Mike Riley: Well, I’ll start, and then Ricardo, you can certainly embellish further… So yeah, the Pi’s - they’re in various configurations. These days you can get up to a Raspberry Pi 5, which is the most powerful. Unfortunately, it’s also the most expensive. Myself, I’m still fairly comfortable with the Raspberry Pi 4, especially the 8-gig model. It’s got plenty of RAM, it’s got plenty of CPU capacity, especially for the smaller jobs. And again, because Go is so small of a footprint, it doesn’t take a lot of resources to run.

[00:12:11.04] So you can run multiple Go applications, within containers no less, on a single Raspberry Pi. And that’s what I’m doing. I’ve got a Raspberry Pi cluster right now that’s doing a number of the jobs that we’ve written for the book. Now, in the book we didn’t want to make it more complex and add Kubernetes to the mix, because that’s a whole other book in and of itself. So we’ve instead recommended that people build their Docker containers and then run it within a Docker container. But needless to say, even running within a Docker container, we can run multiple – in fact, we can run all the projects in the book on a single Raspberry Pi, if necessary.

So obviously, get yourself a Raspberry Pi. You do have to know a little bit of Linux, a little bit of command line. Of course, there’s tons of books, as well as videos and everything else out there for that. And it’s fairly easy to get up to speed, even if it’s basic command line. And we do provide all the command lines necessary, although, again, it really helps to have some background, a little bit of experience with them.

Once you’ve got the Go SDK installed on your Pi, the sky’s the limit. If you already know Go, you can go ahead and start building programs right then and there. Now, there is an extension library specifically written for the Pi, that allows you to access the GPIO pins, which is the reader pins, the header pins on the Pi, as well as the built in LED, so that you can blink that as well.

Now, one of the really cool things that I’m really, really pleased about in the book, and it really didn’t come until the very last minute, was the fact that we were able to use TinyGo. And TinyGo is a specifically designed subset of Go, that is designed specifically for embedded devices. Now, you could, I suppose, run TinyGo on the Pi, but the Pi has got plenty of resources. I’m talking more about the new Pi Zero, which is an embedded device really just for – kind of like an Arduino in a way, in that it’s specifically purpose-built for being able to pick up sensors, or control actuators. And what that really does - and for us, in a way - is one of the nice things of Pi Zero… In fact, you really want to get the Pi Zero W, which is the Wi-Fi edition. Because there’s two models. Wi-Fi is only like two bucks more, so it’s definitely worth it. But it allows you to put that anywhere you can get a Wi-Fi signal.

And in fact, this kind of fulfilled that dream that I talked about earlier, with the Changelog folks. One of the hosts had mentioned that they wanted to find a way to monitor their freezer, because it was having some issues, and they didn’t want to wake up one morning and discover that their food had spoiled. So I thought about that problem and we’ve actually solved that problem in the book.

So you can simply get a Pi Zero, you build a TinyGo program, install it on that Pi Zero, and the Pi Zero will then not only connect to the Wi-Fi, but using the onboard temperature probe on the board, you can then capture in real time what the ambient temperature is around. You tie that into the ability to export that result as a JSON format, consume that in Prometheus, and then Prometheus can provide that to Grafana, so you’ve got a full real-time map for as long as you want it. If you want it for two weeks, two months, whatever; you set the parameters. But now you’ve got a complete fluctuation, or – what I use it for is internal monitoring of my temperature, because I’ve got different temperature zones in the home. So I can see, as the day comes, you get the cyclic cycles, and it’s really kind of neat to see how that happens.

[00:16:00.14] But I also have it so that if that temperature starts to hit a threshold, I then tie into Grafana’s alerting capabilities, so that it contacts myself. And if you do decide to go with the Grafana’s free cloud version, they’ve got another add-on feature for incident management called OnCall. And it actually saved my bacon once, because we had a pretty brutal winter here, where the heater went down, and I didn’t know about it because I was away at the time. But then all of a sudden I get these alerts saying “You’ve got a problem.” And fortunately, I’ve got my kids on the OnCall as a backup. And so they’re like “Hey, dad, I think there’s a problem with the heater.” I’m like “Yeah, there sure is.” Fortunately, we –

