4 ms·

14,975,435MWh for one year (8760 hrs) = 1709 MW load. Or about 1.7 mil servers using 1kW each (total, including its share of data center cooling). That's a lo

by rmah 2y ago

14,975,435MWh for one year (8760 hrs) = 1709 MW load. Or about 1.7 mil servers using 1kW each (total, including its share of data center cooling). That's a lotta servers. Another way to look at it.. they say they have about 3.14 bil daily active users across all platforms. So that's about 13Wh per daily user session. At $0.15/kWh, that's about $0.002 of electricity which doesn't seem that horrible.

fifilura 2y ago
It also corresponds to 2-3 nuclear reactors. I am also not shocked. But somewhat surprised by the choice of unit in the article.
- bamboozled 2y ago
  What is the output of the three reactors worth of electricity? Hot takes and shit posts ?
  - manvillej 2y ago
    the newest 4 nuclear reactors in the USA produce 4500 megawatt hours. https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Plant https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Pla... these are newer and more efficient. so they are likely producing more than the average reactor. So its closer to 1.25 reactors worth of electricity. but data centers are a HUGE driver of electricity consumption.
    - fifilura 2y ago
      Whenever there is a post about nuclear it is always about the latest and greatest. Or in the future - "how many SMRs is that?". I should call it the law-of-something. I feel reasonably comfortable with the estimated 2-3 reactors. I could have said 1-4 but... (there are lots of reactors around 4-500MW too)
      - SECProto 2y ago
        If you said 1-4, I would not have replied - that is an appropriate bound for how many traditional nuclear plants it takes to provide 1.7GW 2 to 3 is accurate for Canada, where the last nuclear plant to connect to the grid was 31 years ago. For the US, the range is 1.25 to 3.5 (four reactors in the last 30 years). For China, the range is 1.05 to 8 (56 reactors built in the last 30 years, but the 200 and 300 MW reactors are kind of special, so 1.05 to 4 is a reasonable range). For France, 1.15 to 2.0 (five reactors in the last 30 years).
- SECProto 2y ago
  > It also corresponds to 2-3 nuclear reactors One reactor, if you use a modern design! https://en.m.wikipedia.org/wiki/EPR_(nuclear_reactor) https://en.m.wikipedia.org/wiki/EPR_(nuclear_reactor) (Or maybe 2 anyway, to deal with downtime)
  - preisschild 2y ago
    One might argue that the EPR is actually a bit too big and thus has become a lot more difficult to build
    - SECProto 2y ago
      One might argue that, but unless the one arguing it has been involved with the construction of an EPR, I would not put a lot of stock in it. My viewpoint (having involvement with many large (>$100M) construction projects, but not power plants): I don't believe the constructability issues with the EPR have anything to do with the size, specifically. I suspect the root issue is design and construction being too separated on the EPRs.
      - sl-1 2y ago
        But unit size does have other considerations outside the project size. For example provisioning capacity for maintenance downtime. Too big units make it quite hard to get enough capacity for the grid to stay up while maintaining hard. For example OL3 in Finland has to limit its max power because if it trips, the grid needs to shed enough load to stay up: (in finnish) https://www.fingrid.fi/kantaverkko/sahkonsiirto/olkiluoto-3-kantaverkkoon/olkiluoto-3n-jarjestelmasuoja/ https://www.fingrid.fi/kantaverkko/sahkonsiirto/olkiluoto-3-...
        SECProto 2y ago
        Yes, definitely a consideration - ol3 provides a significant fraction of Finland's power (~8% according to wiki). Having this large a fraction of any grid trip off would be disastrous. Grid managers have to consider similar things anywhere there is a single source providing a large fraction of the power. Kind of beside the point of meta using the equivalent of 1-4 nuclear plants of capacity.
wrycoder 2y ago
1.5 B$ per year at $0.15/KWh.
- fifilura 2y ago
  Some of those servers are located in north Sweden, where electricity is cheaper (for now [1]) I expect it to be even cheaper for example on Iceland with an abundance of hydro power. [1] They are also trying to redefine steel-processing there and that will use a lot of electricity if they succeed.
Someone 2y ago
> So that's about 13Wh per daily user session Plus ~1/2000th of a server, divided by the life time of a server on their premises in days. Let’s say a millionth of a server. At $1000 a server, that adds about $0.001 a day, if my math is correct. Actual numbers may be higher. What do they pay for their hardware, at the scale they buy it? Plus installation and management, plus the cost of the building.
razakel 2y ago
>3.14 bil daily active users That's about 2 in 5 people on the planet. Astonishing.

