - here you can clearly see that the only areas where you have real contact are around the screws...

Precision machining, fits like a charm

Fuck - forgot the plug, having a collision




Looks not really beautiful

Decided to not use the water channel behind the voltage converters and put in the thermosensor instead

My balcony

[el_rlee]

This is the water cooled miner running at 325MHz now - can somebody point me on where to see if a still have potential to go higher?
What are the "rejected", what the "discarded" shares? Where do I see the hardware error rate?



That's the air cooled miner next to it:



Here are the pool stats - I don't really understand why the water cooled miner is not faster by the number the frquency is higher (~15%)

[invader]

You should check accepted diff1 shares against hw errors, thats what you really want to know.
According to your screen you have ( 9032 / 735510 ) * 100% = 1.22% error rate
Try the latest firmware that was posted in "Avalon ASIC users thread" and overclock it to 375+

It's such a pain to see this "design flaw" that is not allowed you to connect 4th channel that is right behind stepdown converters.
If i was you, i would put connector on the other side of PCB and rearrange pins in male connector accordingly, or "enlong" one row of pins to make it able fit perpendicular. When i look at your photos it makes me want to reapply thermal paste to my unit.. Also thanks for sharing your experience with water cooling.

[el_rlee]

You should check accepted diff1 shares against hw errors, thats what you really want to know.
According to your screen you have ( 9032 / 735510 ) * 100% = 1.22% error rate
Try the latest firmware that was posted in "Avalon ASIC users thread" and overclock it to 375+

It's such a pain to see this "design flaw" that is not allowed you to connect 4th channel that is right behind stepdown converters.
If i was you, i would put connector on the other side of PCB and rearrange pins in male connector accordingly, or "enlong" one row of pins to make it able fit perpendicular. When i look at your photos it makes me want to reapply thermal paste to my unit.. Also thanks for sharing your experience with water cooling.

Thanks for your reply.

The 4th channel which is not in use does not at all result in any temperature difference I could feel by hand... same applies for the difference between the first and the last module.

It seams the problem is not at all to get the heat of the aluminum block, but the high thermal resistance between the chip and the heat sink. I guess two channels would be totally sufficient.

The chiller allows a set value down to 2 degree Celsius, I just shit my pants a little in fear of condensation. I live in a really hot and humid place.

What's a "healthy" hardware error rate?

[SolarSilver]

This is the water cooled miner running at 325MHz now - can somebody point me on where to see if a still have potential to go higher?
What are the "rejected", what the "discarded" shares? Where do I see the hardware error rate?

You can put it onto --avalon-auto and it will sort out the freq automagicly by looking at the % of hardware errors. Air cooled systems run at 365 MHz here, water cooled should go up to 400+ (450 MHz max)

As for discarded:

Discarded is totally meaningless, I wish it wasn't in the display at all. It's a number from when getwork was the protocol in use and means nothing in the local work generation era.

Hardware error rate:

https://bitcointalk.org/index.php?topic=140539.msg2641395#msg2641395