[aiappreciator]

H100s consume like 350W. 3000 H100s = 350kW 350*0.2h = 70kWh, about $20 at $0.3/kWh (lets assume data centers have expensive electrical costs).

[SushiHippie]

Ah Thanks! Taking the data from here[0] for the USA, this would be about 25,69kg of CO2.

[0] https://ourworldindata.org/grapher/carbon-intensity-electric...

[alwa]

Which, per the EPA [0], is roughly the tailpipe CO2 you’d emit by burning a little under 3 gallons of gas in your car, right? Which is to say, driving 65ish miles in the US [1]?

It amazes me that computational feats of this magnitude can be so energy efficient in the scheme of things.

[0] https://www.epa.gov/greenvehicles/tailpipe-greenhouse-gas-em...

[1] https://www.epa.gov/greenvehicles/tailpipe-greenhouse-gas-em....

[NhanH]

It’s not that surprising actually, computation moves nothing and all energy has to turn into heat. If computation would use as much energy as moving physical objects, the heat would burn everything down

[nl]

If you are on GCP you can choose low carbon data centers: https://cloud.google.com/sustainability/region-carbon

[YetAnotherNick]

Or about 3s of 747 cruise[0]

[0]: https://www.carbonindependent.org/22.html#:~:text=At%20a%20c...

[shaklee3]

You can't really use 350W. It's very rare for a workload to consume the full TDP.

[pinkcan]

this is napkin math, for sure there are other components in the system that do not directly contribute to the computation – cooling systems, DC lights