BITMAIN‘s Liquid Cooled Miner C1: It is not only cool, but it is cool

[MrTeal]

problem with running in series is that you would likely see temps on each unit increase in succession, with the last unit(s) having the coolant entering at already-high temperatures and offering insufficient cooling. To avoid this, the flow rate (and thus static pressure from the pump) would probably be too much for the C1 waterblocks.

better would be to use a properly-valved system to run them in parallel. A valve on each C1's loop would adjust for any pressure/temperature difference between units.
 ___________________
 ___   _   _   _   _   __|
      ||  ||   ||  ||  ||

Depends on the restriction of the blocks, but I'm guessing they aren't terribly restrictive. Flow rates wouldn't have to be excessively high to get reasonable performance in a BTC mining application. The specific heat capacity of water is ~4200J/(l*°K), or 4200(W*s)/(l*°K). To find out what flow rate you need to move a certain amount of heat with a specific temperature rise, it's just P/(C*ΔT)
So, if we want the outlet temp of the 4th unit (and inlet of the 5th) to be 30°K above the inlet of the first unit, with the first 4 units dumping 4kW of heat into the water (which is way more than you'd actually see) you'd get the following numbers.

Flow = 4000W / (4200(Ws)/(l°K) * 30°K)
Flow = 0.03175l/s
Flow = 1.9l/min

Not sure how many of these dogie has, but it'd be a cool test to do anyway. You wouldn't even need to use that rad, you could just run three of the 360mm ones he sells in series, and maybe a couple of the pumps to generate sufficient head.

*Edit: That's a minimum flow rate to keep the inlet water temperature of the last unit at a reasonable (say 65C with 35C inlet water) temperature. Depending on the block design you might need a higher rate to actually get 1kW per device from the block into the water.

[dogie]

How would you run 5 independent loops off a single channel radiator?

In series, just like I used to do with 4x GPUs. Can easily dissipate 800W+ of heat on the standard 240 rad I was using back then.

We don't have 800W of heat, we have 5KW of heat - you simply cant get the flow rate required to do that in series before you get to the radiator due to the internal diameter of the tubing [and pathways in the heatsinks].

You also dont need "specialized" coolant for any sort of watercooling, distilled water changed regularly will work just as well, at $.99 per gallon at your local store if that much.

As usual, I recommend the safe options because lets be honest, what percentage of people are going to decide to spend a few hours to drain, flush, clean and refill their multiple C1s?

You will notice I stated a standard 240 rad... that 9x120 rad with push pull could certainly handle 5 of these in series without issue. Infact... I might just do it.
You can also use a power powerful pump as Mr Teal pointed out. Ive been in the WC scene for a LONG time, built over 100+ machines needing to dissipate 1000+w inside a normal ATX case, and those parts were much more heat sensitive in delta T than ASICs generally are.

By my rough napkin calculations, a radiator with reasonable airflow should be able to dissipate a maximum (ideal) of 900w per 120MMx38MM Rad, giving you a theoretical maximum of 8100W dispensation for that radiator. Obviously I am intentionally ignoring things like airflow and water flow rate, but you get the idea.

problem with running in series is that you would likely see temps on each unit increase in succession, with the last unit(s) having the coolant entering at already-high temperatures and offering insufficient cooling. To avoid this, the flow rate (and thus static pressure from the pump) would probably be too much for the C1 waterblocks.

This is what I'm saying. Its not a case of ramming a more powerful pump on it because the heatsink's paths are so restrictive. The minor losses alone would be absolutely huge, nevermind the major losses while trying to pump at high enough velocities to get the last 2 C1s cool. But its also irrelevent, because you're going to need an insane insane insane head pressure to overcome the losses AND get a sensible flow rate.

tldr, just buy 5 of these $50 kits and its k...

[seriouscoin]

LOL at the armchair engineering talks....

Here an idea for you "engineers": use BTU..... and stop using marketing BS by PC cooling vendors.

btw, why waste time on this when you can earn money by just buying BTC at this low price?

