Secondly, the 'rube goldberg' device pictured above is about 75% efficient at transferring heat to the loop at low temps and it drops to about 50-60% at higher temps above ~160. The second prototype is submerged in a silicon fluid and well insulated, Sri (senior scientist) is modeling the device now in Comsol
http://www.comsol.com/comsol-multiphysics?gclid=CNnmyLeI2MACFSsV7AodCTsALQ and its looking like we'll easily break into the mid 90% efficiency for heat capture with the device when submerged.
Are you saying that currently only 55% of the heat stays in the loop when you want to use it for refrigeration/AC? 45% waste?
**No, I'm saying the current prototype is not very efficient at managing heat. All the plumbing that moves the coolant is uninsulated and looses a fair amount of heat - the hotter it gets vs ambient the bigger the loss. We're not investing in the first prototype anymore, its already proven the parts can manage the heat we need to make.
As for the economics - free energy for computation isn't going to be economic? Remember we are using energy once and getting two benefits - heat and computation. You pay for the heat, the computation is free (from an energy perspective) except for the cost of the device.
Thanks-
Correct, it will not be economic.
Here's a simplified comparison for 1MW worth of computing power:
A: Classic datacenter
Capex:
- hardware: $1,000,000
- Warehouse+infrastructure: $200,000
Opex(yearly):
- Electricity: $570,000 (1.3MW at $0.05/kwh)
- Employees:$100,000
5 Year total: $4,500,000
| B: Distributed computing (using your method)
Capex:
- Hardware: $5,000,000*
- Cooling recovery system: $5,000,000* (~$1/W at 5MW)
- Distributing/shipping: $500,000
Opex:
- Employees: $500,000**
5 year total: $13,000,000
|
*Assuming it would require 5 times the amount of hardware because it's being used 25% of the time due to none of your applications requiring 100% load 100% of the time and a 20% efficiency loss due to the hot chips. This is also assuming the system is 100% efficient at recovering the heat but your post suggest it might be closer to 50%.
**If sized correctly to the load it should run far more than the 20-30% utilization rate the typical datacenter sees. There are about 120 different distributed computing networks out there now (a good Wiki search) and they are growing in number every day. Bitcoin is a distributed computing network. The idea idea is to compute whenever there is a need to generate heat, when the system is sized right that need can be pretty consistent. Again - we would not continue to develop this design, it is just the first prototype.
**I've got no idea how many people it takes to properly run a distributed computing network but I'm sure it takes more than the 2 people needed top operate a large scale bitcoin mine. (guessing at least 10 employees)** If memory serves Stanfords folding at home network is 250k distributed machines across the planet and everyone who donates their computer idle time is a part of that network. They have a handful of techs that keep the software and their server running as well as spend time on the forums to answer questions and talk through bug fixes. Not too many.
So basically you will have ~$9.3M extra startup costs so you can save ~$170k/year.. In this scenario it would take more than 50 years to pay off.
Overall cool idea, just entirely unpractical like solar roadways.
**Not the model - a valiant effort but, seriously, who would do that? And... why would we publish the business model on a forum?
Side note: Do you plan on having certified technicians around the country ready to fix peoples heater/AC/refrigeration systems when they stop working? Will you be reimbursing customers for all their spoiled food when the computer crashes for a few hours?
**Nope - there are service models that don't require techs all around the world, at scale this will be more like an appliance, not a computer.