Nice work. You sure sound like a computer guy to me. I sometimes wonder what the contribution to global warming will be from the exhaust of all these data centers. It is a strange dichotomy. On the one hand we are sequestering CO2 by the megaton to control AGW. On the other we are pumping hot air on a metastatical scale into the atmosphere to bring on the advent of Skynet. Make it make sense.
It’s the old catch 22, solving a crisis of that scale requires more processing power than the human brain can handle, it requires AI. But AI itself may contribute to the problem.
Great job, very technical information described in a relatively easy way to understand.
I am an electrical systems engineer with a background in nuclear power plant engineering and operations so I get all the power demands. UPS and big ass lead acid batteries were my life for over 30 years. I now see data center power demand is comparable.
I’ve always been interested in computer stuff so thanks!
The main problem is that AI algorithms are no different from what we did in the '60s to 'train' and generate AI algorithms. They are hugely inefficient. The Chinese now claim some better systems. We'll see.
As I said before it’s a paradox, they are massively inefficient, but they are what we have. The next solution may be too difficult to process without the help of these inefficient systems. I, like you, am taking the Chinese claims with a grain of salt.
The US states with Renewable Portfolio Standards face an interesting challenge. Virtually all generation growth will be required to be renewable, but data center load growth demands stable, reliable power supply. Data center developers don't trust renewables plus storage.
Would RPS states permit construction of non-grid fossil generation for data centers?
Technically OHIO is an RPS state, but it's a bit unfair as their standard is decidedly un-RPS. Ohio is allowing some data center serving gas generation construction
Is the Ohio RPS the result of legislation or PUCO regulation? The later is easier to revise than the former.
I have to believe that NARUC and NASUCA are nervous about the renewable driven rate increases, and no state has yet gotten serious about storage to displace fossil & nuclear backup.
I saw that, we think alike! They burn power being made and burn power being used! They also require lots of cooling on both processes, nobody ever thinks about process cooling.
technically it's not the software that fails, it's that the processor starts getting out of sync and missing process cycles, so it drops "bits". I'm dwelling into the edges of my knowledge here, but processors work on a cycle that process bytes out of the process buffer. This all counts on everything working on a systematic cycle. Heat slows the process and my cause some steps to fail to execute. Software, processed incorrectly, will fail.
Thank you. Is this tantamount to hot hardware leading to random thermal agitation of the 0/1 voltage switching that is the software...unwelcomed noise in the system.
It gets down to physics. The speed at which a transistor can switch is governed by the time it takes electrons to transit its channel. This time is governed by electron & hole mobility, which is a function of an electrons mean free path (the distance an electron / hole can travel before it hits something). The hotter the semiconductor becomes the more thermal agitation there is in the Crystal lattice and the shorter the mean free path becomes, as a result switching slows down. This changes a chips internal timing, which can reduce timing margin and result in erratic operation.
A different consideration is power management, a processor chips power dissipation is both static & dynamic. Dynamic power is that consumed toggling billions of points within the chip between 1’s and 0’s. It can be expressed as N x frequency x node capacitance x voltage change^2, where N is the number of state changes, frequency is the processors clock frequency, and voltage change is the difference between a node being at a 0 and a 1. The quest for higher performance has led to reductions in capacitance & voltage, with these reductions being compensated for by higher clock speeds to boost performance. However more demanding computations drive up N, which increases dynamic power, if not managed the processor’s execution may suffer or it may be physically damaged. So systems manage increased N, first by increasing cooling (make the fan run faster) possibly combined with dynamic voltage scaling. At some point those knobs run out of gas and the system will slow the processor clock to reduce dynamic power and recover timing margin. You will notice this as more fan noise and reduced execution speed.
Nice work. You sure sound like a computer guy to me. I sometimes wonder what the contribution to global warming will be from the exhaust of all these data centers. It is a strange dichotomy. On the one hand we are sequestering CO2 by the megaton to control AGW. On the other we are pumping hot air on a metastatical scale into the atmosphere to bring on the advent of Skynet. Make it make sense.
