The Week in Green Software: AI’s Power Problem
MP3•Episode home
Manage episode 427081547 series 3336430
Content provided by Asim Hussain and Green Software Foundation. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by Asim Hussain and Green Software Foundation or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://player.fm/legal.
In this episode of Environment Variables, host Chris Adams is joined by Asim Hussain to dive into the complexities of AI's growing energy demands and its environmental impact. They discuss innovative approaches to sustainability, such as using fungi to manage building waste in data centers and the potential for greener materials and practices. The conversation also covers software optimizations to reduce AI's carbon footprint, emphasizing that energy inefficiency cannot be outsourced. They highlight the importance of integrated sustainable practices in tech development, particularly in the face of increasing AI power consumption projections.
Learn more about our people:
Find out more about the GSF:
News:
- Mushrooms eat building waste at Meta's Gallatin data center - DCD [03:35]
- Building with Mushrooms to Reduce Drywall Waste — or Cooking Up a New Future for Data Center Construction - Meta Sustainability [06:59]
- Solutions for AI's Energy Inefficiency Can't Be Outsourced - Bloomberg [15:42]
- BitNet: Scaling 1-bit Transformers for Large Language Models - Microsoft Research [16:34]
- Balance effects of AI with profits tax and green levy, says IMF | International Monetary Fund (IMF) | The Guardian [31:09]
- Fiscal Policy Can Help Broaden the Gains of AI to Humanity [31:24]
- [2406.09645] Carbon accounting in the Cloud: a methodology for allocating emissions across data center users [37:52]
Events:
- HotCarbon - Workshop on Sustainable Computer Systems - July 9, 2024 [48:33]
- IEEE CLOUD 2024 International Conference on Cloud Computing - July 7 to 13 [48:53]
- Masterclass: Become a Sustainable UX Designer - July 8 [49:44]
- The Software Measurement Landscape - Workshop 2 - July 9 [50:12]
Resources:
- Software Carbon Intensity (SCI) Specification [10:58]
- Episode 56 - Mushroom recycling with Joanne Rodriguez, Mycocycle - DCD
- Mushrooms eat building waste at Meta's Gallatin data center - DCD
- Options to make software greener without changing the code, and how to remember them | Chris Adams [30:17]
If you enjoyed this episode then please either:
- Follow, rate, and review on Apple Podcasts
- Follow and rate on Spotify
- Watch our videos on The Green Software Foundation YouTube Channel!
Connect with us on Twitter, Github and LinkedIn!
TRANSCRIPT BELOW:
Asim Hussain: And now there's a big hoorah in our space because like AI's now gone and
TRANSCRIPT BELOW:
Asim Hussain: And now there's a big hoorah in our space because like AI's now gone and
our automation has now gotten the point where even our jobs are now we're all really like nervous and upset and but you know this has been the pressure that we've been applying to the rest of the world with all the industry for decades and decades and it's just now coming to us and affecting us so you know we don't really have a leg to stand on I'd say.
Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.
I'm your host, Chris Adams.
Hello, and welcome to Environment Variables; This Week in Green Software, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Adams. Today, in our news roundup, we're diving into some of the pressing issues at the intersection of AI and sustainability.
So, with AI rapidly advancing, the energy demands of training and running these models are also seen to be skyrocketing. Posing significant challenges for the environment. We'll also be touching on some legislation for promoting sustainable business practices amongst AI companies and the potential for a green levy to drive investment for greener eco-friendly technologies.
We'll also be talking about some of the latest papers that have been published for people trying to understand and get grips on cloud carbon emissions. And finally, we'll touch on some of the exciting events in the green software community, including conferences, workshops, and masterclasses aimed at fostering sustainable development practices.
Joining me today for today's news roundup is my longtime friend, Asim Hussain of the Green Software Foundation. Asim, for people who've never listened to the podcast before, can I give you the floor to introduce yourself and some of your background?
Asim Hussain: Yeah, sure. So my name's Asim Hussain. I am the executive director of the Green Software Foundation. And yeah, I've been at the intersection of sustainability and software. I've been very lucky to be thinking about the same question about the intersection of sustainability and software for quite a few years now. And yeah, I've been well, mate. I've been, I've just come back from a vacation, which has been long time coming. I do have a little bit of a, cold, so that's what would explain the slightly nasally, annoying nasally sound the audience members are going to have to experience for this podcast.
Chris Adams: So this is with your hot beverage and Nurofen chaser
Asim Hussain: hot beverage. Yeah, that's what I like to do. That's how I like to start every podcast episode is a coffee and a Nurofen, every conversation with Chris Adams has to have a, both a coffee and a Nurofen.
Chris Adams: Wow, that's, one thing to take away with me. All right, folks, I should just briefly introduce myself before we dive in. I am the executive director of the Green Web Foundation. It's a small Dutch nonprofit based, well, in the Netherlands, where we are working towards an entirely fossil-free internet by 2030.
I also am one of the hosts of this podcast here. As well as one of the organizers of ClimateAction.tech, which is an online community, where actually me and Asim first met online before he basically headed off to set up the GSF in its current state. Alright, Asim, are you sitting comfortably?
Asim Hussain: I'm standing uncomfortably.
Chris Adams: uncomfortably at your swanky standing desk.
That's good enough for me. Should we start with news then?
Asim Hussain: Yeah, let's go for
Chris Adams: Alright, okay, so Asim, I was thinking of you when I saw this paper, this story. The first one is a story about mushrooms eating building waste in data centers. So this is a link to the Data Center Dynamics website talking about specifically the use of these, of essentially building-waste-eating mushrooms at the Meta data center and other ones.
And the general spiel of this is that there are now a number of companies which are, essentially deploying fungi, various kinds of fungi, to deal with all the building waste that ends up being created when you might kind of read, when you're essentially demolishing a building or creating a new one.
