Quote from: antirack on May 05, 2014, 05:08:52 AM

Quote from: KS on May 04, 2014, 11:17:22 AM

Quote from: Jutarul on May 01, 2014, 10:41:54 PM

What this likely means for mining equipment providers is that in the intermediate term the ability to migrate and maintain beats efficiency. That's why developments by Allied Control are so interesting which focus on ease of deployment.

Here you're opening a whole new can of worms. I think that would depends on your OPEX. CAPEX is high with the tanks but if maintenance (OPEX) is low enough, it would be a decent target for countries where labor and space are expensive (but what of electricity costs, what of heat dissipation power usage?). If OTOH you have cheap labor and space, the tanks aren't looking that good anymore.

Also, unless there is a way to stack them, they won't really help you improve the density of your deployment (they have other issues with connectors, plumbing, etc) vs a regular datacenter (more on that after the end of June, I guess). I'm not really sold on the tanks, I think the idea is neat, but the numbers don't convince me so far (probably need more data).

The OPEX and CAPEX of immersion cooling is lower. Unless you are talking about some home miner of course, not the 1.2MW container mining units.

Quote

Isn't immersion cooling incredible expensive?

No it's not. 3M fluids and immersion cooling have a reputation of being more expensive. What most people forget is that you only need very little fluid for a lot of heat, and that a tank full with hundreds of boards costs a lot less than dozens of racks with machines with cases or fans.

To give you an example, we cool 4kW (4000W) in a small space of 1 liter volume with only 200cc of fluid. Even heatsinks and fans for hardware with 4000W will cost more than just 200cc of fluid. And that is just one side of the equation. To make a long story short, it is less expensive to build and to run. That's the value proposition of passive 2-phase immersion cooling.

If you factor in device cooling (heatsinks, cold plates, pumps, hoses, radiators, speciality racks, air conditions, etc), it costs less to build a container with a few tanks from the first day. But where immersion really shines is when you start reusing all of this for your next-gen hardware, you basically get everything for free, you already own everything you will ever need. A single person can deploy hundreds of boards within hours, instead of days or weeks.

The reduced electricity cost for cooling is a bonus on top of all that.

Source:
https://docs.google.com/a/allied-control.com/document/d/1zQwlGoDfHgqlPFlLuUviKzkSUepWMTKcwGaNI_nqNjg/edit

Quote

they have other issues with connectors, plumbing, etc

A 200kW tank for the AM gen-3 boards has exactly one water hose in and one water hose out. That is for 32,000 AM BE200 chips. No plumbing issues.

I am totally sold Grin

Well, OPEX would be about 20-30% cheaper for me in a DC for a 240KW farm (includes electricity and rack costs). It gets slightly worse if you include CAPEX (but I haven't because I'm not sure of the useful lifetime of the hardware).

With 120 racks (2KW per rack), I would also get a better rate than the one I currently have (I could theoretically also build my own, but I would still be smaller than the ones with multiple DCs and they would still have better yet electric rates - a big factor here is the electricity "transport" cost, also there'd be all the associated building management "stuff" to deal with).

Setup would be more expensive in a DC than for a tank, but the OPEX difference is so big that it wouldn't matter.

I'm in a densely populated, high electric cost area. The tanks would look much better if I lived, say, in Washington state.