Spencer Kimball: And that’s great, to have that experience to make product decisions; it’s pretty fundamental. But the idea would be hey, you wanna stand up your database pre-production, but you have developers that are pinging it, and so forth… You certainly don’t have to pay for that. You don’t have to have this big server that’s sitting there, that’s almost completely idle for months…

And then you launch the first version of your software, you get something into the app store, maybe it’s in TestFlight or something, and you have a hundred beta users that are poking at it, and so forth - you’re still under the free tier, for sure. It’s only when you really scale to get more product-market fit and you start having sustained high throughput, then you start to get into overages, and you can pay for exactly what you’re using over that free tier threshold.

And then, eventually, if your startup continues to succeed, you’re gonna want to move to sole tenancy, a dedicated cluster, as opposed to the Cockroach Cloud free tier and the overages where you’re sharing a multi-tenancy cluster with other users.

[59:56] So for infosec reasons, so that you don’t have noisy neighbors and you have very guaranteed throughput, exactly what you expect, and there’s no variance in terms of your latencies and response times and so forth. And also, in order to truly scale to big sizes, where the cost is more economical - that’s where you’d move to the dedicated cluster, and there you can scale to really as far as your ambitions.

Spencer Kimball: And that’s great, to have that experience to make product decisions; it’s pretty fundamental. But the idea would be hey, you wanna stand up your database pre-production, but you have developers that are pinging it, and so forth… You certainly don’t have to pay for that. You don’t have to have this big server that’s sitting there, that’s almost completely idle for months…

And then you launch the first version of your software, you get something into the app store, maybe it’s in TestFlight or something, and you have a hundred beta users that are poking at it, and so forth - you’re still under the free tier, for sure. It’s only when you really scale to get more product-market fit and you start having sustained high throughput, then you start to get into overages, and you can pay for exactly what you’re using over that free tier threshold.

And then, eventually, if your startup continues to succeed, you’re gonna want to move to sole tenancy, a dedicated cluster, as opposed to the Cockroach Cloud free tier and the overages where you’re sharing a multi-tenancy cluster with other users. So for infosec reasons, so that you don’t have noisy neighbors and you have very guaranteed throughput, exactly what you expect, and there’s no variance in terms of your latencies and response times and so forth. And also, in order to truly scale to big sizes, where the cost is more economical - that’s where you’d move to the dedicated cluster, and there you can scale to really as far as your ambitions.

Alya Abbott: And that was really – it was framed as kind of a positive, but what we saw is a huge influx of folks, communities who can’t afford something like paying for a pro plan on Slack, leaving Slack and importing their data and moving to Zulip. And – I mean, for us, we have a really robust sponsorship program for communities and open source projects, nonprofits, education… All kinds of non-business uses for Zulip. We really try to enable folks to benefit from our software.

So we do sponsor free Zulip cloud standard plans for folks… We have I think over 1,500 sponsored organizations at this point, so it’s a really robust program.

Nikita Shamgunov: Definitely. If you want free forever, we have our software and we give our software to up to four servers - like I said, it’s a cluster software - to install whenever you want, and run forever. We call this our software free tier; it grew three times over the past year from a number of active users standpoint. It’s basically one of the best column stores on the planet. Data is highly compressed, it’s stored on disk, very fast reporting, everything is updatable, transactional system of record.

So where other companies that run on premises - the Verticas, the Greenplums - they wanna charge you for that, you get it free, and you can put billions and billions of data points in the system, and get very fast SQL response from it.

In the cloud, our free tier is time-based, so I encourage people to log in… You can play around with the system. That would allow you to not use any software, and consume everything as a service… But because we’re running it on our infrastructure, we’re limiting access to free for a period of time. We’ll be announcing more changes there. We’ll give the system to you for free forever, for limited usage, but that hasn’t come out yet, so that’s something we’re working on. So that would be probably the best place to start. And of course, go to forum.memsql.com to learn about the systems.

Florian Forster: Yeah. So that’s the thing I think went great with the cloud service. So free tier is definitely a thing we will reiterate on, even though you could run it on your own. I mean, it’s not like there is like a feature gap, or something like that.