And, it essentially takes all this waste, and the fungi are able to Essentially, deal with the toxins, and then create something like, kind of, fungi-style bricks that can then be used as a kind of circular building material going forward. And Asim, given that you're our kind of resident mushroom fan, I wanted to just like, see what you thought about this, or if you had any particular immediate like, hot takes or things when you saw this one?
Asim Hussain: No, that's all. it's a great application. In fact, it's not an uncommon application of, you know, what people are applying, you know, fungi in this technology for. It's actually one of the, one of the, one of the, one of the very exciting kind of broader sustainability solutions in this space that there is.
I mean, there's a couple of different types of fungi. You're going to have to pause me at some point. There's a couple of different types of fungi, but there's one particular type was kind of saprophytic, which is effectively what Fungi, the purpose it has in kind this life that we lead is it basically, it's the thing that destroys things that have died. And if it didn't exist, then we'd be basically living on top of this massive mountain of logs that aren't decomposing. So that's one of the things that they're really, really good at. And it's been, it's, there's been a lot in many different spaces, been a lot of active research, a lot of startups, a lot of organizations exploring how to use Fungi to decompose things that they don't normally decompose. And it's actually quite an interesting technique. Because even people, what's so fascinating about the the fungi space is it's, driven in large part by citizen scientists, which is one of the things I love about it. And there's a lot of citizen scientists out there who are doing things like trying to find a strain of fungi, which can decompose certain types of plastics and you would literally do this. You would literally grab a selection of these plastics. You put them in a, you know, you can go online. It's as simple as this. You put it in a blender, you blend up a plastic in your home blender. And then you just have like, as you can probably see behind me, I have lots of jars out there, which have like different strains of fungi, and you would just put that with other material in the jars.
And then you'd collect lots of different strains. Like. Every single kind of mycelium is like of a different strain of it of the same one. So you can have like millions and millions of different types of strains. You go into the forest and you see a type of mushroom you've never seen before. And you're like, ah, maybe that will absorb this plastic. And so there's a lot of interest in this trying to like find those strains of fungi that can kind of absorb and transform, you know, different materials. Obviously certain fungi only in the forest, they only work on certain types of trees. They have like a relationship with them, but you can actually find strains of fungi, which do different things. And kind of the interesting thing about it is about turning them into, bricks like that as well. There's actually organizations out there trying to replace packing material for boxes. But what you do is you basically, you create like, so you take some time. It's not like a foam that you stick in and like 30 seconds later, there's a thing you basically have to Have the thing you want to pack in a box Inside the box you put like a substrate which can be the thing that mycelium grows on you then like almost like impress inside this substrate the shape of
Chris Adams: like the
Asim Hussain: book or something the mold and then you inject it with the mycelium and you put it in basically an oven for like a month. And it comes back out and then you basically spray it off and like the actual mycelium has grown into the shape of the thing. And then you've got something which you can put in the packing crate. And then at the end of the day, you just, it's a mycelium. You just break up and put it into your garden and you, and it decomposes.
There's lots of like wonderful stuff. There's a, great guy called Paul Stamets, who's quite a character. But he's done a lot of, he's done some great TED talks in the use of like fungi and mycelium in, kind of waste and getting rid of waste even getting rid of oil and there's a lot of kind of very active stuff in this space.
Chris Adams: Can I stop you one second there, mate? Cause, cause there's, so you just said Stamets, right? So Stamets hit, so in real life, there's someone called Stamets and in the whole, and that's, so the whole, when they had this whole Star Trek discovery thing, where there was this guy called Stamets who was using like the mycelial internet space thing, that's a direct reference to this dude, basically.
Asim Hussain: That's a direct reference to Stamets, which is kind of, which joins two of my biggest nerd bubbles together in the most beautiful way, but yeah, there's the engineer, and in Star Trek Discovery, how they, instantly, there's a new type of drive, and they instantly can, move to one part of the universe to another, and it's called a spore drive, and you need to kind of enter this kind of psychedelic realm to connect transmission, so yeah.
Chris Adams: My word, Asim, I was not expecting us to dive down that myco rabbit hole for that, but that was a lot of fun. Thank you very much. So you basically said. By doing this, so in this case of packaging, this basically removes the need for like, say, fossil-based expanded polystyrene in packaging, and in the case of materials here for buildings, you would use that instead of having to get a bunch of virgin materials, for example.
This would be like a circular, that's the approach that they'd be using here, right?
Asim Hussain: So what, so the specific approach they're using in this particular article is, I believe it's more about, it looks like they're using, basically trying to get rid of the drywall that they have inside the data centers. So I don't know if it's a particularly from a decommissioning a data perspective or renovating a data center perspective, but they're ending up with a lot of material, which typically you would just dump in a waste fill, but now they've basically got a form of mycelium, which can eat drywall and generate something that's, decomposable, maybe edible.
Yeah
Chris Adams: All right, cool. And so we, this is, mentioning, referencing the metadata center in, I think, Tennessee. But we've also seen Microsoft, as far as I'm aware, Microsoft has also been a bit of, it's been dipping their toes into this field as well. And one of the reasons why you might care about this is that, well, last year, Microsoft's reported emissions, when they released their sustainability report, it was like, up 30 percent and a significant chunk of that came from buildout of data centers.
So we are now starting to think a lot more about the embodied energy in the facilities that are created so that we can actually have data centers, so we can actually use compute, a lot of the kind of compute power available to us, or even some of the AI power, or the kind of sources of AI and stuff like that, because you need, they need to be in a building somewhere to get this stuff built.
And like, this kind of made me wonder actually, Asim, surely I imagine some of this might show up in an SCI score, a Software Carbon Intensity score, if you're purchasing cloud from a certain place. If you've seen a massive buildup of data centers, surely that might have an impact on the embodied carbon for the compute you might be using, right?
