DataTank Mining: 1.2MW 3M Novec Immersion Cooled 2PH Mining Container

[Collider]

@antirack; @DTM:

Please remember to always post significant updates on your assets at havelock (in the updates section )
Not everyone is reading bitcointalk threads.

[1714260854]

1714260854

[Holographic]

Been skimming through this thread and trying to find some updates regarding the IPO in between the fighting and bickering. No luck.

Can anyone help me out and just give a quick status update? I thought the IPO was supposed to start June 24? Thanks in advance for any useful info

[KarmaShark]

Been skimming through this thread and trying to find some updates regarding the IPO in between the fighting and bickering. No luck.

Can anyone help me out and just give a quick status update? I thought the IPO was supposed to start June 24? Thanks in advance for any useful info

I believe the IPO has been pushed a couple weeks:




"How to buy Shares/Units:
Units will be sold in 4 rounds on HavelockInvestments.com, discounts for early investors will be available (public offer date preliminary set to July 7).

Angel investors can also purchase units directly from DataTank Mining, please contact us via email. A minimum purchase amount of 100kW (10,000 units) applies for direct shares."

[DataTankMining]

The last progress update was here:
https://bitcointalk.org/index.php?topic=655464.msg7469776#msg7469776

The public sale of capacity units on HavelockInvestments.com has been pushed back (preliminary date Monday July 7, after the U.S. Marshall's  Service Bitcoin auction and the July 4 weekend).

It is now also copied to the first post and will be done so in the future for consistency.

[electerium]

So basically you are saying immersion cooling is too expensive but you have no idea what it costs to build a 1MW farm?

data tank's strongest feature is it's density. But what use is density if chips need to be replaced every 6 months? That kind of turnover is ridiculous and expensive. And when the prospective says that they need 3-6mo just for deployment, how exactly does 2PH scale relative to the likely 400+ PH of the entire network?

I have no doubt that these guys who engineered and created data tank are smart businessmen. I am positive that they will make money, but I am equally positive that investors won't unless something crazy happens like sha256 ASICs randomly stalling moore's law, and there is some abrupt plateau in chip performance @28-20nm process.

This technology is far too ahead of it's time. When network PH stabilizes or plateaus, this kind of technology will be insanely valuable. But I dont think (according to puppets measurements) we will likely see that anytime before the network hits 2,000 PH.

[Ozziecoin]

data tank's strongest feature is it's density. But what use is density if chips need to be replaced every 6 months? That kind of turnover is ridiculous and expensive. And when the prospective says that they need 3-6mo just for deployment, how exactly does 2PH scale relative to the likely 400+ PH of the entire network?

I have no doubt that these guys who engineered and created data tank are smart businessmen. I am positive that they will make money, but I am equally positive that investors won't unless something crazy happens like sha256 ASICs randomly stalling moore's law, and there is some abrupt plateau in chip performance @28-20nm process.

This technology is far too ahead of it's time. When network PH stabilizes or plateaus, this kind of technology will be insanely valuable. But I dont think (according to puppets measurements) we will likely see that anytime before the network hits 2,000 PH.

Then buy DTMB.

[antirack]

data tank's strongest feature is it's density.

DataTank's strongest features are that (a) it costs less, (b) deploys faster, (c) can be reused, (d) reduces hardware to chips on boards, (e) allows a lot of room to grow (current condenser designs are close to 500kW per tank at 35C, they will do even better in cold Washington state).

Density is important, but only plays a secondary role. It also goes beyond the device level that you see on the surface (ie. smaller space required). At many places space is not a problem, but a 10 ft electrical cable costs 10 times less than a 100 ft cable to keep it simple and make an example = cost savings again.

More important for the job on hand is this in my opinion (_in addition_ to the cost savings on the infrastructure/cooling):

One single DataTank container is around 576 SP30 from Spondoolies Tech. I have done this example a few days ago already in a previous post.

576 2U enclosures cost a lot of money
576 power supplies cost money
17,280 (576x30) heatsinks cost money
4608 fans cost money
Etc

All that needs to be installed, quality controlled (several times), packaged and shipped from China to Israel to Washington State.

