What happens if a supercomputer is used to mine bitcoins?

This has happened. I wrote about incidents of supercomputers being used to mine cryptocurrency a few years ago: Subverting computing research for fool's gold.

Bitcoin

NSF sponsored-researcher misused supercomputers to mine bitcoin via BBC June 2014:

A researcher has been suspended from a US national agency for misusing supercomputers to mine bitcoins at two universities. In a report, the National Science Foundation (NSF) revealed that more than $8,000 worth of bitcoins had been generated from NSF-funded computers... It is claimed that the unnamed researcher used remote access software and may have been attempting to cover his tracks.

So, he knew it wrong, tried to hide what he did including remotely accessing the supercomputer from Europe, and got caught.

@EllieAsksWhy @psnively $150,000 spent to make $8,000 of personal profit. What a waste.

— Emin Gün Sirer🔺⚔️ (@el33th4xor) June 14, 2014

The researcher’s access to all NSF-funded supercomputer resources was terminated, and after the NSF Office of Inspector General investigated (pp. 30-31), he was suspended from any further activities government-wide.

Dogecoin: Klondike operations

In February 2014, yet another researcher, at Harvard University, was caught mining a different cryptocurrency, Dogecoin. He used Harvard's Odyssey, a high-powered network cluster, which uses thousands of CPU cores, to mine Dogecoin. Harvard Crimson described the incident rather colorfully:

“Any participation in “Klondike" style digital mining operations or contests for profit requiring Harvard-owned assets to examine digital currency key strength and length are strictly prohibited for fairly obvious reasons,” 

Unusually high electric power consumption was what alerted Harvard to the situation. Since significant resources were wasted as part of the illicit mining activities, blowing through a lot of funds budgeted for research, the assistant dean for research computing wrote this notice: Harvard Dogecoin Mining on Pastebin. It has since been deleted. This was the URL https://pastebin.com/P8kvWqSG

All's well that ends well. Maybe? A few days after the "Klondike mining operation", several current students received the following email invitation from a mysterious HBS alumnus named John Harvard Bitcoin: 

“Are you the secret DOGE miner? I just started the Harvard Bitcoin Club and would like to invite you to join. The goal is to start an ‘open club’ that promotes Bitcoin education.”

This led to the establishment of periodic gatherings of the Bitcoin Club. Founder John Harvard Bitcoin paid for meeting materials and reimbursed the cost of attendee meals in Bitcoin. 

Supercomputers versus GPUs

Graphics processing units (GPU) are often integrated on a regular processor (chip). Sometimes, especially for gaming or applications like Second Life, a computer will have a "traditional" processor chip AND a stand-alone "dedicated" GPU, activated only as needed for graphically-intensive applications. GPUs are very fast for certain purposes.

Would a supercomputer need a lot less time to mine the same amount of bitcoin as a home PC or Apple computer?  The answer is no, probably not. Stan Hanks, former Enron Broadband Services CTO and founder of Network Mercenaries, told me why, in August 2014:

The bitcoin algorithm doesn't lend itself to parallel decomposition, and concurrent parallel execution on a slow processor just yields a bunch of slow results in parallel. The custom ASIC systems just flat kill everything else, and each generation completely obsoletes the prior generation to the point of it being completely ineffective to operate.

If you don't have a cozy relationship with a chip house doing miners, and aren't willing to do an in-house implementation, you're pretty much out of the game at this point.

So, the only benefit was stealing electricity. The bitcoin algorithm is ill-suited to benefit from what a supercomputer offers, i.e. a whole lot of conventional Intel type processors each working on different parts of a problem simultaneously. Instead, each processor in the supercomputer worked away on its own, slowly like any PC, without benefit from parallelism.

I do not know how subsequent generations of supercomputers would compare with ASIC miners. It has been a long time since I enthusiastically followed the semi-annual Top 500 announcements of fastest supercomputers. Unless there has been some sort of fundamental, alternative design development for supercomputers, there would be no change.

Centralized, non-distributed computing

Stan Hanks responded to my follow-up question about whether profitable Bitcoin mining is concentrated among a few big "mines" (or maybe mining pools) as follows:

I've been working with a lot of miners (because I'm sitting on top of a contract for 20MW of power at the lowest prices for electricity in North America). The only ones that are making money are the in-house operations from companies that fab chips, and one "renegade" outfit from Poland that hires chip designers to make their custom stuff that is not for sale commercially.

Note that, at the time, over 80% of Bitcoin mining was controlled by less than a dozen Hong Kong and Chinese miners. The mainland Chinese miners had official or informal controlling access to hydroelectric dams in rural far west China. They diverted huge amounts of electricity (generated by the dams and intended for consumer and commercial use) to ASIC miners.

The thousands of "mining machines" were onsite at these dams. They were watched over in rotating shifts by men who were either hired for such purposes, or much worse, were forced, instead of monitoring dam operations as they were supposed to do. 

China has since outlawed bitcoin cryptocurrency mining and use entirely.

Why worry about watching over ASIC miners? 

Because they get very hot and have a notorious reputation for catching fire. This is a real hazard as hundreds of them are often physically adjacent; fires due to overheating spread among them quickly.

Regarding my miner concentration concerns, Stan wrote:

In theory, that shifts at some point - the compensation algorithm becomes biased towards making money off of the clearing of transactions, versus mining new coins. Problem is, it takes a vast network of "miners" to make that work, all of which are operating at a loss...

Me:   The other problem with such a shift is that it undercuts one of the ideological bulwarks of Bitcoin: low fee transactions. What now, when Bitcoin starts doing the same thing...?!

Stan:  It's all in the original documents. Problem is, most of the original early adopters stopped reading when they got to the "and you can mine this valuable commodity FOR FREE using this code" and didn't read the whole thing.

The real long term value is in the blockchain, and the fee structure for clearing transactions. Oddly, at a significantly higher rate than market for VISA interchange. That, I think, was not well thought through at the outset. Given the time when this was created, the clearing rate looked progressive; the commodification of credit card clearing wasn't obvious to the author.

* Stan's description of his pioneering role in technology is remarkable! He was a member of the team that debugged the TCP/IP network protocol, co-wrote the code that opened the Arpanet to NSF-funded schools in 1983, and was on the NSF task force for commercialization of the internet!  In January 2024, Stan Hank's credibility was further confirmed to me with his response to a live interviewer's comment at the 2:40 (two minute and 40 second) mark; she said: 

This is, this is, this is it!  I can translate: Stan invented the Internet.