Unless things change, first zettaflop systems will need nuclear power, AMD's Su says

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
Things are starting to slow down at the top

"The AMD CEO makes the case that it may be time to make heavier use of AI and machine learning in HPC. And she's not alone in thinking this. Nvidia and Intel have both been pushing the advantages of lower precision compute, particularly for machine learning where trading a few decimal places of accuracy can mean the difference between days and hours for training.

Nvidia has arguably been the most egregious, claiming systems capable of multiple "AI exaflops." What they conveniently leave out, or bury in the fine print, is the fact they're talking about FP16, FP8, or Int8 performance, not the FP64 calculations typically used in most HPC workloads.

"Just taking a look at the relative performance over the last 10 years, as much as we've improved in traditional metrics around SpecInt Rate or flops, the AI flops have improved much faster," the AMD chief said. "They've improved much faster because we've had all these mixed precision capabilities."

One of the first applications of AI/ML for HPC could be for what Su refers to as AI surrogate physics models. The general principle is that practitioners employ traditional HPC in a much more targeted way and use machine learning to help narrow the field and reduce the computational power required overall.

Several DoE labs are already exploring the use of AI/ML to improve everything from climate models and drug discovery to simulated nuclear weapons testing and maintenance.

"It's early. There is a lot of work to be done on the algorithms here, and there's a lot of work to be done in how to partition the problems," Su said."

Source: https://www.theregister.com/2023/02/23/amd_zettaflop_systems_nuclear/
 
Nuclear power is A - available, and B - nearly unlimited. If it will be required, it will be done.

But will it?

"On October 19, 2016, TVA's Unit 2 reactor at the Watts Bar Nuclear Generating Station became the first US reactor to enter commercial operation since 1996. "

Nuke power has really fallen off since three mile Island. Not really a viable option since practically none are being built and the current infrastructure is aging rapidly.

"As of September 2017, there are two new reactors under construction with a gross electrical capacity of 2,500 MW, while 39 reactors have been permanently shut down"
 
Nuclear lost lobbyist war in the collective West, first in the States, last in France which seems to be the state of play right now. However if interests deemed it necessary, it will be reintroduced. Pushing information processing limits as a requirement, may be the ticket to bring nuclear interests back to the forefront over the coming decades. Tech is way better today than in the '70's so new plants will be lower risk, while at the same time they could repurpose old sites.
 
Nuclear lost lobbyist war in the collective West, first in the States, last in France which seems to be the state of play right now. However if interests deemed it necessary, it will be reintroduced. Pushing information processing limits as a requirement, may be the ticket to bring nuclear interests back to the forefront over the coming decades. Tech is way better today than in the '70's so new plants will be lower risk, while at the same time they could repurpose old sites.
That's what they said in the early 2000's.. new nuke tech - it will be a renaissance". Well, thanks to government regulation and the fact that no one wants a reactor in their back yard - it never happened and likely never will. Renewables appear to be the new emphasis, nuke is really still dead - at least in this country. It will take a change of heart in the political and public arenas to get the nuke permits flowing again. New nuke tech in an of itself is not it's savior.
 
But will it?

"On October 19, 2016, TVA's Unit 2 reactor at the Watts Bar Nuclear Generating Station became the first US reactor to enter commercial operation since 1996. "

Nuke power has really fallen off since three mile Island. Not really a viable option since practically none are being built and the current infrastructure is aging rapidly.

"As of September 2017, there are two new reactors under construction with a gross electrical capacity of 2,500 MW, while 39 reactors have been permanently shut down"
This is an administrative problem, not a technological one. Still a problem none the less, but it's a people one. Those we can fix IF we want to.

Are we going to? Hey your guess is as good as mine.
 
Agreed that it is fairly dead at the moment, in the West in particular, but this information processing power limit may be the motivation required for renaissance.
 
Agreed that it is fairly dead at the moment, in the West in particular, but this information processing power limit may be the motivation required for renaissance.
It's not going to be the processing power that brings nuclear back to the fore, but the critical failure of "renewable" energy like wind and solar to meet the world's overall increasing power demands. More nuclear is the only practical way forward.
 
Things are starting to slow down at the top

"The AMD CEO makes the case that it may be time to make heavier use of AI and machine learning in HPC. And she's not alone in thinking this. Nvidia and Intel have both been pushing the advantages of lower precision compute, particularly for machine learning where trading a few decimal places of accuracy can mean the difference between days and hours for training.