It needs to be packaged, shipped and quality controlled multiple times in fact (at the manufacturer, at Spondoolies when it comes in, at Spondoolies when it goes out).

So how many days does it take the Israeli factory to build, assemble and prep 576 (15 palettes? maybe more) of hardware? Quite a while. Building it in China and shipping to WA still no big difference, somebody has to assembly all that metal.

How much faster and cheaper can FedEx (we are in a rush in Bitcoin mining) deliver 1 palette with boards compared to 15 palettes? I wouldn't be surprised if the shipping cost alone could buy a new car (I will look into this). You also ship twice, once from China, then from Israel. Shipping cost is probably a double digit percentage on these systems, I wouldn't be surprised if it's more than a quarter!

And how long does it take on the other side at the mine to unpack all that, get it ready for installation, and distribute it in 144 racks? (assuming you can install 4 in a rack, that's 10kW/rack)

144 racks from China, that need to come in in a few shipping containers, also cost $40k to $70k, depending on what quality you want (I assume he cheapest). How long to _assemble_ 144 racks? 1 person can probably assemble a few of them in a day. You'll need call a lot of friends if you want them up in a week.
 
If you run a complete TCO comparison (total cost of ownership), you will find out that you saved yourself a lot of money already at the beginning, and that you pay much less operating costs, and that you are online a lot earlier than otherwise.

In the case of DataTank, the deployment of these SP30/Rockerbox boards takes literally a single person a half day. And the moment the chips are soldered to the boards they can leave the factory towards the mine. And because of the efficient cooling, they will actually consume less electricity (reduced leak currents). That's not even related to the missing fans. As a result, you can either save on OpEx, or crank them more (higher hashrate). As Bitcoin miner you probably want the latter, especially with cheap electricity cost and no cooling overhead.

Of course some critics may say you could also reinvent the wheel, design the ASICS in a totally different way, higher some scientists and show the world (Bitcoiners and everyone else) how to do it. But that is what we have here and today, and I truly believe it's the future. We have been doing this for almost 3 years. We are doing it because it makes total sense. And it's not an idea, it exists.

-- // --

Of course the 3 months _initial_ build time is long, but the DataTank plan is long term. After that it only takes a few hours and everything is re-used. Again and again. And don't forget that after the initial phase, the production pipeline is setup and we can produce 1 or a hand full or even more per week.

In the western world, from the idea of deploying 1MW to actually doing it there is a difference of a few months and a lot of work. Case in point: lead time of a transformer is 8-12 weeks minimum, installation time will take another 1-2 weeks, commissioning by the power company another week. From start to finish a lot of money changes hands.

That's why cloud mining is so attractive nowadays to the small miner (I am not talking about the ripoff sites). You save all the trouble. But even the cloud mining companies still need to run their hardware somewhere, and these are the people hat actual deploy real hardware and make the experience I am point out above.

[Ozziecoin]

Of course the 3 months _initial_ build time is long, but the DataTank plan is long term. After that it only takes a few hours and everything is re-used. Again and again. And don't forget that after the initial phase, the production pipeline is setup and we can produce 1 or a hand full or even more per week.

Re: build time. My understanding is that DTMA will purchase chips at the time it is ready to deploy? Is this correct?

[NotLambchop]

...a 10 ft electrical cable costs 10 times less than a 100 ft cable to keep it simple and make an example = cost savings again.

Ridiculous.  No one who ever bought Ethernet patch cables, or guitar patch cords, or electrical extension cords would ever say that. With typical lengths (oh, Ethernet cables, 10ft/100ft), the price difference is not x10 but ~2.5.

More important is this:

One single DataTank container is around 576 SP30 from Spondoolies Tech. I have done this example a few days ago already in a previous post.

576 2U enclosures cost a lot of money

...and are not necessary for a permanent mega-scale installation.  The (open) chassis housing multiple boards could well be integrated into custom racks.

576 power supplies cost money

Sure, but if you're suggesting that it's cheaper/more practical to have a few gigawatt monsters, those can exist for air/water cooled setups too.

17,280 (576x30) heatsinks cost money
4608 fans cost money
Etc

Err...  Isn't that Novec stuff ~$350/gal?  How many gallons in a 40ft container?  And all that heat exchanger plumbing and cooling towers, etc, etc.?
Come on

All that needs to be installed, quality controlled (several times), packaged and shipped from China to Israel to Washington State.