[seriouscoin]

How would you run 5 independent loops off a single channel radiator?

In series, just like I used to do with 4x GPUs. Can easily dissipate 800W+ of heat on the standard 240 rad I was using back then.

We don't have 800W of heat, we have 5KW of heat - you simply cant get the flow rate required to do that in series before you get to the radiator due to the internal diameter of the tubing [and pathways in the heatsinks].

You also dont need "specialized" coolant for any sort of watercooling, distilled water changed regularly will work just as well, at $.99 per gallon at your local store if that much.

As usual, I recommend the safe options because lets be honest, what percentage of people are going to decide to spend a few hours to drain, flush, clean and refill their multiple C1s?

You will notice I stated a standard 240 rad... that 9x120 rad with push pull could certainly handle 5 of these in series without issue. Infact... I might just do it.
You can also use a power powerful pump as Mr Teal pointed out. Ive been in the WC scene for a LONG time, built over 100+ machines needing to dissipate 1000+w inside a normal ATX case, and those parts were much more heat sensitive in delta T than ASICs generally are.

By my rough napkin calculations, a radiator with reasonable airflow should be able to dissipate a maximum (ideal) of 900w per 120MMx38MM Rad, giving you a theoretical maximum of 8100W dispensation for that radiator. Obviously I am intentionally ignoring things like airflow and water flow rate, but you get the idea.

problem with running in series is that you would likely see temps on each unit increase in succession, with the last unit(s) having the coolant entering at already-high temperatures and offering insufficient cooling. To avoid this, the flow rate (and thus static pressure from the pump) would probably be too much for the C1 waterblocks.

This is what I'm saying. Its not a case of ramming a more powerful pump on it because the heatsink's paths are so restrictive. The minor losses alone would be absolutely huge, nevermind the major losses while trying to pump at high enough velocities to get the last 2 C1s cool. But its also irrelevent, because you're going to need an insane insane insane head pressure to overcome the losses AND get a sensible flow rate.

tldr, just buy 5 of these $50 kits and its k...

LOL, master of Mechanical Eng eh? ..... What a clueless dumbass.

The fluid pressure in a closed loop is all the same. The system as a whole has pressure losses but the water pressure at the first and last C1 is the same.

I would ask your "imaginary university" for a refund if i were you.

[seriouscoin]

problem with running in series is that you would likely see temps on each unit increase in succession, with the last unit(s) having the coolant entering at already-high temperatures and offering insufficient cooling. To avoid this, the flow rate (and thus static pressure from the pump) would probably be too much for the C1 waterblocks.

better would be to use a properly-valved system to run them in parallel. A valve on each C1's loop would adjust for any pressure/temperature difference between units.
 ___________________
 ___   _   _   _   _   __|
      ||  ||   ||  ||  ||

Depends on the restriction of the blocks, but I'm guessing they aren't terribly restrictive. Flow rates wouldn't have to be excessively high to get reasonable performance in a BTC mining application. The specific heat capacity of water is ~4200J/(l*°K), or 4200(W*s)/(l*°K). To find out what flow rate you need to move a certain amount of heat with a specific temperature rise, it's just P/(C*ΔT)
So, if we want the outlet temp of the 4th unit (and inlet of the 5th) to be 30°K above the inlet of the first unit, with the first 4 units dumping 4kW of heat into the water (which is way more than you'd actually see) you'd get the following numbers.

Flow = 4000W / (4200(Ws)/(l°K) * 30°K)
Flow = 0.03175l/s
Flow = 1.9l/min

Not sure how many of these dogie has, but it'd be a cool test to do anyway. You wouldn't even need to use that rad, you could just run three of the 360mm ones he sells in series, and maybe a couple of the pumps to generate sufficient head.

*Edit: That's a minimum flow rate to keep the inlet water temperature of the last unit at a reasonable (say 65C with 35C inlet water) temperature. Depending on the block design you might need a higher rate to actually get 1kW per device from the block into the water.

When you have multiple restrictive blocks in parallel, you will not have bottle neck in the system. I dont know why ppl even consider series in this case.