Evan Weaver: So we are also free. In fact, we can be more free than other cloud database vendors, in particular like the – the Fauna architecture, it’s a true API, it’s multi-tenant, the isolation is internal to the database kernel… So you’re never paying for idle capacity.

With most cloud databases you sign up and you get like a 30-day trial, because they deployed some container or VM and it’s costing them $100/month. That means some salesperson looked at your email, decided you were worth spending $300 on a database date and you get your 90 days. Then after that, they call you and they’re like, “Do you wanna pay or do you wanna go away?”

Fauna - we have no fixed costs for a new user. We only have to pay for resources actually consumed. So anyone can sign up for Fauna for a free forever database. You don’t need a credit card, or anything. Then if you start to scale, then you can start to pay for it by the resources consumed. So the actual economics of it are much better, both for us as the vendor and for the customer, than your typical managed or containerized deployment of any kind.

[52:06] I think some people do care about open source. Fauna is not open source, it’s only cloud; it’s only proprietary. But the majority of the market, in our experience, cares about the interface more than they care about the codebase. The number of people who are gonna crack open their database and make some fix is very, very small. Most people treat databases as some kind of artifact that’s handed down from the gods… Opening it up and changing the implementation is the last thing you would ever try to do. You’ll exhaust every other opportunity to fix your problem before you would take that risk.

Tom Wilkie: Thank you very much, yeah. You know, I miss so much out of what’s happened because unified alerting is a huge step in the Grafana story. I’m really pleased as the way the company came together. We used to have two alerting systems - we had the Grafana alerting system and the Prometheus alerting system. And they were worlds apart. On one hand, the Grafana alerting system is probably the easiest one that exists out there; it’s very accessible, very easy to get started with… And on the other hand, the Prometheus system is probably one of the most sophisticated and powerful ones.

So I think it was really exciting how the team could combine the power of the Prometheus system, with multi-dimensional alerts, with alert managers routing, grouping and deduping and silencing… And bundle all these features into Grafana in a way that makes them easy to use and gives you that level of user experience that people have come to expect. And best of all, we haven’t duplicated any features. We’re just using alert manager under the hood. We’re using the same API as Prometheus under the hood. So it’s true to our open source roots as well, and that’s – the team did a fantastic job with unified alerting.

I think the thing you said about cloud, the generous free tier, for instance - we launched that in January, I think…

Dustin Schau: Yeah, good question. Incremental Builds requires cloud tooling to be an effective solution today. In the future we definitely want to explore that space to see what’s possible in open source. You’ve mentioned a couple things that I think are really interesting.

Earlier there was, you know, “Could we use distributed builds locally?”, and the answer is “Not today.” But it is something that we wanna keep exploring, because again, it’s a really exciting space. One note here is that there’s like a super-generous free tier, just like other platforms, so we will be able to use all of these features in Gatsby Cloud on the free tier, in general.

Johannes Schickling: Yeah, that’s a good one. So think about that, and for your data across any kind of database - one for developers, but also for non-developers, for more business users. We see a lot of companies using Prisma Cloud basically to roll out access to their entire customers support team, to the entire marketing team, to the entire sales team, that they get access to the actual application data without the development team needing to build a lot of admin back-end interfaces, or these admin panels. So that’s certainly a workflow that we see a lot, but also workflows around typical development processes.

[55:40] Thinking about continuous deployment, we actually will roll out an integration with GitHub, that you can simply push your changes to your GitHub repository and Prisma Cloud will deploy the database changes on your behalf. So if you’re familiar with Netlify for example, how you can simply push your changes to GitHub and Netlify automatically deploys your new website. Think about the same thing, but actually for your database and your database migrations, where changes that are non-destructive - they can be configured in such a way that it’s just like rolled out.

And if there are database changes that might actually introduce some breaking changes, that’s sort of batched up and you can manually decide to roll it out into production. So various workflows around that, and this is what we’re offering for free with Prisma Cloud, that people can just get better workflows. And if people want the same sort of workflows, but more in a company setting, where all of that lives in your own cloud, or where all of that lives on premises, this is where you can get the same features as part of the enterprise edition.