Your containers don't need to be shipped, installed or quality controlled?

It needs to be packaged, shipped and quality controlled multiple times in fact (at the manufacturer, at Spondoolies when it comes in, at Spondoolies when it goes out).

Again, not when you have a bulk order of 576 units to be installed at one location.  Each unit isn't packaged individually, in a colorful box with glossy instruction manuals and a few CDs of pr0nz gaems drivers whatever it is.  Bulk shipping is a thing, it's not rocket surgery.  Further, what you'll be shipping will be 100% standard, easily serviceable by semi-trained personnel.  Who would not have to put on scuba gear, power down the entire farm, and dive into boiling Novec to pull out one bad board

So how many days does it take the Israeli factory to build, assembly and prep 576 (15 palettes? maybe more) of hardware? quite a while.

No frickin idea, depends on the factory.  You act like Israel is some third-world country.

How much faster and cheaper can FedEx (we are in a rush in Bitcoin mining) deliver 1 palette with boards compared to 15 palettes? I wouldn't be surprised if the shipping cost alone couldn't buy a new car (I will look into this).

Faster?  Usually not much.  FedEx doesn't hand-carry each box to its destination.  They have big trucks and big planes.  How much faster is it to deliver one letter instead of ten?

I mean sure, there are advantages, but what you get in the end is highly specialized gear that can't be resold without mods, two separate production lines (one for consumer gear and one for stripper boards for submerged use), and a slew of unforeseen complications that come with new technology.  Don't over-hype.

[jimmothy]

...a 10 ft electrical cable costs 10 times less than a 100 ft cable to keep it simple and make an example = cost savings again.

Ridiculous.  No one who ever bought Ethernet patch cables, or guitar patch cords, or electrical extension cords would ever say that. With typical lengths (10ft/100ft), the price difference is not x10 but ~2.5.

The cost difference between 1000ft of cable vs 10000 ft of cable is ~10 times. They are not buying individual 10 ft cables from best buy.

...and are not necessary for a permanent mega-scale installation.  The (open) chassis housing multiple boards could well be integrated into custom racks.

Custom racks. Sounds cheap and reusable.

17,280 (576x30) heatsinks cost money
4608 fans cost money
Etc

Err...  Isn't that Novec stuff ~$350/gal?  How many gallons in a 40ft container?  And all that heat exchanger plumbing and cooling towers, etc, etc.?
Come on

I'd also like to know how much fluid per KW they are using.

I mean sure, there are some advantages, but what you get in the end is highly specialized gear that can't be resold without mods, two separate production lines (one for consumer gear and one for stripper boards for submerged use), and a slew of unforeseen complications that come with new technology.  Don't over-hype.

There is no need to resell hardware because with 2x efficiency and $0.02/kwh they will be the last ones to dump their old hardware.

Who said anything about consumer gear?

And what do you mean "a slew of unforeseen complications"? You know they have been running an immersion DC for months right?

[NotLambchop]

...a 10 ft electrical cable costs 10 times less than a 100 ft cable to keep it simple and make an example = cost savings again.

Ridiculous.  No one who ever bought Ethernet patch cables, or guitar patch cords, or electrical extension cords would ever say that. With typical lengths (10ft/100ft), the price difference is not x10 but ~2.5.

The cost difference between 1000ft of cable vs 10000 ft of cable is ~10 times. They are not buying individual 10 ft cables from best buy.

If he's talking about 10ft vs. 100ft, I can safely assume he is talking about cable with connectors on its ends, not bulk length.  At which point, the connectors and labor become a significant part of the cost.  The x2.5 number I gave is for shit cat5 10packs.

...and are not necessary for a permanent mega-scale installation.  The (open) chassis housing multiple boards could well be integrated into custom racks.

Custom racks. Sounds cheap and reusable.

A hell of a lot cheaper than 40ft container full of Novec.  At $350/gal, my guess is approximately 50 times more expensive than the table wine you drink

17,280 (576x30) heatsinks cost money
4608 fans cost money
Etc

Err...  Isn't that Novec stuff ~$350/gal?  How many gallons in a 40ft container?  And all that heat exchanger plumbing and cooling towers, etc, etc.?
Come on

I'd also like to know how much fluid per KW they are using.