Asim Hussain: It could do, it depends on what your, cause in the SCI, there's kind of two components. One is the software boundary, which is "what are you going to include and not include in that score?" I remember us having quite a few conversations in the early days of the SCIs, which "should you include the, like the concrete that was used to pour the floor of the data center?" And I'm not too sure we really, I don't, think anything particularly made its way into the specification, but it has to be, if you're measuring something, it has to be something which drives a choice or a behavior. So, you know, I suppose what I'm going to with this is, if there was a data center, which was particularly built like a zero carbon, maybe built with mycelium or something.
I don't know, but like, if there was a data center, it was particularly built with that choice, then maybe it is something you want to include in the score, because then that can drive an action of choosing one option over another. But if every single data center is effectively built exactly the same way, the discussions we were having was, well, that's just overhead of adding a sec, effectively a coefficient, which wouldn't really drive a decision-making factor. So, I suppose as what I'm going to this is excitingly, if there are data centers that are being built, which are going to have
vastly different embodied carbon profiles, and then you, and then if that was included in an SCI score... and I think as we move forward with SCI, because one of the things that's happening... SCI on its most, we, launched kind of the version one and now 1,1 of the SCI and it's, and it was, it's very bare, very basic. It's designed to be built upon. And so now what the teams are having conversations around is like, if you were to apply SCI to AI, specifically, what does that include? And one of those questions
they want to answer is what,
Chris Adams: building,
Asim Hussain: yeah, like what you include so, so when someone reports an AI score, we can actually start getting to like some apples to apples comparisons. So that's, a conversation that, you know, it's interesting. Maybe we should bring that, bring it up again is do you include the embodied of the data center? But then you also get into the headache of "my God, it's hard enough trying to figure out what the embodied of a chip is."
Chris Adams: Yeah.
Asim Hussain: And now going to ask to figure out like what the embodied, so there's has to be some practicality aspect to this as well.
You know, you know, we have to, there should at least be some models. I don't know.
Chris Adams: So there are models that, as I understand it, there are models for working out the, essentially like the carbon emissions for a kilo of concrete, for example, or stuff like that. Some of these exist. And,
Asim Hussain: exists from a
LCA,
Chris Adams: and there are companies, and we know there's like, in Europe at least, I know there's a company called LeafCloud.
That are explicitly using, are doing, making reused, or, they're reusing heat, but also a very specific kind of data center, which isn't like a very large outta town thing. They have like essentially shipping containers put into places like say greenhouses, where the heat is being reused, for example, and where they're not having to build a whole bunch new buildings. There's also, I think in Switzerland, there's one company, because we maintain, where I work, we maintain a directory of green data centers. And one of them is a, they basically reused a old factory building with a waterwheel that used to be kind of like a clothing factory, and now it's a data center.
So they've essentially reused the whole building shell. They haven't built a load of stuff as a result. So this is one place where this might show up, but in order to do this, you need to have access to the numbers for this. And that's still a bit of a challenge because, yeah, we don't have the, we don't have easy access to these numbers, and like you do say, it's a challenge just thinking about chips, let alone expanding the boundary to the actual buildings instead.
Asim Hussain: Yeah. I mean, at least you have some information when you're running software, like what, you know, you can, now that we've done, a lot of that workings out so you can figure out, you know, perhaps it's this chip, but I think given the secrecy around data centers, I don't know, I think there's going to be,
I don't know.
Chris Adams: It's gonna be a challenge, because while we have this practice of, essentially, water usage and electricity use, so many things being under NDA, it'll be very hard to come up with some numbers without using, like, a basic number. Okay, alright. We have totally gone past talking about mushrooms and data centers into all these other things, but I guess this is part of the whole thing about sustainability and technology.
It's all interconnected.
Asim Hussain: It's all interconnected.
Chris Adams: Shall we go to the next story?
Asim Hussain: Yeah. Let's go for it.
Chris Adams: So this is a piece from Bloomberg, actually, so this is the topic of this is Solutions for AI's Energy Inefficiency Can't Be Outsourced, and this is an opinion column from Bloomberg talking about this projected demand some of the figures which are pretty, pretty impressive, they basically say, in the US at least, it's, there's a projection saying that AI, the growing demands of energy are like, is projected to make up around 8% of the US' power consumption, electricity consumption up from 3% in 2022.
Now these numbers seem a little bit high and they are citing a kind of this arms race of different kinds of organizations, essentially building out these massive data centers but also buying loads and loads of chips, but it does talk about some of the approaches that we're seeing now to kind of rein in some of this growth.
So one of the things was this idea of one-bit architecture, which is essentially, I'm not going to pretend to understand it. And I'm not sure if you are similarly informed in this one, but
Asim Hussain: I'm going to definitely pretend to understand it.
Chris Adams: In that case, I'll hand over for you to confidently bluff it around, just like a ChatGPT would actually, Asim, the floor is yours.
Asim Hussain: I'm going to, I'm asking GPT. No, I'm guessing, and this is, I haven't really, I've, seen it, but because it's one bit and a bit can only be one or zero, I'm guessing what this is, that, you know, instead of like pumping in a number between one and 256 as one of the inputs to a node in a model. Maybe you could just try one or zero. And then output one or zero and then see if that actually still gives you some pretty reasonable results and from what it looks like it might do and you know for those of us I kind of it's like there are extreme inefficiencies you can do when you're working at the bit level in terms of programming and computation and instructions on the chip and things like that because it's so much lower level than the architecture.
I presume that's what it is, which actually to me is. It's really exciting from a, from the level of, this is a software architectural solution, which is effectively, I think what we've been advocating for, a large part of the time, which is, you know, we do need to I hate to use the word code because I don't want people to dive down the, you know, building more. I, well, AI is one of the, one of the few areas I would say where actual code efficiency is extremely important. But yeah, this is kind of, it's interesting. Now that there has been a pressure applied to optimize, the optimization has happened.
Chris Adams: Yeah, that's, so basically, Asim, I think you're about right. Now, when I, remember when I skimmed over this paper before, one of the key ideas was, the one bit approach was, essentially, when you, would be able to, you would use this to encode the difference between different parts of, like a dataset rather than showing absolute numbers.