Pam Selle: Unless it’s absolutely constantly… I think it would be a challenge. Maybe your Giphy instance maybe could hit the limit, but you get a million invocations on AWS, two million free on Googe Cloud, and one million on Azure.

Dustin Kirkland: That’s a good question. I mean, it’s extremely important to the long-term sustainability of Ubuntu. The success of Canonical is very importantly tied to Ubuntu being continuing to be available to everyone in the world. Canonical itself has a couple of businesses, a couple of parts to the business. First of all the company is really split into two halves. One half works on laptops, desktops, phones and devices, now sort of IoT connected devices. The other half of Canonical works on servers and clouds and virtual machines and containers. But the two work… We share a lot of technology. Our businesses are a little bit different.

At its core, Canonical is a services company. We offer suites of services that surround Ubuntu. We don’t charge for Ubuntu. The bits that make up Ubuntu on your laptop or desktop or in the cloud are always free to you; you’ve never paid for Ubuntu itself. But there are many enterprises and organizations and some individuals that choose to buy auxiliary services from Canonical. And those services, at its very core, often start with support services, so we do have commercial support where we try to help ensure large institutions, typically, that have built a business around Ubuntu and they heavily depend upon Ubuntu. We can try to help ensure their uptime, the security of their packages. They have a phone number to call, they have an interface to connect with our support engineers, who can then help troubleshoot or debug maybe a kernel performance regression, or a security issue, or something like that. That’s our support services.

Then we also provide consulting services. Some of the biggest telephone companies in the world - AT&T, Deutsche Telekom - are building their data center infrastructure services around Ubuntu OpenStack, and we engage with them at a very large scale to provide consulting services where we help design and implement that cloud. [01:16:12.28]

And then we have a large number of customers where Canonical provides what we call level 3 support. So that customer might take the first and second phone calls from the customer. IBM is a good example. IBM has its own customers, let’s say the financial industry, Wall Street or something like that, where they’re using Ubuntu. They call IBM for the first level and the second level of support, but when that IBM engineer needs help from a Canonical engineer, IBM can then pick up the red phone, call Canonical and receive first-rate Ubuntu support from an Ubuntu expert. So those are sort of the support services.

Then we also do… Let’s call it OEM work, or work with OEM, which was originally hardware manufacturers, ensuring that Ubuntu, as it ships on Dell PCs and HP servers and IBM servers and lots of others, now network switches, as Ubuntu ships those devices, we help that engineering team build the image of Ubuntu that runs on that switch or on that server. But we’ve also seen that same model work very well with the major public clouds where Canonical curates, secures, provides tests, QAs, the images in those clouds. So all of those are commercial engagements that help sustain Canonical, helps pay the salaries of the 700 of the 756 of us working all around the world.

David Flanagan: I mean, I’ve sold my soul a few times over now, but yeah, I’m in a good place. Yeah, DevRel’s – things are tough right now. I’ve got a lot of friends that have lost their jobs, and the entire tech market is in a major phase of cutbacks… But DevRel’s not going anywhere. I think we all agree, developers - we smell market bull***t a mile away. And the problem was all these companies hired DevRels to be marketers, and not to do proper developer advocacy and relations. And then they put out market [unintelligible 00:42:11.20] content, developers smelled the bull***t, and now they’re all like “Oh, we’re not getting the results we expected.” Like, no s**t. Like, you can’t just take the same thing you’ve been doing for the last 20 years, call it DevRel, and then expect to have a massive upturn in your business.

And there’s also the challenge that a lot of people are looking at like the Linux Foundation and the Cloud Native Foundation as free marketing, donating their projects… And that’s causing a lot of hassle too, because these projects think “Oh, we’re now an incubating CNCF project. We’re going to get all of these contributors. People are going to fix all the bugs and we’re going to get all this adoption.” But you’re still just one project in a sea of many, in an evolving landscape, with developers that are frustrated trying to keep up with this race that they didn’t sign up for. No, it’s not going to work.

So I think we’re going to see a lot of change over the next two to five years. Developer advocacy and relations is going to be stronger than ever, with the right people, with authentic, credible messaging, with companies that get it. And everyone else will go back to doing what they were doing before. Dodgy mailing lists, dodgy advertising, and… Yeah.

