Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

El Capitan: New supercomputer is the fastest

spectrum.ieee.org

Post reply on HN
Print view View original

81–90 of 106 posts

Re: El Capitan: New supercomputer is the fastest

#81

Post by sliken »

olao99 wrote: post #7

I fail to understand how these nuclear bomb simulations require so much compute power. Are they trying to model every single atom? Is this a case where the physicists in charge get away with programming the most inefficient models possible and then the administration simply replies "oh I guess we'll need a bigger supercomputer"

Well there's a fair bit of chemistry related to the explosions to bring the sub-critical bits together. Time scales are in the nanosecond range. Then as the subcritical bits get closer obviously the nuclear effects start to dominate. Things like berrylium are used to reflect and intensive the chain reaction. All of that is basically just a starter for the fusion reaction. That often involved uranium, lithium deturide, and more plutonium.

So it involves very small time scales, chemistry, fission, fusion, creating and channeling plasmas, high neutron fluxes, extremely high pressures, and of course the exponential release of amazing amounts of energy as matter is literally converted to energy and temperatures exceeding those in the sun.

Then add to all of that is the reality of aging. Explosives can degrade, the structure can weaken (age and radiation), radioactive materials have half lives, etc. What should the replacement rate be? What kind of maintenance would lengthen the useful lives of the weapons? What fraction of the arsenal should work at any given time? How will vibration during delivery impact the above?

Seems like plenty to keep a supercomputer busy.

Re: El Capitan: New supercomputer is the fastest

#82

Post by pama »

sliken wrote: post #75
pama wrote: post #45

Earlier quoted context omitted.

Agreed. However also note that if it was only matrix multiplies and no full transformer training, the performance of that Meta cluster would be closer to 16k PFlops/s, still much faster than the El Capitain performance measured on linpack and multiplied by 4. Other companies presumably cabled 100k H100s together, but they dont yet publish training data for their LLMs. It is good to have competition, I just didnt expe…

I'd expect linpack to be much closer to a user research application than training LLMs. My understanding of LLMs is that it's more about throughput and has a very predictable communication patterns, not latency sensitive, and bandwidth intensive. Most parallel research, especially at this scale is more about different balance of operations to memory bandwidth, and much more worried about interconnect latency. I would…

All HPC and LLMs tend to get fully optimized to their hardware specs. When you train models with over 405B parameters and process about 2 million tokens per second calculating derivatives on all these parameters every few seconds, you do end up at the boundary of latency and bandwidth at all scales (from host to host, host to device, and the multiple rates within each device). Typical LLM training at these scales multiplexes three or more different types of parallelism to avoid keeping the devices idle and of course they have to also deal with redundancy and frequent failures of these erratic hardwares (if a single H100 fails once every five years, 100K of them would have more than two failures per hour.)

Re: El Capitan: New supercomputer is the fastest

#83

Post by handfuloflight »

colonCapitalDee wrote: post #55
pkaye wrote: post #27

Earlier quoted context omitted.

Are they always designing new nuclear bombs? Why the ongoing work to simulate?

Basically yes, we are always designing new nuclear bombs. This isn't done to increase yield, we've actually been moving towards lower yield nuclear bombs ever since the mid Cold War. In the 60s the US deployed the B41 bomb with a maximum yield of 25 megatons, making it the most powerful bomb ever deployed by the US. When the B41 was retired in the late 70s, the most powerful bomb in the US arsenal was the B53 with a…

How do you know all this?

Re: El Capitan: New supercomputer is the fastest

#84

Post by GemesAS »

olao99 wrote: post #7

I fail to understand how these nuclear bomb simulations require so much compute power. Are they trying to model every single atom? Is this a case where the physicists in charge get away with programming the most inefficient models possible and then the administration simply replies "oh I guess we'll need a bigger supercomputer"

Modern weapon codes couple computationally heavy physics like radiation & neutron transport, hydrodynamics, plasma, and chemical physics. While a 1-D or 2-D simulation might not be too heavy in compute often large ensembles of simulations are done for UQ or sensitivity analysis in design work.