And one of the things that this allowed you to do was allowed you to just use addition rather than multiplication in some cases. Now, I'm not an AI specialist and I'm not a hardware specialist, but the general idea was by representing things in a more somewhat simple fashion here, you avoided having to make some of the expensive calculations that you would otherwise need to do.
And this basically reduced the energy that you might need to run some of these calculations. So this was like one example. I was quite impressed to see this inside Blumberg because it was a very quite new research, but also really, technical and actually quite promising. So yeah.
Asim Hussain: It's, it, the title is interesting though, isn't it? Cause it's not like, it's not like software making, it's kind of talking about energy inefficiency can't be outsourced.
Chris Adams: Yeah,
Asim Hussain: I just thought it was an interesting, it's like no one really knows or cares or thinks about the software side of the, this whole equation.
To me, this is just like a software optimization. So you would just say like software can be optimized to reduce any AI's energy footprint. It's not, not expressed in that way. It's interesting. Kind of expressed
Chris Adams: there's a couple of things that I think are also really interesting about this piece, in my view, was that it talks about the kind of economics around some of this, and basically the idea of outsourcing this is essentially how we have a bit of a tendency in the technology industry to say, "well, we're just going to have, like, We realize that data centers use loads and loads of power, so what we're going to have to do is just somehow get loads more power."
And so you basically have people talking about, oh, obviously the solution is to deploy loads and loads of nuclear, for example, right? Never mind that these take between minimum 10 years to get built, right? So, what are we going to do in the meantime? A lot of the time it's likely to be coming from things like gas, if you're going to be using something like that.
So that's an issue there. But it's also worth thinking a little bit about these figures that were mentioned in this story. We've seen numbers like 8 percent of the USA's energy consumption by 2030. It's worth bearing in mind that these numbers are often coming from the utility providers in various states, all right?
So like, say, in, say, Virginia, I think it might be, I forget the name of the actual monopoly provider, but there's only one provider over there. And basically I think that's Pacific General. I think that's actually on the, other coast, basically.
Asim Hussain: We've played this game before where you can name all
Chris Adams: was Caiso last time, which was, which is, that's the Independent Systems Operator, which is not the energy company.
The energy companies are somewhat different because it's a, because in many, because they have very specific, if you're in a state where you've got a single provider, they are allowed The only, the reason you only have a single provider is that you have basically had that state agree to have a, what's referred to as a natural monopoly.
So, they basically, the agreement is, we will give you a guaranteed 10 percent net profit plus for your organization, alright? But you need to basically, yeah, as long as you agree to share your plans, For the new infrastructure you're going to build over the next few years, but also you need to justify this in each of these cases.
And when you think about this, if you're going to get a 10 percent net profit from that for any of the energy you, get. Now, what you, if you want to increase your profits, what you need to do is you need to say, "well, I have loads more demand coming. I need to like double my expected demand to double my profits inside this."
Asim Hussain: Oh, I see.
Chris Adams: is one of the things, because this is a lot of the existing providers, they're used to saying, "well, we've got all this extra demand. What we need to do, we need to build a bunch of new gas, fired power stations. And 'cause we know we're gonna make a 10% guaranteed profit on all the infrastructure we, build.
That's basically, you know, we are incentivized to say it's gonna be really, high" basically. So. You, it's, really worth looking at a paper by one, one gentleman, John Kumi, who's actually, who's spoken a lot about this, because 20 years ago, we had the similar thing when you had people in the coal industry saying, "well, coal was what powers the internet, so you need to have more coal fired power plants if you want more internet."
We have a very similar thing happening. In this case as well actually. So it's worth bearing in mind that yes, we do see these kind of apocalyptic forecasts for energy, but you also see that when you do have constraints on this because it's so difficult to build, then we do end up with a renewed interest in energy efficiency.
And even at the kind of like energy level, right, there are different ways that you can basically meet demand. You can meet demand by adding new supply, but you can also meet demand by investing in energy efficiency. And that's, and this is very much what it looks like, so a lot of the ideas you might see at the energy sector, I think, are at least applicable, or at least relevant in what we talk about with cloud, because essentially you're looking at a kind of commodity that you pay for on an hourly basis, or something like that.
Asim Hussain: Well, that's kind of one of the... All I see is there's, a significant amount. There's not a significant, there's a fixed amount of investment and focus that organizations can put into something. And if you present them with an option, either put all this engineering effort to make something more efficient, which costs 10 or buy renewable energy, which costs five and then, well, I'll choose the five one. So I think that's kind of, that's. That's why kind of investment goes in one way or the other. Whereas I suppose what's happening now is that energy is now, we're reaching the point where energy is, and I'm just throwing out numbers here for energy is now costing 12, but the development still costs 10.
So like, well, maybe we'll put some money into development
Chris Adams: Yeah.
Asim Hussain: So that's kind of, and that's interesting that's why you want things like we were talking about levies and money last time. That's kind of why you want to change that balance of it a little bit. And that's also significantly why in the SCI specification, the decision was made not to include any energy offsets or anything like that.
Just because if you gave somebody the option of spending 1 instead of spending 10, they would spend 1. And we want people to spend the 10 to actually make things more
Chris Adams: Yeah, to address the consumption issue, rather than just think about the intensity.
Asim Hussain: Exactly, yeah. But this is exa I'm really excited. And the other thing I was thinking about as you were talking was I was just remembering about my time at Intel. I think this maybe links a little bit to the Nvidia's statement as well. So, I think I might have, I always love telling people this story because I just think it's such a cool word and it's such, it tickles my sci-fi bone so much. But there's a statement they used a lot, which was dark silicon. Have I told you this? I
Chris Adams: No, you haven't.
Asim Hussain: people. There's a dark silicon. And I was like, "Oh, that sounds good. What's the, what's dark silicon?" And what dark silicon is when they're kind of looking at a chip and they put load on it.
