What country does Generative AI use as much power and water as?

Generative AI is the new bitcoin. Pundits used to compete for which country they could compare Bitcoin's power use with. Some nominated were: Greece, Ireland, * Netherlands, and New Zealand. I put a section on this in my Sustainable Computing course.

I tried asking Copilot:

"What country does Microsoft Copilot use as much power and water as?"

It replied:

"Microsoft Copilot, like all datacenters, consumes energy and water. Let's delve into the specifics:

1. **Power Usage Effectiveness (PUE)**: This metric gauges how efficiently a datacenter utilizes energy. The closer the PUE number is to "1," the more efficient the energy use. Microsoft tracks PUE globally and across operating geographies. While newer datacenters and those operating at peak utilization tend to be more efficient, the goal is to get as close to "1" as feasible¹.

2. **Water Usage Effectiveness (WUE)**: WUE measures the efficiency of water consumption for humidification and cooling in datacenters. It's calculated by dividing the liters of water used by the total annual power needed to operate datacenter IT equipment (measured in kWh)¹.

Now, let's address your question. Microsoft's water consumption for its cloud data centers increased by **34%** in 2022, totaling **6.4 million cubic meters** of water. Generative AI workloads are partly responsible for this surge³. However, it's essential to note that this data pertains to Microsoft's overall operations, not specifically to Microsoft Copilot.

In summary, Microsoft's datacenters, including those supporting Copilot, have a significant environmental impact. However, pinpointing a specific country for comparison would require more detailed data. 🌐💧🔌

Source: Conversation with Bing, 22/02/2024
(1) How Microsoft measures datacenter water and energy use to improve Azure .... https://azure.microsoft.com/en-us/blog/how-microsoft-measures-datacenter-water-and-energy-use-to-improve-azure-cloud-sustainability/.
(2) Microsoft's water consumption jumps 34 percent amid AI boom. https://www.datacenterdynamics.com/en/news/microsofts-water-consumption-jumps-34-percent-amid-ai-boom/.
(3) Microsoft 365 admin center Microsoft 365 Copilot usage - Microsoft 365 .... https://learn.microsoft.com/en-us/microsoft-365/admin/activity-reports/microsoft-365-copilot-usage?view=o365-worldwide.
(4) Measuring usage, adoption, and the impact of Microsoft Copilot. https://techcommunity.microsoft.com/t5/copilot-for-microsoft-365/measuring-usage-adoption-and-the-impact-of-microsoft-copilot/ev-p/4051091."

Cryptomining Consumes More Energy Than Gold Mining

Krause and Tolaymat (2018), estimate that "mining" of the four commonly used cryptocurrencies (Bitcoin, Ethereum, Litecoin and Monero), from 1 January 2016 to 30 June 2018, produced between 3 and 15 million tonnes of carbon dioxide emissions. They also estimate this is more energy intensive, in terms of value produced, than conventional mining of copper, gold, platinum or rare earth oxides.  Only production of aluminum was more energy demanding than cryptocurrency. One worrying trend is that as the cryptocurrencies drop in value the energy needed to mine them increases. The good news is there are alternatives, such as Ripple, which does not use mining.

ps: Last semester I asked my ICT Sustainability graduate students at the Australian National University: "Are Bitcoin and Blockchain Bad for the Environment?". 

Reference

Krause, M. J., & Tolaymat, T. (2018). Quantification of energy and carbon costs for mining cryptocurrencies. Nature Sustainability, 1. url

https://doi.org/10.1038/s41893-018-0152-7

ICT Sustainability and Bitcoin

The Australian National University is offering my course "ICT Sustainability" (COMP7310) in Semester 1, 2018. Enrollment are now open, so I made a short video to promote the course. This is a little different to the usual university course as it is entirely on-line (students can enroll form anywhere and do not have to be in a degree program), plus there are no examinations. I have added a small section on block-chain and bitcoin to the course, asking students to consider the energy use of this technology. The notes are available free and anyone is welcome to run their own version of the course. Athabasca University (Canada), run the course as Green ICT Strategies: COMP 635.

ps: The video was made using the Kdenlive free open source editor on a low power laptop running Linux. Stock video footage is from Unripe Content. The audio commentary was created with the Text-to-Speech YAKiToMe service. Images are from Wikimedia Commons.

Are Bitcoin and Blockchain Bad for the Environment?

Harald Vranken at the Open University of the Netherlands (2017) estimates that bitcoin's ‘proof-of-work’ algorithm is using up to 500 MW of energy. Unlike other computing protocols, which could be improved using a more efficient algorithm, or a faster processor, the inefficiency of Blockchain is an essential part of the protocol. A question I may to put to my ICT Sustainability students next semester at ANU is "Are Bitcoin and Blockchain bad for the environment?".

In the original paper proposing Bitcoin and Blockchain, Nakamoto (p. 1, 2008) wrote "... the longest chain not only serves as proof of the sequence of events witnessed, but proof that it came from the largest pool of CPU power...". Vranken (p. 3, 2017) traces the evolution of bitcoin "mining" computers, which started using general purpose CPUs in 2009, then GPUs in 2010,  FPGAs in 2011 and Application-Specific Integrated Circuits (ASICs) in 2013. Each change in technology brought about an improvement in energy efficiency, from CPUs with an efficiency of up to 0.1 Mh/J (million hashes per Joule of energy). Four years later the ASICs were up to ten-thousand times as efficient, at 10,000 Mh/J.