Ppl really overthink this. You can have multiple loops with common coolant and heat exchange (ie. 1 pump for each C1, one giant rad - NOT in series....)

[dogie]

Not sure how many of these dogie has, but it'd be a cool test to do anyway. You wouldn't even need to use that rad, you could just run three of the 360mm ones he sells in series, and maybe a couple of the pumps to generate sufficient head.

*Edit: That's a minimum flow rate to keep the inlet water temperature of the last unit at a reasonable (say 65C with 35C inlet water) temperature. Depending on the block design you might need a higher rate to actually get 1kW per device from the block into the water.

Lets try some maths:

flowrate = Q / ((heat capacity)* density *(Tout - Tin))

Q = 5KW
Heat capacity of water = 4.18 kJ/kg/K
Density = ~1000kg/m^3
Tout max = 65C to keep chips < 75C [pretty optimistic heat exchanger]
Tin = 30C [depends on ambient, I wouldn't be getting mine below 40C]
flowrate = ? m^3/s

flowrate = 5 / (4.18 * 1000 * 35)
flowrate = 0.0000342 m/3^s = 0.0341 L/s = 122 L/h

With a 6mm internal diameter [forced to this size by the heatsink channels]...

A= 0.0000283 m^2
flowrate = A*v
v = velocity m/s
0.0000342 = 0.0000283 * v
Therefore v would be 1.21 m/s.

Assumes perfect heat exchanger, which its not so we lose effective deltaT. Do remember though the size of the pump you're going to need to achieve 120 L/h with that level of restriction.

Assume 40% effectiveness due to single flow.
New v = 1.21/0.4 = 3.0 m/s

Minor losses:
v of 3.0 m/s
K of 40 [conservative, if you think of the number of times we have 180 degree turns in the  5 blocks + the radiator]
head loss (m) = 18.4m

Major losses:
Assume 20m of internal flow path = L
D = 0.006m
V = 3 m/s
Major losses = 30.6m

Total losses = 49.0m.

So if you can generate 50m of pump head, you can move the fluid.

[MrTeal]

When you have multiple restrictive blocks in parallel, you will not have bottle neck in the system. I dont know why ppl even consider series in this case.

Ppl really overthink this. You can have multiple loops with common coolant and heat exchange (ie. 1 pump for each C1, one giant rad - NOT in series....)

I never said the full series solution is the best, just that people were overstating the amount of flow you need to keep the coolant at acceptable temperatures. 2l/min isn't much.

Having a large series string with a couple pumps does have other benefits as well in terms of reliability if there's a pump failure, and possibly easier routing of tubing.

[seriouscoin]

When you have multiple restrictive blocks in parallel, you will not have bottle neck in the system. I dont know why ppl even consider series in this case.

Ppl really overthink this. You can have multiple loops with common coolant and heat exchange (ie. 1 pump for each C1, one giant rad - NOT in series....)

I never said the full series solution is the best, just that people were overstating the amount of flow you need to keep the coolant at acceptable temperatures. 2l/min isn't much.

Having a large series string with a couple pumps does have other benefits as well in terms of reliability if there's a pump failure, and possibly easier routing of tubing.

The reliability benefit does not outweight the performance (temp consistency of the miners). If you're looking for reliability use flowrate + temp sensors for each loop.

[MrTeal]

Depends on the restriction of the blocks, but I'm guessing they aren't terribly restrictive. Flow rates wouldn't have to be excessively high to get reasonable performance in a BTC mining application. The specific heat capacity of water is ~4200J/(l*°K), or 4200(W*s)/(l*°K). To find out what flow rate you need to move a certain amount of heat with a specific temperature rise, it's just P/(C*ΔT)
So, if we want the outlet temp of the 4th unit (and inlet of the 5th) to be 30°K above the inlet of the first unit, with the first 4 units dumping 4kW of heat into the water (which is way more than you'd actually see) you'd get the following numbers.