Greg Serochi: That is the best way. We’re going to have pretty soon some free access to Gaudi. So right now you need a developer credit. We’re working to get some nodes available for free on the developer cloud, so people will have the ability to try out our tutorials and examples and code examples, and be able to see and experience the Gaudi usage, and see how easy it is to run.

Jason Bosco: Yeah, so people do ask us regularly for an unlimited free tier in Typesense Cloud. Right now we give out a free tier for 30 days, and after that you have to start paying… But I think the difference between Algolia and Typesense is that Typesense is open source. So if you wanted it for free, you could definitely run it yourself, and it’s fully featured, and there’s a community UI. You can basically run this whole thing if you were willing to put in a little bit of effort; you can get this for free, an unlimited amount of time. So I’d say that is kind of equivalent to Algolia’s unlimited free tier, which does have a lot of restrictions, and you can only put so many records, so many searches… With Typesense, this “free” tier - it’s unlimited everything, except for, of course, the infrastructure costs that you’ll have to pay any cloud provider… Or if you wanna it on your machine, then it’s going to be completely free, except for the electricity. So that’s how I think about it. So if someone says they absolutely want a free tier, I just tell them, maybe sign up for one of the cloud providers. They offer a better free tier for at least like a year, or give you free credits if you open new accounts, and then just run Typesense under your own cloud account and you get it for free.

Jerod Santo: On the money-making front, this is something that I don’t think is Vercel-specific, but it’s more like cloud hosting problems, is generous free tiers, or zero credit card required. Maybe you’ve got to put it in, but just no cost to get started, and then a surprise bill when something happens that you didn’t realize your side project gets a little bit of action, or whatever happens. And then this is bad for your overall image, because people are posting pictures like “Oh, look what Vercel did to me, $5,000 overnight.” As a product guy, you’re certainly trying to solve that problem too, aren’t you?

Katie Gamanji: I actually have many people asking me that… Some of my friends - they are actually developing products, startups, very small startup companies, and they’re asking “Is Kubernetes the right thing for me at the moment?” And usually, what I answer in that circumstance is gonna be probably no, because it’s just only two people, both of them developers, they don’t have any infrastructure developer or cloud platform engineer within their team… So for them, managing and completely deep-diving into the Kubernetes architecture and management maybe is not the answer.

So in that circumstance, I would usually maybe suggest a cloud provider. Again, you have free tiers you can use from different cloud providers… So I would usually recommend that. However, if you are in a circumstance where you have enough engineering resources and you have enough expertise of maybe understanding how to run an infrastructure - not necessarily Kubernetes, but the basics of what exactly an infrastructure is composed of, then probably Kubernetes is gonna be the answer moving forward. Because when we’re talking about Kubernetes, it’s not just about containers, it’s about how these containers are managed, and what kind of leverage you get in a production environment by using Kubernetes.

I’ve been mentioning the scheduling capabilities… For example, you have – maybe I should introduce Kubernetes very briefly. So Kubernetes is pretty much an orchestration platform for containers that is run across a distributed amount of machines. So you can have different instances, and all of them are gonna be put together to run your applications. Now, on which node, on which instance - that doesn’t really matter. That’s always gonna be abstracted by Kubernetes. So one of the capabilities is gonna be the scheduling.

So based on the requirements you have for your application - for example you can choose “This application should have this amount of memory and CPU at all time. This is the very minimum of the resources I need for it to be up and running.” The scheduler would take these requirements into account and place it on a specific node that will have all of these available resources.

Now, the thing is if that particular instance goes down, usually, if you’d be working in a data center, you will need to migrate the application, or you will pretty much need to trigger your load balancer to point to a different data center. Now, with Kubernetes all of this is automated. So Kubernetes will be managing or monitoring the state of your application all the time. It will say if it’s no longer up and running and it will go back to the scheduler, and this will put it on a different instance, again, with enough resources and capacity to run your application. So all of these operations are automated… And this is just one of the functionality it provides.