Vranken (p. 5, 2017) speculates about possible improvements in the energy efficiency of  bitcoin mining hardware, however, the Blockchain design has built into it a mechanism which increases the computation required as hardware (or software) becomes more efficient:

"To compensate for increasing hardware speed and varying interest in running nodes over time, the proof-of-work difficulty is determined by a moving average targeting an average number of blocks per hour. If they're generated too fast, the difficulty increases." (Nakamoto, p. 3, 2008).

This proof of work is used not only to reduce the possibility of fraud, but also to combat inflation of the digital currency (Nakamoto, p. 4, 2008). Those processing the transactions are rewarded with newly generated "coins". If this becomes too easy, then there would be runaway inflation.

Apart from the hardware, the major cost in bitcoin mining is the energy to run the equipment. As Vranken (p. 7, 2017) notes bitcoin's proof-of-work wastes energy, and there have been proposals to replace it with some useful task and alternative schemes to prevent fraud and inflation.However, the current proof-of-work scheme has proved remarkably effective. It would be interesting to conduct a more detailed analysis of how it compares with other more conventional financial systems, in terms of energy efficiency.

The Australian National University is offering my course "ICT Sustainability" (COMP7310) in Semester 1, 2018. I have added a reading on block-chain and bitcoin to the course, asking students to consider the energy use of this technology. The notes are available free and anyone is welcome to run their own version of the course. Athabasca University (Canada), run the course as Green ICT Strategies: COMP 635.

Reference

Nakamoto, Satoshi [Szabo, Nick?] (1 Nov 2008). Bitcoin: A Peer-to-Peer Electronic Cash System. URL https://bitcoin.org/bitcoin.pdf

Vranken, H. (2017). Sustainability of bitcoin and blockchains. Current Opinion in Environmental Sustainability, 28, 1-9. URL https://doi.org/10.1016/j.cosust.2017.04.011

ACS Blockchain Committee

The Australian Computer Society (ACS) has announced the members of their Blockchain Committee for 2017/2018. I am delighted to have been included. The committee is to look into the technical aspects and use of blockchain, including Smart Contracts.

ACS Blockchain Committee

Chair: Dr Vincent Gramoli, Head of the Concurrent Systems Research Group at the University of Sydney.

Vice-Chair: Dr Philippa Ryan, barrister and lecturer, Faculty of Law, University of Technology.

Members:

• Mr Ambarish Natu, IT Architect, Australian Taxation Office 
• Professor Ren Ping Liu, Head of Discipline, Network and Cyber Security, School of Electrical and Data Engineering, University of Technology Sydney. 
• Mr Scott Nelson, member of the National Committee of the IEC in Australia. 
• Mr Steven Pereira, Chief Information Officer, GS1 Australia 
• Mr Tom Worthington, Honorary Senior Lecturer, Research School of Computer Science, Australian National University.

Online Bitcoin Masters

The University of Nicosia is offering a Master of Science in Digital Currency. This is notable for not only being offered on-line, but also having the introductory course free. The university also accepts payment for later courses using bitcoin and uses the blockchain technology the students learn about to issue a digital certificate to graduates. This is an example of dogfooding: the university is demonstrating it is willing and able to use the technology, not just teach about it. The full cost of the Masters is €11,760 (about AUD$16,000). This is about a third less than I paid for a Masters (of Education) in Canada and quite a bit cheaper than at an Australian university degree.

Some Australian universities are offering credit towards a degree for completion of a low cost introductory on-line course. Four go further and offer credit for one quarter of a masters degree for completion of on-line courses. It will be interesting to see if these increase program enrollments and improve program completion rates. However, this may have some negative public education policy implications.

One of the advantages for students is that with a free open course you can have a look at the materials, so I filled in the enrollment form for Introduction to Digital Currencies (DFIN-511). This provided intimidate access to the program's Moodle website. The course has a conventional e-learning format, with twelve weekly topics, each with notes, a quiz and a live forum. The course web page is very plain, easy to read and uncluttered with excess images and formatting (just the way I like it). The live forums are at 5pm UTC, which is inconvenient for Canberra, being 2am, but the sessions are recorded. Also the time is not given in UTC for some forums (I am not sure what EEST, EDT, BST, or PDT are).

While I was not intending to undertake the course now (the latest cohort of students is up to week 8), I completed the pre-course survey. This asks about the background of the student and any experience with the topic of the course. It was implemented using Survey Monkey and at the end left me stuck at a Survery Monkey advertisement. I had to use the browser back button to find my way back to Moodle. It would be less confusing if Moodle own survey module was used.

The first video was a "talking head": the instructor sitting in a bare room just talking. The audio was clear, but an hour long video of one person talking is forty minutes too much. The notes for week 1 are a 2.4 MB PDF file (not too large), containing 50 slides. These are excellent slides, but I would have preferred a set of notes. The slides are not much use without an accompanying audio commentary (which I could not find).

The first quiz had twenty multiple choice questions. This took me six minutes and I scored 8.33 out of 10.00 (83%). This was without having read the notes, but with a reasonable level of background knowledge of the topic. The quiz is well implemented with the Moodle quiz module, providing the student with feedback for incorrect answers. This quiz would be enough to encourage me to study the material carefully, but not so hard as to be discouraging. The quiz allows another attempt, so I can study what I had missed and with the higher grade recorded from the attempts.

I could not find a description of the assessment scheme for the course. The Moodle grade-book showed eleven weekly quizzes, each out of ten, equally weighted for a course total. However, there is no indication of what the pass mark is. Australian universities usually have a grading system with 50% as a pass, whereas vocational education may have a higher requirement.