Flow = 4000W / (4200(Ws)/(l°K) * 30°K)
Flow = 0.03175l/s
Flow = 1.9l/min

Not sure how many of these dogie has, but it'd be a cool test to do anyway. You wouldn't even need to use that rad, you could just run three of the 360mm ones he sells in series, and maybe a couple of the pumps to generate sufficient head.

*Edit: That's a minimum flow rate to keep the inlet water temperature of the last unit at a reasonable (say 65C with 35C inlet water) temperature. Depending on the block design you might need a higher rate to actually get 1kW per device from the block into the water.

Lets try some maths:

flowrate = Q / ((heat capacity)* density *(Tout - Tin))

Q = 5KW
Heat capacity of water = 4.18 kJ/kg/K
Density = ~1000kg/m^3
Tout max = 65C to keep chips < 75C [pretty optimistic heat exchanger]
Tin = 30C [depends on ambient, I wouldn't be getting mine below 40C]
flowrate = ? m^3/s

flowrate = 5 / (4.18 * 1000 * 35)
flowrate = 0.0000342 m/3^s = 0.0341 L/s

With a 6mm internal diameter [forced to this size by the heatsink channels]...

A= 0.0047 m^2
flowrate = A*v
v = velocity m/s
0.0000342 = 0.0047 * v
Therefore v would be 0.0084 m^s.

Check your math, you're comparing W and kJ directly.

Edit: I see you realized your mistake and corrected it.

[dogie]

Check your math, you're comparing W and kJ directly.

Edit: I see you realized your mistake and corrected it.

Yeah missed the preview button while I was writing it out, finished now lol.

[MrTeal]

Check your math, you're comparing W and kJ directly.

Edit: I see you realized your mistake and corrected it.

Yeah missed the preview button while I was writing it out, finished now lol.

I think your losses are pessimistic. Even using about double the flow rate we were talking about to keep inlet temperature down (1GPM) you shouldn't need 50mH2O of head. A common 360mm radiator like the Swiftech MCR360-QP has a pressure drop of ~0.2mH2O at 1GPM. Even three of those in series would only be 0.6m of pressure drop.
I don't know how those blocks are internally constructed, but most common CPU waterblocks are 0.5-2mH2O @ 1GPM. If those two blocks in the C1 are close to the 10mH2O at that speed you would need in order to get a 50m total drop, I would be extremely surprised and they would be a terrible design.

[MasterRadix]

without the shipping cost, I would take one....

[dogie]

I think your losses are pessimistic. Even using about double the flow rate we were talking about to keep inlet temperature down (1GPM) you shouldn't need 50mH2O of head. A common 360mm radiator like the Swiftech MCR360-QP has a pressure drop of ~0.2mH2O at 1GPM. Even three of those in series would only be 0.6m of pressure drop.
I don't know how those blocks are internally constructed, but most common CPU waterblocks are 0.5-2mH2O @ 1GPM. If those two blocks in the C1 are close to the 10mH2O at that speed you would need in order to get a 50m total drop, I would be extremely surprised and they would be a terrible design.

Do remember that the radiators only have essentially 3 180 degree turns and large flow channels in order to slow the fluid down. This works because of the extreme surface area. The waterblocks on the other hand (of which there are 10), has a small diameter channel and the flow snakes forwards and back 4x [4x 180 degree bends]. Each barb, fitting and slight bend then adds additional minor losses. The figures you're quoting are also for 9.5mm ID tubing, where as we're using 6mm.

Even if you pretend for a second there are zero minor losses through the entire system, just look at the major losses. There is about 1.2 metres of path in each block, so 12m of path just from them. The radiator has another metre internally, and you've got all the small bits of tubing and then the larger lengths of tubing. Even taking it down to 15m, that's a major head loss of 22.5m.

And yes, this style of waterblock is MUCH more restrictive than a radiator as you have to snake the channel because you have no internal fins for surface area.

[jelin1984]

How is it the shipping cost???

