Thought Experiment: Could NCAR’s “Derecho” Climate Supercomputer Earn Its Keep as a Bitcoin Miner?

I wondered—since climate modeling isn’t earning us anything, and with indications that the enterprise may be downsized or even discontinued as priorities shift—I started a little thought experiment to find out whether the facility could earn some cash if it were converted to Bitcoin mining.

Now before any climate zealots panics: no, NCAR isn’t going to rip out their weather and climate codes and turn Derecho into a crypto farm. They are not going to fill the Wyoming facility with racks of blinking ASICs and hire a guy named “Blade” to wander around checking coolant loops. But the question is fun precisely because it exposes something most people—even many involved in climate modeling—simply don’t realize:

Supercomputers are terrible at Bitcoin mining.
Monumentally terrible.
Spectacularly terrible.

But the reasons they’re terrible are more interesting than the conclusion itself, and they say something about the way we allocate money to climate modeling versus real-world utility.

So let’s dig in.

---

Meet Derecho: 19.87 Petaflops of Coal-Powered Climate Prediction

The NCAR “Derecho” system, installed in 2023, is a coal-powered 19.87-petaflops HPE Cray EX supercomputer located in Cheyenne, Wyoming – because coal-powered electricity is cheaper there. Its primary mission:

“to support climate and weather modeling, atmospheric research, and Earth system simulation to improve understanding of future climate scenarios.”
—NCAR system description

https://www.cisl.ucar.edu/ncar-wyoming-supercomputing-center

This noble purpose comes at a noble price, too: roughly $35 million for the hardware alone, with ongoing multimillion-dollar annual operating and electrical costs.

Derecho is built out of:

• 2,488 compute nodes
• Two 64-core AMD EPYC Milan CPUs per node
• Totaling 323,712 CPU cores
• Plus a GPU partition, although NCAR rarely foregrounds its specs because GPUs are used for very specific workloads

Climate modelers love Derecho. They speak of it in hushed tones, as though it were an oracle capable of seeing 100 years into the future—never mind that it still can’t reliably predict whether your weekend barbecue will get rained out.

But for all that silicon and expense…
What does it earn?

Nothing.
Well, nothing financially.

And that’s what made the question irresistible.

The Core Problem: FLOPS ≠ Hashes

Climate models are measured in floating-point operations per second (FLOPS). Bitcoin mining is measured in hashes per second (specifically, SHA-256 hashes).

These two computational worlds might as well be different planets.

Climate models need:

• 64-bit floating-point math
• High-precision differential equation solvers
• Gigantic memory bandwidth
• Inter-node communication

Bitcoin mining needs:

• Simple 32-bit integer operations
• SHA-256 hashing performed millions of times per second
• Predictable, repetitive, dumb-as-a-brick computation
• Hardware that burns electricity as slowly as possible for a given hash rate

As one engineer famously put it:

“Supercomputers are built for thinking. ASIC miners are built for banging their heads against the same wall 10 trillion times per second.”

Derecho is a thinker, not a head-banger.
That’s the entire issue.

How Bad Could It Be? Let’s Quantify the Uselessness

A back-of-the-envelope comparison is enlightening.

Modern Bitcoin ASIC Miner (Antminer S21, for example):

• ~200 TH/s
• ~3500 W
• ~60 J/TH efficiency

Derecho CPU node (very rough estimate):

• Maybe 5–10 MH/s per CPU if you could even install mining software
• That’s 0.000005 TH/s
• At ~3,500 W per node, similar to an ASIC
• Efficiency: roughly 600 billion J/TH, give or take a few magnitudes

If that number looks absurd, it’s because it is.
By comparison, a modern ASIC is roughly 10 to 12 orders of magnitude more efficient than a CPU at hashing.

Total Derecho Mining Output (Speculative but Not Unrealistic)

Let’s generously assume:

• 5 MH/s per CPU
• Two CPUs per node → 10 MH/s
• 2,488 nodes → 24.8 GH/s total

Put differently:

Derecho would achieve approximately 0.024 TH/s.

That is not a typo.

A single $20 USB Bitcoin miner from 2013 did better.