Nvidia has arguably been the most egregious, claiming systems capable of multiple "AI exaflops." What they conveniently leave out, or bury in the fine print, is the fact they're talking about FP16, FP8, or Int8 performance, not the FP64 calculations typically used in most HPC workloads.

"Just taking a look at the relative performance over the last 10 years, as much as we've improved in traditional metrics around SpecInt Rate or flops, the AI flops have improved much faster," the AMD chief said. "They've improved much faster because we've had all these mixed precision capabilities."

One of the first applications of AI/ML for HPC could be for what Su refers to as AI surrogate physics models. The general principle is that practitioners employ traditional HPC in a much more targeted way and use machine learning to help narrow the field and reduce the computational power required overall.

Several DoE labs are already exploring the use of AI/ML to improve everything from climate models and drug discovery to simulated nuclear weapons testing and maintenance.

"It's early. There is a lot of work to be done on the algorithms here, and there's a lot of work to be done in how to partition the problems," Su said."

Source: https://www.theregister.com/2023/02/23/amd_zettaflop_systems_nuclear/
Easy fix don't use plutonium or uranium just use thorium. If you use thorium the ecogeeks will all be on your side.
 
Electricity is still a young technology.
I would not be surprised if many more uses for electricity are invented only after "Electrical Scarcity" is over with.
And it certainly helps being deep into the computer era.
 
what about fusion
You still need power to create fusion. The goal is to output more power than is consumed. The lasers and magnets used in the current known processes consume huge amounts of power. Even though it was big news that they hit the goal of positive power output, it was only for a short period of time. If fusion is going to be practical, a huge technological breakthrough needs to be made to make that kind of output both larger and sustainable.
 
You still need power to create fusion. The goal is to output more power than is consumed. The lasers and magnets used in the current known processes consume huge amounts of power. Even though it was big news that they hit the goal of positive power output, it was only for a short period of time. If fusion is going to be practical, a huge technological breakthrough needs to be made to make that kind of output both larger and sustainable.

Yeah, it's a simple logic to understand that before we go for fusion, we need to eliminate energy scarcity, probably with thorium reactors.
Once we have lots of energy to draw from, it'll be no problem at all to continue to experiment with fusion.
 
Yea, we all like to bet on fusion but it isn't 'invented' yet. Everyone should continue the work on it but no sense building the air travel agencies before someone invents the plane. Nor should we abandon any current technologies, keep improving them. Better yields, less waste, less pollution. We need to stop thinking in exclusivity. Use renewables where it makes sense, use fuel where it makes sense etc.
 
You still need power to create fusion. The goal is to output more power than is consumed. The lasers and magnets used in the current known processes consume huge amounts of power. Even though it was big news that they hit the goal of positive power output, it was only for a short period of time. If fusion is going to be practical, a huge technological breakthrough needs to be made to make that kind of output both larger and sustainable.
what if after producing more power than is consumed that you get to enough level to cut the cords to the previous supply of power and the fusion reactor is fed back into tself?

you know, with elegant timing and all so there's no disruption
 
what if after producing more power than is consumed that you get to enough level to cut the cords to the previous supply of power and the fusion reactor is fed back into tself?

you know, with elegant timing and all so there's no disruption
What's the difference? Your net power out is the same.
 
What's the difference? Your net power out is the same.
nah, i mean once it's efficient enough to account for not only the ignition, but also able to overcome the net power being the same to the point of not only being able to cut the cord to the power supply, and backfed into itself, but also still have a net positive reserve to deliver outside the fusion reactor
 
The red tape on reactors is insane, to build one in North America you need to pay for it 100% up front and have a cash account set aside that is fully funded that contains an additional 20% for cost overruns.

That is on top of permits, studies, infrastructure, blah blah blah.

Nobody builds them in NA right now because they have been regulated to a point where they aren’t cost effective.

This is one of the driving factors behind the mini reactors.
 
Last edited:
Nobody builds them in NA right now because they have been regulated to a point where they aren’t cost effective.
If only there was a type of organization that could build and maintain expensive large-scale infrastructure in the public interest.
 