And the key thing with a chip is how much can it expel heat and still function at that level. So looking at heat on a chip. And so when they're running a certain software on a chip, they'll put like a, what you call the heat detecting camera on. And you know how they look like this. It's very red, it's very red on the bits that are hot.
And it kind of looks
Chris Adams: Ah, okay, yeah, look at a thermal house,
Asim Hussain: Thermal imaging. yeah. Yeah. Even though the black might not be like ice
cold, quite hot, but like relatively it's cold. And so the things that they would be really like thinking through is like, how come this software, how come half the chip is black?
Like, why aren't you using the rest of the chip? Like you've maxed out the chip, but half the chip is black. And really what it kind of, you know, what it goes back down to is that, you know, we called it, I think I might call it the silicon gap, which is the gap between what engineers are building and what silicon manufacturers are enabling on their chips. And there's this disconnect between, you know, they're, all building, "why aren't you using this, these more advanced chip sets that are more efficient? Why are you using this stuff on this side of the chip?" And so I think that's something that we need to get down and tighten that gap to use this infrastructure more efficiently, I think over the years, from a developer's perspective, it's always been about time to market.
How do we beat our competition? It's never around, how do we use this chip more efficiently? And so I think that one bit architecture is, it sounds like an example of that. It sounds like an example of, we want to leverage the instruction set on this chip to be as efficient as possible. We need to change. Fundamentally how we're architecting and even thinking through algorithmically this code to take advantage of that. And that's, I think it's also like this, other area, which is completely, we're just ignoring, you know, there is this dark silicon and honestly, the silicon manufacturers are like, why are developers not, I don't understand, we put so much energy and time into like
building
Chris Adams: we're only using a of it, right?
Asim Hussain: a The percentage of it, and there's that also, and I'm going to, I'm rambling on for a second, but just one more, one more point. I thought it was really interesting. One of our organization's Entity Data, they did a really great report. It was kind of two years ago now, I think. I don't think we really circulated.
I don't think we circulated that well. And it was, they just, they just looked at Java, you know, Java, like still, there's still a lot of very antiquated Java applications running out there in the world. And they just said, what is the energy difference if we just upgraded, not the code, but the JVM, the underlying JVM.
And that's all they did was I think they upgraded, I cannot remember. I'll find a link to the article and the paper. It was like several steps up. But they were like, "look, most apps are still running on whatever the JVM was they were built with like 10 years ago. And it was a seven, it was a 60 or 70% energy efficiency improvement.
It was unbelievable. The energy efficiency improvement just from grading the JVM. And that was
Chris Adams: Ah, okay.
Asim Hussain: That, and if you think about what that means, what happens was the chips evolve to have different instruction sets. The JVMs, only the modern JVMs are built to the new one. And so if you're running on the old ones, it's just using the old instructions.
So you're not really leveraging the infrastructure the same way, which is why like recompiling software with like, you know, the latest version of the compiler against the latest version of the chip. It's really important. And it again, that was Intel's, when I was at the time, that was their big push.
They were like, "use the latest bits, use stuff that's compiled now using the latest optimizations." 'Cause they saw a lot of people were still just kind of compiling, leaving that binary, letting it run for like four or five years. And then, that's it. And yeah, I'm going to stop ranting now.
Chris Adams: No, that's actually, I didn't realize, I was somewhat aware of things like the JVM, there's like hotspot or different kinds of flavors of the Java Virtual Machine to run this code, and it's somewhat similar to like in PHP land, like when a new version of PHP came out, because it made much better use of the underlying code, the underlying hardware, you saw a massive increase in performance.
And like, you kind of see something a bit like that with Python as well, with the whole global interpreter lock. Like, I can have a piece of Python that'll be running, and it won't be able to use all the other cores in my machine, in my computer, right? So, rather than lighting up the rest of the silicon, it's got just, it's, most of my computer is dark, basically, in that same kind of approach.
All right, yeah. Cool, alright, so that's like one of the approaches that we have, and this is one thing that you could plausibly do. I've shared a link to a blog post that I've, I was trying to explore this to find a way to explain it, to basically explain the fact that you can reduce the emissions associated with code without actually changing the code, by thinking about what options you might have in terms of, like you said here, like you change the VM or something like that, or change when you run it, or anything like this.
And I'll share a link to that, because I've kind of framed it in terms of If, there are three, three things you can change, basically. You can change the time of running something, which is kind of speaks to carbon awareness. You can change the speed, the amount of compute, computation you might be using, the number of cores you might be using something, or you might change the place, like where in the world you choose to run this for the carbon intensity of the underlying energy.
So I'll add that to the show notes because it might be another nice helpful addition to this. Alright, okay, that was quite, that was fun. Shall we look at the next story?
Asim Hussain: Go on. Yeah.
Chris Adams: Okay, the next story is, this is actually from The Guardian, and this is talking about the balancing some of the incentives of the kind of profits that are projected to come from deploying AI with something like a green levy on these profits, basically.
And this is actually, came from that, Left leaning organization, the International Monetary Fund,
Asim Hussain: Oh, wonderful. Great. Great to see them in this space.
Chris Adams: And basically what they are, the argument from the IMF is basically saying, well, you've got all these very profitable AI firms, and they, and we know there's both the social and environmental impact that's taking place here. So what you should actually have is some kind of green AI win for tax, essentially, that will be used to fund some of the, sustainability initiatives.
And like, to be honest, I have a lot of sympathy for this because what we've seen from the largest providers in the last like year is that given the choice between investing in efficiency or investing in more capacity and building loads more data centers, we've seen all the big providers go for building more capacity and like emphasize profits rather than the environmental impact here.
So it looks like we're not the only people thinking about this. The IMF is thinking about this as well. And they're saying, about this. You need something a levee to do this. Do you have any kind of particular thoughts on this? First of all, Asim, because I was really surprised to see this come up from the IMF of all organizations.