It’s the old catch 22, solving a crisis of that scale requires more processing power than the human brain can handle, it requires AI. But AI itself may contribute to the problem.
Great job, very technical information described in a relatively easy way to understand.
I am an electrical systems engineer with a background in nuclear power plant engineering and operations so I get all the power demands. UPS and big ass lead acid batteries were my life for over 30 years. I now see data center power demand is comparable.
I’ve always been interested in computer stuff so thanks!
Great work! Can't believe you called out those of us on gaming computers... ;)
Did you like that hot rod in the link?
I did, that's a really fancy one!
The main problem is that AI algorithms are no different from what we did in the '60s to 'train' and generate AI algorithms. They are hugely inefficient. The Chinese now claim some better systems. We'll see.
As I said before it’s a paradox, they are massively inefficient, but they are what we have. The next solution may be too difficult to process without the help of these inefficient systems. I, like you, am taking the Chinese claims with a grain of salt.
AI causes global warming 🤧
Those 1s and 0s are costly!
The US states with Renewable Portfolio Standards face an interesting challenge. Virtually all generation growth will be required to be renewable, but data center load growth demands stable, reliable power supply. Data center developers don't trust renewables plus storage.
Would RPS states permit construction of non-grid fossil generation for data centers?
"Enquiring minds want to know."
Technically OHIO is an RPS state, but it's a bit unfair as their standard is decidedly un-RPS. Ohio is allowing some data center serving gas generation construction
Is the Ohio RPS the result of legislation or PUCO regulation? The later is easier to revise than the former.
I have to believe that NARUC and NASUCA are nervous about the renewable driven rate increases, and no state has yet gotten serious about storage to displace fossil & nuclear backup.
Looks like it's a PUCO standard
https://puco.ohio.gov/utilities/electricity/resources/ohio-renewable-energy-portfolio-standard
Kilo, we are on a parallel path:
https://tucoschild.substack.com/p/the-monster-footprint-of-digital
I saw that, we think alike! They burn power being made and burn power being used! They also require lots of cooling on both processes, nobody ever thinks about process cooling.
Unless they plan ahead and put servers in Iceland. Cold and geo power.
Great vodka
I'm curious, what is it about hardware getting hot that causes the software to fail? ...and thanks for the article, we'll done.
technically it's not the software that fails, it's that the processor starts getting out of sync and missing process cycles, so it drops "bits". I'm dwelling into the edges of my knowledge here, but processors work on a cycle that process bytes out of the process buffer. This all counts on everything working on a systematic cycle. Heat slows the process and my cause some steps to fail to execute. Software, processed incorrectly, will fail.
Thank you. Is this tantamount to hot hardware leading to random thermal agitation of the 0/1 voltage switching that is the software...unwelcomed noise in the system.
It gets down to physics. The speed at which a transistor can switch is governed by the time it takes electrons to transit its channel. This time is governed by electron & hole mobility, which is a function of an electrons mean free path (the distance an electron / hole can travel before it hits something). The hotter the semiconductor becomes the more thermal agitation there is in the Crystal lattice and the shorter the mean free path becomes, as a result switching slows down. This changes a chips internal timing, which can reduce timing margin and result in erratic operation.
A different consideration is power management, a processor chips power dissipation is both static & dynamic. Dynamic power is that consumed toggling billions of points within the chip between 1’s and 0’s. It can be expressed as N x frequency x node capacitance x voltage change^2, where N is the number of state changes, frequency is the processors clock frequency, and voltage change is the difference between a node being at a 0 and a 1. The quest for higher performance has led to reductions in capacitance & voltage, with these reductions being compensated for by higher clock speeds to boost performance. However more demanding computations drive up N, which increases dynamic power, if not managed the processor’s execution may suffer or it may be physically damaged. So systems manage increased N, first by increasing cooling (make the fan run faster) possibly combined with dynamic voltage scaling. At some point those knobs run out of gas and the system will slow the processor clock to reduce dynamic power and recover timing margin. You will notice this as more fan noise and reduced execution speed.
Thanks Steve.