[35:56] We have scalability, we have resources which will allow you to scale an application based on different events. For example, if you reach the amount of – maybe the maximum of the CPU or memory your application can consume, you’ll be able to scale forever. But now you can actually scale on external events as well. For example, maybe there is a queue you have in a – actually, an SQS messaging queue in AWS would be able to take those metrics, and based on that, maybe scale further. It can really go the extra mile [unintelligible 00:36:27.04] personalized scale mechanism. And all of this, again, is automated. You have a declarative representation of your application, so your application is represented as code. It’s YAML, it’s not necessarily the most readable thing out there, but it’s out there, and that’s gonna be representing the desires that you want to have within a cluster… And that desired state is always gonna be fulfilled. You have control managers which will always make sure that what you want is actually gonna be in the cluster. And it’s always in a loop to reach that ideal condition that you defined within the manifest.

So these are just some of the functionalities. I didn’t even talk about the Ingress and how you actually manage the reachability to your application, how you have this abstraction across a collection of pods of services, and how can you have granular control of how your application serves different HTTP endpoints with Ingress, or how your Ingress can actually point traffic to different services and different applications. So you have a lot of availability out there, and you have custom resource definitions, and you really can go the extra mile. It’s a tool that is very customizable. But more importantly, again, it has some of the basics very well set, so you don’t have to think about them anymore; you just take advantage of them straight away.

So in that case, if you have a team that would like to run an application within a production environment, would like to kind of take advantage of all of these capabilities that Kubernetes provides, and they have enough resources within an organization to run it, then probably the answer is “Yes, do look into Kubernetes and start rolling it out.”

Anais Dotis-Georgiou: Yeah, so we’ve just released a cloud offering, and there’s a free tier, so that’s probably the easiest way, because you just have to sign up and then you’re good to go. Otherwise, you can install the platform as a single binary, and then I recommend checking out Telegraf, even if you are not interested in InfluxDB. Telegraf is a collection agent. It’s a single binary, and it’s plugin-driven. It’s database-agnostic, so it’s by far our most popular tool. There are over 180-190 plugins… So if you’re looking for a way to collect data and you haven’t found something that you like, I recommend looking into Telegraf, any which way; it’s completely open source.

Adam Azzam: Yeah. You know, one thing that you had said which I did gloss over is - you had said “What’s the experience of deploying?” and then “Tell me about a shiny dashboard that you see”, or something like this. So once I go through all the work and I have my workflow running out on the world, I have this really beautiful UI that I can log into. And when I access that UI, what it displays for me is it basically tells me – it’s almost like standing on the the platform above the factory floor, right? You can see everything that’s in progress. You can see the box of widgets that everything has produced at the end, and you can see everything that was broken in the process.

And so if you’re the type of person that treats your workflows as cattle and not as pets, it’s very easy to see “Yeah, you had 10,000 jobs that succeeded. Here’s the 10 that failed.” You can easily click in, see “Okay, the 10 that failed… Why did they fail?”

If you’re on Prefect cloud, which has a very generous free tier you can run a business on, we do AI summaries of the errors… And so we will summarize like “Look, these workflows are failing for these reasons in natural language”, so that you don’t have to dig through a 10,000-line stack trace just to find out that you had an out of memory error.

Break: [00:37:17.06]

Jerod Santo: Yeah… Well, there’s another factor there, which is we had a decade of hyper-growth, 0% interest rate-based investment, and the free tier for a startup is an epic win for growth. I mean, it’s a massive way to get lots of people. I think – was it Jason Bosco from Typesense, Adam, that we talked to about how they didn’t have a free for open source, or a free for whatever offering with their cloud, and how that is one way of going about getting a whole bunch of users, that you could just buy them, because you have investment money… And like those times are changing. I think that’s more of the – I don’t know if cynical is the right word, but maybe the hairier side of what those things are. Of course, we wanna always just be like “We love developers, we love open source, and we want to see people tinker and have fun, and here it is…” And I do believe that is sincere in many cases; it is also a hyper-growth strategy that’s afforded by large investments, and continues to run based on larger investments, which eventually run out, or need their money back. So like you said, eventually Salesforce is like “This doesn’t make sense for Heroku anymore, and so we’re not going to do it.”

