Consider this : Power estimates of the Stampede3 put it at somewhere around 30-50MW. An entire power plant puts out 500mW. So this thing takes 1/10 of a power plant (about 1/20 of a nuclear reactor but we don't build those any more). Meanwhile, people in Texas are suffering brown-outs and black-outs because their whole infrastructure is broken. During the famous winter storm of 2021 when people did not have power, Texas did not even make the school shut down the existing supercomputers! They called it "a critical datacenter" and left them all running. Now consider they have multiple supercomputers running in TX, each one using enough power for 20,000 homes (or more in an emergency). But yeah, let's build bigger and more power hungry supercomputers, not upgrade our power infrastructure, and not even TRY and make these computers more power efficient. Even the most conservative of estimates puts the largest supercomputers passing the 500MW line in about 15 years. A whole power plant for one computer? It's NOT sustainable. Both Academia and Gov't need to pivot from raw computing FLOPS to significantly driving the power efficiency up.
This is a bizarre complaint. Texas’ infrastructure problems are a political problem created by its own corrupt government. It doesn’t matter individually who is using how much power until that is solved, no one person or entity can “voluntarily” solve the problem on their own. Only government can solve the problem, and it’s the broken government that created the problem.
Peak demand during the Texas 2021 winter storm was 77,000MW. Even if Stampede2 consumed 50MW in the first place (it doesn’t), improving efficiency by 50% wouldn’t make a difference. The power grid still would have collapsed under the conditions it faced.
Improving efficiency is important, but not at all for the reason you mentioned. You can’t fix regulatory corruption with more efficient computers.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
5 Comments
Back to Article
brucethemoose - Wednesday, July 26, 2023 - link
This is similar to Fujitsu's A64FX supercomputer, which some US government researchers spoke highly of.jjjag - Wednesday, July 26, 2023 - link
Consider this : Power estimates of the Stampede3 put it at somewhere around 30-50MW. An entire power plant puts out 500mW. So this thing takes 1/10 of a power plant (about 1/20 of a nuclear reactor but we don't build those any more). Meanwhile, people in Texas are suffering brown-outs and black-outs because their whole infrastructure is broken. During the famous winter storm of 2021 when people did not have power, Texas did not even make the school shut down the existing supercomputers! They called it "a critical datacenter" and left them all running. Now consider they have multiple supercomputers running in TX, each one using enough power for 20,000 homes (or more in an emergency). But yeah, let's build bigger and more power hungry supercomputers, not upgrade our power infrastructure, and not even TRY and make these computers more power efficient. Even the most conservative of estimates puts the largest supercomputers passing the 500MW line in about 15 years. A whole power plant for one computer? It's NOT sustainable. Both Academia and Gov't need to pivot from raw computing FLOPS to significantly driving the power efficiency up.Ryan Smith - Wednesday, July 26, 2023 - link
"Power estimates of the Stampede3 put it at somewhere around 30-50MW"You need to check the source of those figures. That's roughly the power budget for Aurora; Stampede3 is a fraction of that size.
shelbystripes - Monday, July 31, 2023 - link
This is a bizarre complaint. Texas’ infrastructure problems are a political problem created by its own corrupt government. It doesn’t matter individually who is using how much power until that is solved, no one person or entity can “voluntarily” solve the problem on their own. Only government can solve the problem, and it’s the broken government that created the problem.Peak demand during the Texas 2021 winter storm was 77,000MW. Even if Stampede2 consumed 50MW in the first place (it doesn’t), improving efficiency by 50% wouldn’t make a difference. The power grid still would have collapsed under the conditions it faced.
Improving efficiency is important, but not at all for the reason you mentioned. You can’t fix regulatory corruption with more efficient computers.
tipoo - Friday, July 28, 2023 - link
Is there any sort of public third party overview of how Ponte Vecchio is performing vs the competition?