by rmah 2y ago

fifilura 2y ago

It also corresponds to 2-3 nuclear reactors. I am also not shocked. But somewhat surprised by the choice of unit in the article.

bamboozled 2y ago
What is the output of the three reactors worth of electricity? Hot takes and shit posts ?
- manvillej 2y ago
  the newest 4 nuclear reactors in the USA produce 4500 megawatt hours. https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Plant https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Pla... these are newer and more efficient. so they are likely producing more than the average reactor. So its closer to 1.25 reactors worth of electricity. but data centers are a HUGE driver of electricity consumption.
  - fifilura 2y ago
    Whenever there is a post about nuclear it is always about the latest and greatest. Or in the future - "how many SMRs is that?". I should call it the law-of-something. I feel reasonably comfortable with the estimated 2-3 reactors. I could have said 1-4 but... (there are lots of reactors around 4-500MW too)
    - SECProto 2y ago
      If you said 1-4, I would not have replied - that is an appropriate bound for how many traditional nuclear plants it takes to provide 1.7GW 2 to 3 is accurate for Canada, where the last nuclear plant to connect to the grid was 31 years ago. For the US, the range is 1.25 to 3.5 (four reactors in the last 30 years). For China, the range is 1.05 to 8 (56 reactors built in the last 30 years, but the 200 and 300 MW reactors are kind of special, so 1.05 to 4 is a reasonable range). For France, 1.15 to 2.0 (five reactors in the last 30 years).
SECProto 2y ago
> It also corresponds to 2-3 nuclear reactors One reactor, if you use a modern design! https://en.m.wikipedia.org/wiki/EPR_(nuclear_reactor) https://en.m.wikipedia.org/wiki/EPR_(nuclear_reactor) (Or maybe 2 anyway, to deal with downtime)
- preisschild 2y ago
  One might argue that the EPR is actually a bit too big and thus has become a lot more difficult to build
  - SECProto 2y ago
    One might argue that, but unless the one arguing it has been involved with the construction of an EPR, I would not put a lot of stock in it. My viewpoint (having involvement with many large (>$100M) construction projects, but not power plants): I don't believe the constructability issues with the EPR have anything to do with the size, specifically. I suspect the root issue is design and construction being too separated on the EPRs.
    - sl-1 2y ago
      But unit size does have other considerations outside the project size. For example provisioning capacity for maintenance downtime. Too big units make it quite hard to get enough capacity for the grid to stay up while maintaining hard. For example OL3 in Finland has to limit its max power because if it trips, the grid needs to shed enough load to stay up: (in finnish) https://www.fingrid.fi/kantaverkko/sahkonsiirto/olkiluoto-3-kantaverkkoon/olkiluoto-3n-jarjestelmasuoja/ https://www.fingrid.fi/kantaverkko/sahkonsiirto/olkiluoto-3-...
      - SECProto 2y ago
        Yes, definitely a consideration - ol3 provides a significant fraction of Finland's power (~8% according to wiki). Having this large a fraction of any grid trip off would be disastrous. Grid managers have to consider similar things anywhere there is a single source providing a large fraction of the power. Kind of beside the point of meta using the equivalent of 1-4 nuclear plants of capacity.

wrycoder 2y ago

1.5 B$ per year at $0.15/KWh.

fifilura 2y ago
Some of those servers are located in north Sweden, where electricity is cheaper (for now [1]) I expect it to be even cheaper for example on Iceland with an abundance of hydro power. [1] They are also trying to redefine steel-processing there and that will use a lot of electricity if they succeed.

Someone 2y ago

> So that's about 13Wh per daily user session Plus ~1/2000th of a server, divided by the life time of a server on their premises in days. Let’s say a millionth of a server. At $1000 a server, that adds about $0.001 a day, if my math is correct. Actual numbers may be higher. What do they pay for their hardware, at the scale they buy it? Plus installation and management, plus the cost of the building.

razakel 2y ago

>3.14 bil daily active users That's about 2 in 5 people on the planet. Astonishing.