# Exascale Supercomputers Power Consumption

I don't know if its the right place to ask. I have read a lot of articles about exascale and found out that it may consumes approximately 20MW power envelope. Is it a daily basis or a yearly basis or in an instant? Please enlighten me. Here are the papers I have used.

http://www.cs.utexas.edu/~skeckler/pubs/SC_2014_Exascale.pdf

page 1:

One of the main challenges in achieving this goal is power consumption, which a range of HPC system operators have suggested be limited to 20 MW for a full exascale-capable system to mitigate cost of ownership and new power delivery infrastructure costs [2]

and

http://www.computermachines.org/joe/publications/pdfs/hpca2017_exascale_apu.pdf

page 1

An exascale supercomputer is envisioned to comprise of on the order of 100,000 interconnected servers or nodes in a target power envelope of ∼20MW, with sufﬁcient memory bandwidth to feed the massive compute throughput, sufﬁcient memory capacity to execute meaningful problem sizes, and with user intervention due to hardware or system faults limited to the order of a week or more on average [3]

• I think your basic confusion is between power (measured in watts) and energy (usually measured in joules but often in kilowatt-hours for the commercial grid). These are different things altogether and you need to understand what the terms mean. – Elliot Alderson Aug 29 '18 at 14:40

MW (megawatts) is not a unit related to time, it is the power usage in a moment in time.

According to the formula P = V * I, P (in watts) is the product of voltage times current. I doubt the computer uses standard 120V of 220V but let's assume it does, it means the computer uses 20MW = 110V * I => I = about 180 kA (kiloamperes).

To translate this in time, the unit kWh is used (kilowatt-hour), meaning the average number of kilowatts used over one hour. Thus if this computer would be on full power for one hour, it would consume 20 MWh.

• So what does it mean? An exascale supercomputer can consume 20MW in an instant time? Please help me understand this. – Cubilla Aug 29 '18 at 14:24
• @Eliyah MW is power, not energy. It is a rate, not an amount. MW multiplied by time would be an amount. – evildemonic Aug 29 '18 at 14:54
• @evildemonic I think your use of the word "amount" is confusing. One can talk about an amount of power or an amount of energy, but you seem to imply that only energy can be an amount. – Elliot Alderson Aug 29 '18 at 16:12
• @Eliyah, what does it mean? It means that you will have to pay $2,000 for each hour when this conglomerate is turned on (assuming 10c per kWh of electricity), or$48,000 each day. Multiply this for 365 days if you want a yearly budget. – Ale..chenski Aug 29 '18 at 16:41
• @ElliotAlderson Good catch, I should have used the word "quantity" to be less ambiguous. – evildemonic Aug 29 '18 at 17:44

Remember that watts is an power measurement, joules is an energy measurement. Watts is joules per second. So 20MW is 20 million joules per second. This is useful because it tells you how many computations you can make and how many transistors you can have.

No matter the computing device, whether it be a brain, or a exascale computer, every gate, memory cell (or neuron or whatever your computing with) takes energy to run or switch. If you want to make an exascale computer or 1 quintillion (10^18) operations this will take a lot of energy. Source: http://userweb.eng.gla.ac.uk/douglas.paul/SiGe/limits.html

20MW is a lot, it's 20000kW, the average American home uses roughly 1200W or 1.2kW. (or 897kWh per month, or 1245W) That means one exascale computer would take the same power as 16000 homes!

It also means it would take the power distribution size of the same size as 16000 homes.

• 897 sounds quite a lot, we use a bit more than 300 around here (not us) so does that maybe reflect the use of electricity for heating there? – PlasmaHH Aug 29 '18 at 17:26
• It's the average for national use, but it does sound excessive, I suppose that it must be the 1% that are using all of the energy. I think my bill is closer to 300 – Voltage Spike Aug 29 '18 at 17:27