Asim Hussain: Yes, I thought it was a very important point. I love the fact because the use of the word levy, and I remember us having the conversation last podcast about like use of the term levy rather than
tax instead of, explaining it, but then in the article actually uses the word tax, like all over the place. Yeah, I think it's really important. Like one of the things I'd say is that like, why is everybody so excited about AI in the first place? In my most cynical moments as as a software engineer, I would say, Our purpose in life is to either find solutions that help people waste more of their time or get rid of jobs and automation.
If you think about kind of why we have been like one of the most highly paid sectors for quite a significant amount of time, it's because building automations, yes, you could argue and helps you deliver kind of projects faster, but it also helps you to do more with fewer employees. It decreases the.
You know, the earnings potential, the, a lot of this stuff from that perspective, and now there's a big hoorah in our space. Cause like AI is now gone and
our automation has now gone the point where even our jobs are now we're all really like nervous and upset. And, but, you know, this has been the pressure that we've been applying to the rest of the world with all the industry for decades and decades.
And it's just now coming, to us and affecting us. So, you know, we, don't really have a leg to stand on, I'd say. We also. There's nothing we can do. It's happening is the only other thing I'd say. There's no, you know, we have to just accept it and move forward. But yeah, I really liked the idea of that because like, let me put it another way.
Like there's this huge kind of bro down AI tech bro showdown with like, well, I think Sam Altman posted something a year ago now, which deeply disturbed me. And he said he wouldn't be surprised, I think within the next couple of years, if there's a one person unicorn startup, which is a single, like, billion dollar organization run by one human being. And I was thinking to myself, like, that might actually be true. And I would, you know, I do, there is a chance that would be true. might happen in the future, but how do I feel about that? What is the human impact to that? I mean, what is the green impact to that? So I'm now going beyond green because I think that there's like this AI is going to make a few people and organizations immense amounts of power and wealth. How do we have ways to, redistribute all of that and to kind of add a level of fairness to, the rest of society? Is it okay? And so from a green level, absolutely. But I'd also argue from a societal level as well, like, like, you know, like when we talked, you know, how about this? When we spoke about the green transition, it was impossible for us to have like a proper conversation about the green transition without having a real conversation and talking about how we're going to transition the people who are employed in the fossil fuel industry over to other areas. I don't see having, I don't see us having that conversation here as well.
Like it's just ignored. And so I think that's something that we need to have is like, is if you're, if you want to have. The opportunity to get this much power and money and wealth, I think it should come with a certain amount of social responsibility to, you know, be a green levy in terms of the green ones, but I think it actually should be broader than that. It should be, you know,
Chris Adams: So address some of the. Some of the inevitable costs that might be incurred upon society to provide to, like, ease that transition. Okay,
Asim Hussain: yeah.
Chris Adams: I was not expecting you to go there, Asim, but
Asim Hussain: Well, I feeling it.
Chris Adams: Bang yeah, that's usually me kind of jumping up and down, actually. All right, okay, that
Asim Hussain: How do you feel about it?
Chris Adams: I think, so you said this idea, like, I feel uncomfortable about a billion dollar startup with a single person.
And I, okay, how many, we're not that far from it, I think, because if, actually, no, we've been, so if you think, how many people work for WhatsApp?
Asim Hussain: WhatsApp! I was WhatsApp, I think it 11
Chris Adams: was purchased for $12.6 billion by Facebook a few years back, right? So that's not that far away on a kind of per person basis, but that's not a single person.
But you've got to realize that like, you know, if that was probably a lot better for the people who own shares in that than the people who are working for this. And we have seen multiple cases. We've seen cases like when a company has to choose between keeping on staff to work on something and getting rid of them, and then spending multiples of the staff's wages on buying their own shares to kind of in increase the cost, increase the share price. We've seen the decision that people have been taking, and there is the, I think that this is a thing that needs to be addressed. 'Cause the current, if we're going to assume that, if we're gonna accept that digital is gonna be the, this thing which is just as important to our lives as access to water or you to or energy or anything like that then you probably want to have a discussion about okay well how are the dividends shared well how is the upside shared in an equitable fashion so that we don't end up with people rushing outside with guillotines or in the very least right like it's not good for social cohesion basically so that's my the view that i might actually have on some of this
Asim Hussain: No, that's a, really, I think that's something that people, I've been, I don't know, should I say this?
Chris Adams: I'm going to stop you because it's coming up to 40 we've got one story, so we can, talk about societal accounting, all to carbon accounting, this new paper, which I think is really interesting and we both were nerding out about it before this call. So this is a new paper from Google, Carbon Accounting in the Cloud, a methodology for allocating emissions across data center users.
So, We can totally talk about the societal aspects of cloud computing here. But this one here is really interesting because this looks like one of the most interesting papers about how you apportion responsibility for your use of cloud services when, using these things. And for the longest time, we've had a real struggle because we haven't had access to any of these numbers.
And this paper really. Lays out a bunch of really interesting ideas with lots of really helpful diagrams, and it dies into how inside Google, people allocate carbon emissions for both internal use, but also for cloud customers. It's a really, fun paper, but it is also quite a significant piece of work.
Read. Like, me and Asim were quite excited about it, but we realized this is almost like a kind of book club kind of paper to read through, basically.
Asim Hussain: Yeah. Yeah. Yeah.
Chris Adams: So, Asim, I'll hand over to you because, there's a couple of things I'd like to draw attention to, but I suspect there's some things that caught your eye as well, or at least maybe you might have some context about why the two of us might be so excited about this.