I mean sure, there are some advantages, but what you get in the end is highly specialized gear that can't be resold without mods, two separate production lines (one for consumer gear and one for stripper boards for submerged use), and a slew of unforeseen complications that come with new technology.  Don't over-hype.

There is no need to resell hardware because with 2x efficiency and $0.02/kwh they will be the last ones to dump their old hardware.

Who said anything about consumer gear?

And what do you mean "a slew of unforeseen complications"? You know they have been running an immersion DC for months right?

Err... You know they've been running conventional data centers for decades, right?
And yeah, I'd like to be able to sell hardware if i wish to liquidate, or have a chance to unload it at a profit.

I know that sounds crass, but I just wanna use hardware, I don't feel like we need to get married.
But you...  I see you already got a stiffy for this handsome 40-footer full of unicorn peckersnot.  You marry it, just send me pix!

[antirack]

Err...  Isn't that Novec stuff ~$350/gal?  How many gallons in a 40ft container?  And all that heat exchanger plumbing and cooling towers, etc, etc.?

This is all included in the price.

Novec is not cheap, but it's not even close to $350/gal.

And again, we can use as little as 200cc per 4000W.

The value proposition of our solution is therefore that it is less expensive than what you would need to spend for heatsink and fan once 1MW of mining gear is deployed.

That's the whole point.

There are some very professional people that have built a facility in Prineville, Oregon. They have invented the Open Compute Project just for that. They essentially strip anything from their machines that doesn't contribute to efficiency, including the paint. They also don't use UPS and their facility is cooled with good old fans and water and nothing else.

Result:

28MW Facebook facility
One of the least expensive facilities today (100% free cooling)
$210 Millions spent
Thousands of new jobs created
Cost of $7500/kW (vs. $500 of DTM)
PUE 1.09 (vs. <1.01 of DTM)
Millions of gallons of water used every year (DTM wastes zero)

http://www.fbpuewue.com

Sure, Facebook is not Bitcoin mining, but how much more efficient than using fans at a relative cool place could it be? They have all their details publicly available, check their facilities in the Arctic Circle where it's even colder and you will be surprised.

This is not a hype. Passive 2-phase cooling will change the way hardware is cooled and the cloud is built. No matter with 3M or other fluids. Bitcoin is the incubator, just like it's taking chip design to new levels. I have published and article in Electronics Cooling just about this topic, and it made the cover page in March.

http://www.electronics-cooling.com/2014/02/bitcoin-2-phase-immersion-cooling-and-the-implications-for-high-performance-computing/

What makes it work is simple physics and recent changes in the way how chips are made (a lot happened since the 1970 when IBM looked at it first). We are just the first to create some noise, but that doesn't mean we have invented all this.

Bitcoin also takes this to an accelerated path: nobody could have imagined Bitcoiners being the first to make chips with hundreds and hundreds of watt, or being amongst the first to ship 20nm or 16nm devices to the end user, and yet the first 20nm chip comes from KNC.

Enough for today

[NotLambchop]

[snip] ...and your kids will dance to it while it saves you money!

What is it you do for a living antirack?

*whole.  Pitch isn't half as pro with misspelinks,

[antirack]

[snip] ...and your kids will dance to it while it saves you money!

*whole.  Pitch isn't half as pro with misspelinks,

I apologize, engrish is not my native language 

[NotLambchop]

^
Nor mine, but you understood the question, right?

...
What is it you do for a living antirack?
...

[Collider]

^
Nor mine, but you understood the question, right?

...
What is it you do for a living antirack?
...

He works for Allied Control / Data Tank Mining, as documented in this thread about one page ago. (are you that lazy?)

What is it you do for a living?

It seems like you hold the position that immersion cooling isn´t the way to go but you have no idea how much a regular high density datacenter costs?

From experience i know that an online argument is pretty much won when the only thing the other side complains about is their spelling....


The fact of the matter is, that high density datacenter space is expensive AND very limited, as there is a good amount of competition between potential buyers.