[tbolt]

Sigh, my reserved post was deleted, need to wait for Theymos to come on now to restore it. For now here are the links:

What you need to get started:

____________________________	_________________	______________________________
1KW PSU + ethernet cable	[amazon.com]	| /7LuMfC]$152.19 + $0
Special C1 watercooling kit	[syscooling.com]	| /bBpr2P]$50 + ~$50
1L* of coolant	[aquatuning.us]	| /kEuPTd]$7.29 + $8
2x Quick disconnect fittings*^	[aquatuning.us]	| /K6Ugng]$26.30 + $8%

   


*Get 2L if you think you might spill or want spare.
*^Optional but highly recommended.
^%No additional shipping charge if purchased with the coolant.

Dang.

Aquatuning requires a $75 USD minimum order.

anyone know of a different source for the fluid and quick disconnects
that they would recommend?

[MrTeal]

I think your losses are pessimistic. Even using about double the flow rate we were talking about to keep inlet temperature down (1GPM) you shouldn't need 50mH2O of head. A common 360mm radiator like the Swiftech MCR360-QP has a pressure drop of ~0.2mH2O at 1GPM. Even three of those in series would only be 0.6m of pressure drop.
I don't know how those blocks are internally constructed, but most common CPU waterblocks are 0.5-2mH2O @ 1GPM. If those two blocks in the C1 are close to the 10mH2O at that speed you would need in order to get a 50m total drop, I would be extremely surprised and they would be a terrible design.

Do remember that the radiators only have essentially 3 180 degree turns and large flow channels in order to slow the fluid down. This works because of the extreme surface area. The waterblocks on the other hand (of which there are 10), has a small diameter channel and the flow snakes forwards and back 4x [4x 180 degree bends]. Each barb, fitting and slight bend then adds additional minor losses. The figures you're quoting are also for 9.5mm ID tubing, where as we're using 6mm.

Even if you pretend for a second there are zero minor losses through the entire system, just look at the major losses. There is about 1.2 metres of path in each block, so 12m of path just from them. The radiator has another metre internally, and you've got all the small bits of tubing and then the larger lengths of tubing. Even taking it down to 15m, that's a major head loss of 22.5m.

And yes, this style of waterblock is MUCH more restrictive than a radiator as you have to snake the channel because you have no internal fins for surface area.

I never said it wasn't more restrictive than a radiator, just that it would be pretty terrible if ten of them in series had 50mH2O (71PSI) pressure drop at 1GPM.

Since Bitmain is selling these without matching hardware, it would actually be really handy if they could make flowrate vs pressure drop chart for them, similar to the one in this datasheet.
http://www.wakefield-vette.com/Portals/0/resources/datasheets/180-12,180-20.pdf

You don't happen to have a flow meter and manometer, do you?

[dogie]

Since Bitmain is selling these without matching hardware, it would actually be really handy if they could make flowrate vs pressure drop chart for them, similar to the one in this datasheet.
http://www.wakefield-vette.com/Portals/0/resources/datasheets/180-12,180-20.pdf

You don't happen to have a flow meter and manometer, do you?

And variable speed pump, no

[Rabinovitch]

Well, if C1 price is 1 BTC then I would like to buy S4 for 2-2.2 BTC. Bitmain?.. My coupon will burn out soon and I can't use it to buy an expensive S4...

[richardamullens]

As a home miner, I don't want to pay duty on the controller when I can buy a Raspberry Pi and use cgminer.
Also, I don't want to scrap the controller (as well) when it is no longer financially viable to operate the kit.

Please make a device like this (or air cooled) which can be powered by 1 ATX PSU and which has a USB connection - so I have a logical upgrade path from Antminer U2.

Perhaps such a device could also make use of the airflow from the ATX PSU.

Better still if you could sell a unit that came in less than the threshold for duty payments.

Otherwise, it's great to see you innovating.

[MrTeal]

Since Bitmain is selling these without matching hardware, it would actually be really handy if they could make flowrate vs pressure drop chart for them, similar to the one in this datasheet.
http://www.wakefield-vette.com/Portals/0/resources/datasheets/180-12,180-20.pdf

You don't happen to have a flow meter and manometer, do you?

And variable speed pump, no

Heh, send me one and I'll test it for you.