Jerod Santo: As a matter of fact, what I recently did was I used to run Grafana on-site, on-prem, but they’ve got that Grafana Cloud… And I got tired of always having to upgrade anyways, so I just got lazy. They offer a free tier, you just have to be careful not to send too much data, otherwise they start to concatenate, so you don’t have a large history to be able to fall back on. But the Grafana Cloud that they offer at the free tiers - you know, that way I don’t have to worry about versioning and such, and it’s just data that I’m spitting out from my Prometheus on-site server anyway, that’s connected to that cloud. And it also allows me to check it wherever, on whatever device, whether I’m at home or not. And also, I’ve got the alerts set up so that if I do get threshold issues, CPUs running hot, whatever, I’ll get the alert from the cloud, as opposed to on-prem.

Stella Biderman: Yeah. So right now, the largest open source – well, okay, the second-largest open source English language model in the world is called GPT-NeoX. We trained that. I trained that. My organization. And that cost us about $350,000, or would have, if we weren’t given the compute for free. But $350,000 for the second-largest open source language model in the world… At the time we released, it was the largest. Later, someone else trained a bigger model, with sponsorship from the Russian government.

So GPT-3 came out in 2020, and for about two years almost nobody was training and open sourcing language models. Google was doing it with similar models, but not the same kinds of models that GPT-3 was, and we were doing it. It was really not that expensive. We got into it on compute that we got for free through a Google Research Computing program called the TensorFlow Research Cloud. And with that, we trained a 6-billion-parameter language model, the one that underpins the first version of Dolly, that he was talking about. That’s been extremely widely used, deployed in a whole bunch of different industry and research contexts, and been hugely successful… And it was literally just like compute Google gave us for free. It ran preemptively on their Research Cloud – basically, the idea of TRC is that they have a research cluster that they don’t always use all of. And so other researchers, independent researchers, academics, nonprofits can apply to be able to run preemptable jobs on their research cluster, and just use the compute that they’re not using at the time. And using that, we trained this model in two and a half months.

Chris Benson: [00:36:26.02] And I love seeing that from them. We’ve long talked about that if you’re probably looking a little ways down the road, AI ever integrating more and more with the software around it, to the point where it’ll be kind of ludicrous to have software that doesn’t have some sort of AI capability in it in the future. It’s feeling more and more like software in that way, that when we hit the million open models on Hugging Face, and then just seeing these capabilities coming up… You know, when you said that, that reminds me of all the major cloud providers will kind of offer a limited free tier, so that you can go do some stuff with it, and that’s kind of how Hugging Faces offering that with open models feels to me, in terms of being able to go use something when you might not have the resource otherwise. So yeah, it’s good stuff… But boy, gosh, the world is changing faster, isn’t it…?

Rahul Nair: Sure. So Intel Developer Cloud is Intel’s production-ready cloud specifically for AI and machine learning workloads. And of course, when we say AI and machine learning, it’s [unintelligible 00:19:22.01] so many other compute-heavy workloads can run really well on IDC.

So for this particular hack, we provided anyone logging into Intel Developer Cloud registering on IDC as a standard or free tier user - you get a shared Jupyter Hub instance, where you get access to Intel’s data center GPUs, Intel Xeon processors. And I would say this system, for a free tier user, I don’t think any other service provides. I mean, there are many services with a Jupyter Hub frontend, but the amount of compute and the amount of memory and RAM and even file storage that you get in the systems, I’m not seeing a single cloud service provider providing that. And we have seen a lot of people really using it, and giving us feedback on how we could even improve it.

Today on IDC we have a lot of models or LLMs already, and other tens or even hundreds of local models that we are planning to add further to boost this, that are Stable Diffusion models, LLM models, and things like that. Beyond that, for productionizing the workload - Dan, in your case, you’re using the Gaudi 2 accelerators. Those are specifically designed for workloads that request high bandwidth, like LLM and GenAI workloads, and this… I lost my train of thought, but yeah. We have Gaudi 2 accelerators, which we are seeing incredibly competitive, and sometimes outclassing the best out there for your particular workloads.

Along with Gaudi accelerators - those are specifically designed for GenAI and AI workloads - we have general-purpose GPUs, the data center [unintelligible 00:21:01.09] both with 48 gigs and 128-gig versions. So folks in the hackathon, they actually used both our fourth generation Xeon; that is the latest Xeons that we have, which - what it particularly does is that it accelerates your machine learning workloads. We have dedicated instructions in the CPU to sometimes even take your workload to 2x the performance that you got in an earlier generation.