If the network keeps growing much more (and i don´t see a stopping point yet) a miner will find himself competing with the big companies for very limited high density hosting space.

It only seems smart to develop a high density DataTank, as you then essentially own your (ultra) high density hosting facility yourself and profit from lower OPcost and bulk electricity rates.
(Alongside many other things)

[Ozziecoin]

Of course the 3 months _initial_ build time is long, but the DataTank plan is long term. After that it only takes a few hours and everything is re-used. Again and again. And don't forget that after the initial phase, the production pipeline is setup and we can produce 1 or a hand full or even more per week.

Re: build time. My understanding is that DTMA will purchase chips at the time it is ready to deploy? Is this correct?

Antirack - can you confirm chip purchase timeline?

[kjs]

Of course the 3 months _initial_ build time is long, but the DataTank plan is long term. After that it only takes a few hours and everything is re-used. Again and again. And don't forget that after the initial phase, the production pipeline is setup and we can produce 1 or a hand full or even more per week.

Re: build time. My understanding is that DTMA will purchase chips at the time it is ready to deploy? Is this correct?

Antirack?

Yes that is correct.

Antirack is Alex Kampl, VP Engineering @ Allied Control & DataTank mining.

That's all folks

[NotLambchop]

^
Nor mine, but you understood the question, right?

...
What is it you do for a living antirack?
...

He works for Allied Control / Data Tank Mining, as documented in this thread about one page ago. (are you that lazy?)

With so many fine offerings available on Havelock, who has the time?

It seems like you hold the position that immersion cooling isn´t the way to go but you have no idea how much a regular high density datacenter costs?

wat?  

From experience i know that an online argument is pretty much won when the only thing the other side complains about is their spelling....

Spelling is just the cherry on the cake.  "A cable x10 as long costs x10 as much" caught my attention, since that comes close to epitomising right-sounding but in reality dead-wrong selling points.  Reading further, I realized the rest of his points were equally valid.
I went on to point out the weakness of each one, punctually:

...a 10 ft electrical cable costs 10 times less than a 100 ft cable to keep it simple and make an example = cost savings again.

Ridiculous.  No one who ever bought Ethernet patch cables, or guitar patch cords, or electrical extension cords would ever say that. With typical lengths (oh, Ethernet cables, 10ft/100ft), the price difference is not x10 but ~2.5.

More important is this:

One single DataTank container is around 576 SP30 from Spondoolies Tech. I have done this example a few days ago already in a previous post.

576 2U enclosures cost a lot of money

...and are not necessary for a permanent mega-scale installation.  The (open) chassis housing multiple boards could well be integrated into custom racks.

576 power supplies cost money

Sure, but if you're suggesting that it's cheaper/more practical to have a few gigawatt monsters, those can exist for air/water cooled setups too.

17,280 (576x30) heatsinks cost money
4608 fans cost money
Etc

Err...  Isn't that Novec stuff ~$350/gal?  How many gallons in a 40ft container?  And all that heat exchanger plumbing and cooling towers, etc, etc.?
Come on

All that needs to be installed, quality controlled (several times), packaged and shipped from China to Israel to Washington State.

Your containers don't need to be shipped, installed or quality controlled?

It needs to be packaged, shipped and quality controlled multiple times in fact (at the manufacturer, at Spondoolies when it comes in, at Spondoolies when it goes out).

Again, not when you have a bulk order of 576 units to be installed at one location.  Each unit isn't packaged individually, in a colorful box with glossy instruction manuals and a few CDs of pr0nz gaems drivers whatever it is.  Bulk shipping is a thing, it's not rocket surgery.  Further, what you'll be shipping will be 100% standard, easily serviceable by semi-trained personnel.  Who would not have to put on scuba gear, power down the entire farm, and dive into boiling Novec to pull out one bad board

So how many days does it take the Israeli factory to build, assembly and prep 576 (15 palettes? maybe more) of hardware? quite a while.

No frickin idea, depends on the factory.  You act like Israel is some third-world country.

How much faster and cheaper can FedEx (we are in a rush in Bitcoin mining) deliver 1 palette with boards compared to 15 palettes? I wouldn't be surprised if the shipping cost alone couldn't buy a new car (I will look into this).