Asim Hussain: Well, I just think, I think it's really exciting. Well, I get really excited when any organization does such a deep, thorough analysis of their emissions. The thing I'm kind of like going through as I go through this is like, part of me is like, I'm going to try and like represent some of this stuff as an impact framework manifest File because I can read an impact framework manifest file and I can compare it and I can look at it and I can know what's going on. Like one of the first things that the paper outlines, the approach that Google developed to quantify location based emissions of its individual products. And now I'm now like, I now need to dive into this paper to understand, well, yes, I'm understanding the, my definition of location based. However, I'm also seeing references to CFE, which to me doesn't factor as
Chris Adams: CFE being carbon-free energy
Asim Hussain: carbon-free energy So like, there's like a lot of nuance to this stuff. and yeah, I I would probably, as an experiment for me to try and understand the paper, I might try and represent some of this stuff as a manifest file. 'cause for me that's quite useful as a way of, learning something in, in a way, in such a way where I can compare and contrast it to other methods and methodologies as well.
But it looks very, interesting, very exciting. And when Google first worked on its carbon dashboard as well. They were the only organization to have done it the slightly different way, which is kind of like bottom up, like from products and services up towards the top, and now IBM has actually done actually to give IBM some credit as well. And they've got great white paper as well. If you're not ready, I'll send it over to you, Chris. they've done another approach is kind of bottom up as well. And so they're the only two organizations that have gone bottom up. The other ones have gone top down and I'm always excited when people go bottom up, because then you get the data with the granularity, the level of products and services that people actually use, which they drives.
Which is what you need to drive emissions reductions. So that's it. I haven't read, I haven't read almost any of this So I'm interested to get Chris's, if you've got any time, I don't know. What are your hot
Chris Adams: So, the thing that really caught my eye from this is that it talks about some of the ideas that Google has been doing that other organizations aren't so open about. So, for example, you have like a given amount of capacity that might be available inside this. Now, what Google have done previously is they've kind of said, well, We know that we've got a certain amount of, kind of, energy that's coming from green sources and we've said that we want to have a percentage of our energy that is always running at, say, 100 percent carbon free, for example, right, and they, Google have an approach where they say we count something as carbon free if it's matched at the time of use and if it's I think it's on the same grid.
So it's not literally a solar power, but a set of panels on the data center. It might be a wind farm that's where you could plausibly deliver the power to that place. And they use this to kind of represent the amount of clean capacity as something which they call a virtual capacity curve, because it changes over the time of day, basically.
So they talk about in this paper, there's some production, there's some production loads, like that always have to run and always have to respond very, quickly. And there's things where they've got a Bit of freedom in how they move it around and I think this is actually quite interesting because they talk about where they have some flexibility inside this and they talk about how they account for a bunch of that because it's the first time I've seen a paper, A, talk about this, but also talk about the fact that there's like, a set amount of kind of idle power then there's amount of power that will kind of ramp up and down based on the amount of use you're introducing they do they speak about a bunch of really interesting things inside this and the thing that i think there's like there's a couple of figures which i really find like really quite fascinating actually and the fact that there's like one thing like If you are at all interested in, like, Sankey diagrams, they've got this really cool Sankey diagram of saying, well, this is all the kind of power that goes into running machines, running the overhead.
This is how it gets proportioned across all the different services. And this is how these end up being allocated to both our internal use, but also cloud customers and stuff like that. It's a really, fun read. And I'm probably going to spend like, I think, an afternoon or maybe the weekend making some more notes on this.
'cause there is a bunch of stuff which is beyond my can, like some of the equations are. I, don't have the, I don't have the ability to kind of make sense of those. I am looking forward to reading this nonetheless, because it's really nice to see something like this, not least because by putting this into the public domain, it's now raised the bar for some of the other providers to be more transparent about this.
Because if you're looking at, say, Amazon, you're looking at Amazon's calculator, you don't have scope 3 emissions, so that could be up to 99 percent of your emissions not accounted for in the numbers. So if they look suspiciously good on the Amazon dashboard, maybe they are suspiciously good. But also you look at the resolution.
This is something where they providing information at both the location and a product line value. So let's say I'm using Cloud Run or one particular kind of storage. I can see it at that kind of resolution. And that's that kind of location. In some other providers, you might see Europe and then compute.
So, there is nowhere near that kind of resolution. So, people talking about this is how we do it. This is how it's possible. This is what you should be expecting from other providers. I think it's really, good. And they also do mention the fact that they're using high time resolution. So, they say, "We're using data from electricity maps to help us work out these hourly curves, so that we know at what times of day, what the kind of carbon intensity for the power might be, so that we know that we've got this much kind of green compute that we can plausibly use," and in a defensible and transparent way, say, "yes, this really is running on renewable energy, according to the way that we talk about this."
And like, they do refer to like, they, you don't need to agree with the approach that they use in order to at least understand where they're coming from, because there's plus points and minus points with using the approach here versus having something totally location based like we've spoke about before.
So, yeah, that's kind of my initial thoughts when I see this, actually.
Asim Hussain: We need to get this to the point where it's kind of easily absorbable and understandable by folks in such a way that they can actually, because you know, otherwise it's just, oh, right. There's a great paper we'll use, you know, but I think you need to understand the nuance of a lot of this stuff.
I don't
Chris Adams: Well, there's two things, but what I think, so... you know, there's a, the real time cloud working group that Adrian Corkcroft is
Asim Hussain: Oh yeah. Yeah.
Chris Adams: He's been really pushing on a bunch of this stuff. This feels like one of the abs, absolutely worth sharing to that group because they've been doing a really good job of actually collating this data so it can be used.
And like, a scene like that was the thing that kind of fed into the impact framework stuff, right? So there is a path for this. And like, There's a job to decode some of this and make it easier for lay people because, yeah, like, Asim, we've been talking about this for literally five, six, five years to get to, for us to understand why it was exciting.
But yeah, you do need a job to actually make this easier for people new to the field because there's lots of developers who are kind of coming into this kind of sustainable software field.
Asim Hussain: you know, we should also have and this, we're now talking about GSF work. I know I've actually got to drop like very soon, but this is interesting because as I just asked you that question, I think an answer came to my head as well, which is like standardization. So, you know, as we like to, even as a real time cloud project is kind of evolving, like one of the things it's trying to figure out and work through is like, like you mentioned CFE and your definition of CFE.