It’s all about making the CPU as efficient as possible, and making it as fast as possible, and still maintaining the general purpose utility of a CPU. But then, like I said, the data center [unintelligible 00:21:42.08] a little bit more generic solution, where you can run your AI workload, [unintelligible 00:21:48.06] HPC workloads. Each of these machines, when you’re productionizing, you get a VM, you get an 8-node [unintelligible 00:21:56.22] system that also clusters the systems available. Then comes the Gaudi accelerators, both – there are single-node machines, and also clustered machines if you want to do pretraining or big fine-tuning. All those cool things. And soon, we’ll have Kubernetes service, object store, file store, all those things coming up… So it’s gonna be great.

[00:22:17.29] What I see is that if you’re building a startup, it would be very difficult to find a performant accelerator cloud like IDC out there. I’m sure that there are different hyperscalers… But this, uniquely for startups, from my personal experience, is a really, really awesome solution.

I’d like to know more from you, Dan. You are one of the first customers of IDC, right? What are the things that you thought that really made you decide to choose IDC, for the performance, and also the team side also?

Christina Warren: Okay, thank you. Okay, there was some amazing tech stuff. Some of the first things… Codespaces is now available to everyone. And it’s been in beta, at this point it’s gone through a couple of iterations, but it’s now available to all users, and free accounts get 60 hours of usage a month, pro accounts get 120 hours of usage a month for free. Codespaces, and just the whole movement of cloud development environments, whether using Codespaces or something else - it has fundamentally changed the way that I spin up projects and try things out now. That isn’t to say that I don’t still have local instances of things that are running on my own machines, but many, many times, if I’m wanting to check out something, if it has a Codespace attached, or a devcontainer.json file attached, that I can turn into a Codespace, I would much rather use that as a way to explore and test things, rather than doing the old method of installing the Docker file, or trying to run locally with my dependencies, and whatnot.

So Codespaces is going GA for everyone. I think that was huge. Really excited about that, especially – I’m glad that we have a free offering for people to get started, and to figure out if they like doing development from a cloud VM or not… We also announced - and I think this is really great, especially when it comes to accessibility and helping more people get more out of coding… We announced an experiment that’s taking place within GitHub Next, which is sort of our incubator for some of our next ideas. And it’s called, Hey GitHub, and it’s like a voice assistant for GitHub Copilot. And so you can use Copilot, but you don’t have to type. This, I think, has massive potential.

I was hit by a car a few years ago… I was crossing the street, and I was hit by a car and thrown underneath a bus; it was a whole thing –

Sherol Chen: But if you wanna try it, again, you would choose from those categories. So if you choose AutoML tables, there’ll be the opportunity – it’ll kind of pull up the UI in the cloud console, so of course you’ll have to have a Google Cloud platform account… But you can train for free for up to a certain amount, to kind of try it out for yourself. And then actually the cost isn’t very much. I’m not on the sales side, so I actually don’t know the exact numbers… But every time I’ve talked to a solutions engineer and heard them quote, it’s very low amounts of money once you actually wanna get it going. But yeah, I do know that there’s an amount that you get to use for free.

So if you just wanna check it out, try it out yourself, it’s pretty accessible. And even if not that, if you wanna even try some of the pre-trained APIs, like you wanna upload a cat picture and see if it knows your cat is a cat, you can just do that within the web. So if you just wanna see how well these pre-trained ones work, they have online demos within the website as well.

Evan Prodromou: [20:15] Yeah, so partly because my experience was Wikitravel, I was involved with a number of people who were interested in how free and open source software was going to be affected with cloud computing. This is like 2006-2007.

I was part of a group - The Franklins Street Group - that did some discussion and put out a statement (The Franklins Street Statement) about free and open source software in a kind of cloud computing world, and ways that people can be preserving their freedom in that kind of world.

At the same time, a lot of the social networks that we take for granted today were just starting to break through. In 2007 Twitter had their launch at South by Southwest, their big public launch… The classic epic launch at South by Southwest.