Re: El Capitan: New supercomputer is the fastest

#85

Post by ethbr1 »

sliken wrote: post #81
olao99 wrote: post #7

I fail to understand how these nuclear bomb simulations require so much compute power. Are they trying to model every single atom? Is this a case where the physicists in charge get away with programming the most inefficient models possible and then the administration simply replies "oh I guess we'll need a bigger supercomputer"

Well there's a fair bit of chemistry related to the explosions to bring the sub-critical bits together. Time scales are in the nanosecond range. Then as the subcritical bits get closer obviously the nuclear effects start to dominate. Things like berrylium are used to reflect and intensive the chain reaction. All of that is basically just a starter for the fusion reaction. That often involved uranium, lithium deturide…

I'd never considered this, but do the high temperatures impose additional computational requirements on the chemical portions?

I'd assume computing atomic behavior at 0K is a lot simpler than at 800,000,000K, over the same time step. ;)

Re: El Capitan: New supercomputer is the fastest

#86

Post by ethbr1 »

colonCapitalDee wrote: post #55
pkaye wrote: post #27

Earlier quoted context omitted.

Are they always designing new nuclear bombs? Why the ongoing work to simulate?

Basically yes, we are always designing new nuclear bombs. This isn't done to increase yield, we've actually been moving towards lower yield nuclear bombs ever since the mid Cold War. In the 60s the US deployed the B41 bomb with a maximum yield of 25 megatons, making it the most powerful bomb ever deployed by the US. When the B41 was retired in the late 70s, the most powerful bomb in the US arsenal was the B53 with a…

Small addition: weapon precision has drastically increased since the days of the monster bombs

Less need of 9 megatons against a hardened silo if you have a 1.2 megaton weapon with a 120m CEP.

Re: El Capitan: New supercomputer is the fastest

#87

Post by ethbr1 »

buildbot wrote: post #69
leptons wrote: post #57

Earlier quoted context omitted.

I doubt the US Government is telling everyone about their fastest computer.

Unless it's like, air gapped powered by a naval nuclear reactor, I feel like someone would question why a random US gov building is drawing 20-30MW of power, and exhausting most of that as heat...

Aside from the utility, who would know? There's a lot of land out there, relative to the size of even a very large building.

Re: El Capitan: New supercomputer is the fastest

#88

Post by remram »

buildbot wrote: post #69
leptons wrote: post #57

Earlier quoted context omitted.

I doubt the US Government is telling everyone about their fastest computer.

Unless it's like, air gapped powered by a naval nuclear reactor, I feel like someone would question why a random US gov building is drawing 20-30MW of power, and exhausting most of that as heat...

Not disagreeing with you but how could any building draw power without radiating most of it as heat?

Re: El Capitan: New supercomputer is the fastest

#89

Post by johnisgood »

sliken wrote: post #77
teleforce wrote: post #71

Fun facts, FFT was discovered back in 1965 based on the urgent necessity of discovering and detecting illegal nuke testing activities, just two years after the Partial Test Ban Treaty (PTBT) was signed in 1963 [1]. The first sentence statement in the article mentioning that United States and other nuclear powers committed to the Comprehensive Nuclear-Test-Ban Treaty in 1965 is wrong since the treaty was only signed i…

Another fun fact, the priority for detecting nuclear testing led to seisometers all over the planet. So the detection the exact position and nature of any disturbance on the planet became radically better. This was quite the boon to anyone interested in earthquakes, not only can the earth quake be detected in 2D, but accurately in 3D. The number and accuracy is enough you can see where on each fault is, the thickness…

Cool factoids! If you have any more to share, please do so.

Re: El Capitan: New supercomputer is the fastest

#90

Post by wbl »

CapitalistCartr wrote: post #14
olao99 wrote: post #7

I fail to understand how these nuclear bomb simulations require so much compute power. Are they trying to model every single atom? Is this a case where the physicists in charge get away with programming the most inefficient models possible and then the administration simply replies "oh I guess we'll need a bigger supercomputer"

It's because of the way the weapons are designed, which requires a CNWDI clearance to know, so your curiosity is not likely to be sated.

And if they really want to know: https://www.energy.gov/nnsa/working-nnsa
Post reply on HN