And I can tell you right now, I've heard different definitions of CFE from other organizations, which don't. Count the, it to be, it doesn't have to be the grid. It could be anywhere. So I think some of this stuff, it might be interesting to have conversation with Google, like other ways to standardize some of the terminology, the methodology, the equations to this, as soon as you can create a standard, maybe something we can push into ISO or something like that. That kind of also in a way also forces the, not only simplifies everything for everybody, cause they're like, "well I don't really understand what the standard is, but I can see that it's got wide adoption and it's a standard and competitors have got together and agreed on this standard." So I don't really, you know, and there seems to be a wide body of people who support it. I don't need to look at this. This equation is so thick. I'm staring at this equ
Woo!
Chris Adams: it's quite a man,
Asim Hussain: It's a quite, a yeah, it's quite a lot. Selecting it has like 43 components in the selection.
But...
Chris Adams: the thing that we can maybe talk about is that there are standards. We don't need to be doing all this work ourselves. Like energy TAG is one standard that is essentially written into European law and American laws around hydrogen now, like hydrogen production. So there's things that, you know, we wouldn't be starting from zero.
We could be using some of that stuff, but you're right.
Asim Hussain: We how we push
Chris Adams: Yeah. I need to also. Thank you. Trying to rush us through for the last few bits, because there's a few things inside this that, as hosts, I need to be doing, and I sadly can't talk about the marginal carbon intensity of baked bread, because we had a really lovely follow from Dr Daniel Schien, who responded about the low carbon bread and high carbon bread thing. So maybe we'll, actually, we should commit right now to do an article about the, about how green the energy is, and how you talk about that, because this is what we just spoke about. Right now, Dr schien raised a number of really good points and linked it and shared some really helpful resources with us.
Okay.
Asim Hussain: we should get a proper conversation together with, if we can, like with maybe EM and some
Chris Adams: Yeah, we have some people inside the organization who've been doing that. Alright, let's look at events. So, stuff that's coming up. HotCarbon is a workshop on sustainable computer systems. This is happening on July the 9th. It's free to attend virtually. You can turn up in person if you're in California.
It's really, good. And I, there's, they have videos online and really fascinating papers. It's really worth reading. It's like absolute cutting edge stuff. There's also a, the IEEE. They have a cloud conference on the 7th through to the 13th. This is in Shenzhen, China. And for the first time I've seen, there appears to be something like a sustainable AI track, or sustainable computing. You are? Oh,
Asim Hussain: to be there. Yeah. I'm going to be there. Yeah. Yeah. I'm going to be there. I was invited by, well, the, anyway. So yeah, I'm going to be, I'm going to be, I'm going to be over there. I'm going to talk about sustainable AI in the cloud. There's going to be a whole track, several panels, discussion topics.
And
Chris Adams: Oh, wow! Cool!
Asim Hussain: You know, I don't think we particularly speak too much to, I'm a big believer in that this is a global challenge and a global issue. And yeah, most of our conversations happen in the Western world. So one of the things I'm personally trying to do is to try and
Chris Adams: Bring the other 1,5 1 point something billion people into that. Okay, great. Okay, so other two things we had. There is a, so outside of programming, this was shared with us. The masterclass on becoming a sustainable UX designer. This is one thing that's been led, I believe, by Thorsten. Oh, Thorsten, I'm afraid I forgot your name.
Thorsten Jonas. He's been helping organizing Sustainable UX as a community, which has the unfortunate short name of SUX, S U X, but it's, this is a virtual masterclass you can join. We'll share a link to that inside the show notes that's taking place, and he has a number of these coming up. And there's also Asim, I think this is yours, right?
The Software Measurement Landscape Workshop on the 9th of July? That's something that you've got, you're looking at?
Asim Hussain: That's not, well, no, not, it's yeah, it's not something I'm particularly involved with myself, but the green, the, we have a meetup group in Brighton in the UK, organized by an organization called Root & Branch, which do a lot of green software, the kind of green software startup, and they're doing a workshop on software measurement.
Chris Adams: Folks, ah, so Cardamon were the people behind, sorry, Root & Branch were behind the Rust based software tool called Cardamon? I think Cardamon, it's monitor for some of this carbon related things. Okay, that makes sense now, putting two and two together. Okay. And we have one last wrap up before I need to go, actually.
So, this is the, this week in Green Tech, where it's basically you and me, and occasional guests talking about the news. We have a, there's a new podcast, or a new podcast related to the Green Software stuff called CXO Bytes, and this is, I believe, Sanjay Poddar, the chair of the Green Software Foundation.
He'll be talking a little bit about some of this and speaking at the, I guess, CIO level, something like that. That's what we have. So, there's a first episode coming out in July, for this and, yeah, I figure we should tell people about that because it's something that's coming up too. Alright.
Asim Hussain: That will be, that's a separate, just everybody knows that's a separate podcast, so you'll have to find it and put it in the show, no late. We have to find it and subscribe to it separately. And as Chris mentions, like Sanjay, I'll be speaking more to the c-suite level of an organization and like one of his personal beliefs, and I do subscribe to this as well, if you want change in an organization, it needs to come from both directions or it will just stall in the middle. So he's very keen
Chris Adams: Getting by on it at the management level, at the very top, yeah? The big cheeses,
Asim Hussain: do is big cheeses and we do the witherbread whatever. That's
Chris Adams: Or the mushrooms, if you prefer, the
Asim Hussain: right. There you go. Wither, wither, yeah.
Chris Adams: breaking down barriers to adoption, like you break down dying logs, right? Okay. But Asim, really, enjoyed this, mate, I'd really love to see you again, and get well soon. I hope the Nurofen and coffee thing doesn't have to be a daily thing, and that you get better over the weekend, alright?
Take care of yourself, mate.
Asim Hussain: Wonderful. You too.
Chris Adams: Okay, ta ra!
Asim Hussain: Catch you there.
Chris Adams: Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners.
To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again, and see you in the next episode.
79 episodes