The US government has no interest in building new nuclear power plants. They're the ones who implemented the unreasonable red tape for building them in the first place.
And that's because the American people, out of willful ignorance and "Not In My Backyard" selfishness, demanded that back then. As much as people pretend otherwise, the people at-large control the government. We just keep electing people that campaign on (and take action on behalf of) policies that are not in our interest. If we truly want nuclear power again, then elect people who support it.
 
Last edited:
The US government has no interest in building new nuclear power plants. They're the ones who implemented the unreasonable red tape for building them in the first place.
That would also be seen as government overreach because they privatized the energy sector, so now you would have private for-profit corporations being forced to compete with a government-at-cost service.
Personally, I believe something as important as electrical infrastructure should never have been privatized, but that's just me.
 
I get the overly harsh regulations with something like nuclear, you don't want companies just being able to say "thanks for the money, we're out now" I mean who is in charge of nuclear waste? Sure as hell isn't the private companies that created the reactors dealing with their disposal.

Also whenever you mention anything about nuclear waste issues someone is bound to knee jerk "but breeder reactors" or whatever fancy buzzword they read about, the reality is why yes NIMBYism and the lack of politicians with any level of spine to potentially upset voters does keep nuclear out of the US, the reality is no one wants the liability of them and I for one am kind of happy the government isn't allowing a pass on that. That said, I'm not anti-nuclear at all, I simply understand how corporations tend to work and they most definitely do not do so for the betterment of mankind.
 
I get the overly harsh regulations with something like nuclear, you don't want companies just being able to say "thanks for the money, we're out now" I mean who is in charge of nuclear waste? Sure as hell isn't the private companies that created the reactors dealing with their disposal.

Also whenever you mention anything about nuclear waste issues someone is bound to knee jerk "but breeder reactors" or whatever fancy buzzword they read about, the reality is why yes NIMBYism and the lack of politicians with any level of spine to potentially upset voters does keep nuclear out of the US, the reality is no one wants the liability of them and I for one am kind of happy the government isn't allowing a pass on that. That said, I'm not anti-nuclear at all, I simply understand how corporations tend to work and they most definitely do not do so for the betterment of mankind.
Exactly this, they are too expensive to run and maintain, and corporations will always look to cut costs, and that will always 100% of the time lead to a problem and the potential scale of the problem isn't something you can just "whoops our bad!".
So you will get improperly disposed of waste, and they will overwork 2 inspectors instead of hiring 4, that part can run another 2 months before being replaced because we don't want downtime during peak AC season. There are so many cost-cutting measures we can all imagine and they are all sadly plausible.
This is why GE, Lockheed martin, and a bunch of others are working on those small disposable reactors. Build them, bury them, nothing more than preventative maintenance for 20 years, then the whole thing gets tossed and replaced.
Too small to meltdown in the movie sense.
 
This is why GE, Lockheed martin, and a bunch of others are working on those small disposable reactors. Build them, bury them, nothing more than preventative maintenance for 20 years, then the whole thing gets tossed and replaced.
Too small to meltdown in the movie sense.
what's the output on them?
 
Nuclear lost lobbyist war in the collective West, first in the States, last in France which seems to be the state of play right now. However if interests deemed it necessary, it will be reintroduced. Pushing information processing limits as a requirement, may be the ticket to bring nuclear interests back to the forefront over the coming decades. Tech is way better today than in the '70's so new plants will be lower risk, while at the same time they could repurpose old sites.
Nuclear is thriving in France, I think they're already carbon neutral due to their nuclear plants. As much as I dislike Macron he's pushing a nuclear initiative all across Europe.
 
Easy fix don't use plutonium or uranium just use thorium. If you use thorium the ecogeeks will all be on your side.
I like it but don't tell them about thorium's dirty little secret - it's not a fissile fuel on its own. It requires a neutron source and uranium is the easiest way to do that. Once that happens, it turns into a short lived isotope of uranium - U233.

There's still waste and all the other problems that go along with fission just far less.
 
  • Like
Reactions: erek
like this
I like it but don't tell them about thorium's dirty little secret - it's not a fissile fuel on its own. It requires a neutron source and uranium is the easiest way to do that. Once that happens, it turns into a short lived isotope of uranium - U233.

There's still waste and all the other problems that go along with fission just far less.
They’re making strides

https://www.newsweek.com/energy-sav...r-solves-achilles-heel-nuclear-fusion-1784256

https://iopscience.iop.org/article/10.1088/1741-4326/aca54e/meta#nfaca54ef2
 
Back
Top