Meanwhile, the Bitcoin network right now sits around:

≈ 600 EH/s (600,000,000 TH/s)

Derecho’s contribution:

0.000000004% of the network hashrate

That’s eight decimal places of irrelevance.

If Bitcoin mining were a pie, Derecho wouldn’t even be a crumb.
It would be the memory of a crumb. A nano-crumb.

So How Much Bitcoin Would Derecho Earn?

Let’s be absurdly optimistic.

The Bitcoin block reward is currently 3.125 BTC (post-2024 halving).

At 0.024 TH/s, mining solo would yield…

≈ 0 BTC in the lifetime of the universe.
(Again, not a joke.)

Pools, you say? Okay—using standard share contribution calculations:

Derecho would likely earn between $0.01 and $0.50 per YEAR, before electricity.

Yes, cents.

In electricity costs?

Derecho consumes something like:

• ≈ 6 megawatts

6 MW × 24 hours × 365 days × $0.07/kWh (Wyoming industrial rate) ≈ $3.7 million/year in electricity

So if NCAR converted to Bitcoin mining:

Annual revenue:
≈ $0.10

Annual costs:
≈ $3,700,000

Annual net profit:
≈ −$3,699,999.90

Even the U.S. government would raise an eyebrow at that business model.

But This Is Actually the Point

The amusing part isn’t that Derecho would be a catastrophic Bitcoin miner.

The meaningful part is this:

Derecho is optimized for models whose predictive skill is routinely oversold.

We’ve built a machine so specialized, so narrow in its computational purpose, that it can’t do almost anything else efficiently—not even a very simple, embarrassingly parallel task like hashing.

And that wouldn’t be troubling if the models it runs produced value commensurate with their cost. But after 35 years of supercomputer-enhanced climate prediction:

• ECS (climate sensitivity) still ranges from “mild inconvenience” to “biblical doom”
• Warming projections still overshoot observations by 2× to 3×
• Arctic sea ice predictions have been consistently wrong
• Extreme weather attribution is a political football more than a scientific certainty
• And temperatures stubbornly refuse to follow the tidy model curves displayed to policymakers

At what point do we admit the machines are producing a lot of heat, a lot of papers, and not a lot of clarity?

Derecho’s “Earned Value”: The Broader Question

If the climate models sitting atop these machines were yielding reliable, actionable, real-world forecasts, their costs could be justified without hesitation.

But instead what we have are:

• Large, expensive computers
• Running complex models
• That produce dramatically divergent predictions
• Which are then cited as evidence of “consensus”

There is an odd circularity here:

1. We build models to justify climate concern.
2. We fund supercomputers to run the models.
3. We point to the outputs as proof.
4. And when the predictions fail, we say we need bigger supercomputers.

Meanwhile, the machine itself earns nothing and produces no market-verifiable output.

Bitcoin mining—useful or not—has one virtue:
It is brutally honest. You either produce value or you don’t.

Climate modeling? Not so much.

The outputs are judged not on accuracy but on rhetorical utility.

What If Derecho Were Required to Earn Its Keep?

If you required climate supercomputers to justify their existence with demonstrable output—whether through market value, verified forecasting accuracy, or real-world predictive performance—some interesting changes might occur:

• Bad models would be abandoned.
• Parameter tuning would be scrutinized.
• Ensemble spreads would be interpreted more cautiously.
• Climate projection uncertainty would be communicated honestly.
• Politicians would stop treating 100-year projections as gospel.
• And supercomputers might be sized according to actual needs, not political symbolism.

The fact that Bitcoin mining is absurdly mismatched to supercomputing simply exposes the underlying issue:

Climate supercomputers are not built for utility.
They are built for narrative reinforcement.

And that’s what separates them from machines whose output is judged by the unforgiving reality of markets, physics, and math.

A Final Thought: What If We Reverse the Question?

Instead of asking:

“How much could Derecho earn if converted to Bitcoin mining?”

Maybe the more honest question is:

“How much real-world value is Derecho producing right now?”

Because if:

• its climate projections are unreliable,
• its model outputs are politically curated,
• its forecasts don’t match observations,
• and its missions increasingly serve advocacy rather than science…

…then perhaps its economic value isn’t much higher than its Bitcoin value.

And that should worry us far more than any thought experiment.