Faster?  Usually not much.  FedEx doesn't hand-carry each box to its destination.  They have big trucks and big planes.  How much faster is it to deliver one letter instead of ten?

I mean sure, there are advantages, but what you get in the end is highly specialized gear that can't be resold without mods, two separate production lines (one for consumer gear and one for stripper boards for submerged use), and a slew of unforeseen complications that come with new technology.  Don't over-hype.

So yeah, there are a few things beyond spelling here.

The fact of the matter is, that high density datacenter space is expensive AND very limited, as there is a good amount of competition between potential buyers.

If the network keeps growing much more (and i don´t see a stopping point yet) a miner will find himself competing with the big companies for very limited high density hosting space.

It only seems smart to develop a high density DataTank, as you then essentially own your (ultra) high density hosting facility yourself and profit from lower OPcost and bulk electricity rates.
(Alongside many other things)

If there is indeed demand for DataTank gear, mining farms/ ASIC manufacturers should be eager to buy into this.  Why fund this through an unlicenced Panamanian exchange by pitching the scheme to laymen?  

Re. what I do for a living:  If I was pitching my company, the question would be both reasonable and relevant.  I'm not, so it isn't.

U may now return to disruptive technologing.

*To proof 2 U I haz pioneerink recearch, sea patint bellow (YES!!  Turbine riggerd up to electricitys generator for cooling towerz and popcorn popping!):

Guise!  
I just made a INVENTION!  
It totally patentid copywrited and tratemarked, so DO NOT STEAL!!1!


*U just wish U could brain like me

[Collider]

I hope it was an ironic / comedic post which nobody expects to be taken seriously (thank you for the nice drawing though).

What are you claiming here, that AC/DTM doesn´t have any patents or that only patented structures are able to generate profit?
Last time I checked, you don´t need a patent for a structure to be profitable.

I don´t know why this is important but yes, 10* the cable length does in fact cost (near) 10* as much. You are not talking 1m and 10m here but 500* 10m-100m.
This was also only meant as an example for certain hardware costs, and never to be taken literally.


Are you seriously complaining that a company is trying to have the average miner get into immersion cooled mining?
It is a great solution to many problems that come with high density miner deployment. (as already frequently mentioned)

Usually people are more upset when the big companies develop technologies and keep them to themselves. Offering anyone the chance to profit who is willing to take the risk is a good idea in my opinion.


If you are upset that DTM is trying to raise funds (for a very likely profitable venture) you are entitled to that opinion, however nobody forces you to buy any shares.



To go through your points:

- cable example: see above. It was meant as an example for some hardware costs, but still stands on a large scale (if you really do want to take it literally)

- Enclosures for the sp30 are in fact needed, as they direct airflow and make the unit stackable / rack mountable.
  The custom rack solution you proposed would need a lot of NRE costs and most likely be more expensive than just fitting enclosures.

- PSU costs are included in the fixed costs of the container. They are regular server grade platinum PSUs with some small tweaks for immersion cooling.
  They power every hardware that goes in the DataTanks and are reusable with every new miner generation.

- The "40ft container" is just the big enclosure, the tanks are much smaller. A very small amount of fluid is needed per blade and the cost is included in the overall cost of 0.5-0.7$/W.

- Shipping costs are saved as miners are lighter, take up less volume etc., the container is shipped only once, but can be shipped to a better location if so preferred (try that with your datacenter )

- DataTank gear needs less testing (during manufacturing) as there are less manufacturing steps (money saved here aswell!).
  There is no need to power down the whole farm for servicing (I don´t know how you have come up with that).
  There is also no need to put on any special protective gear (though putting on gloves shouldn´t set you back to many hours )

- It obviously takes more time to completely manufacture finished miners than to just get PCB blades from the factory, ready to be deployed in the tanks. (It not only saves time but also labor = money)

- Havelock is by far the most professional Bitcoin exchange out there. They properly vet any new stocks and the companies behind them, so nobody simply runs off with your invested funds.
  Who is to say professional mining farms / manufacturers aren´t interested in this?
  As a fact, AM has a 0.5MW immersion cooled facility (using the same basic technology), and it was built by AC, the "parent company" of DTM.