Hello. Have you ever been in a situation where you're overhearing conversation and it dawns on you this person well intended is wrong about something minuscule but important to you and duty. The universe compels you to tell them why they're wrong. Watch this. I recycle plastic because it's good for the environment. >> Oh, good. You recycle. How noble and delightfully basic. Just so you know, tossing plastics in the bin isn't a magic eco wand. Most of it still gets downycled. landfilled or shipped overseas. So, yes, helpful, but not the planetary rescue you imagine. >> Let's try another one. I have heard that strawberries actually count as nuts. >> Heard strawberries are nuts. That's not how botney works. Strawberries are fruits with seeds on the outside. True nuts are dry. Single seeded fruits like acorns. Close but irrelevant. Try a dictionary next time. You see how powerful this is? Now, we can be everywhere at once. Wherever someone is slightly wrong about something, we can be there to correct them. I have built a

fully automated mansplainer. The best thing is it all runs on a box that fits right in my backpack. This is the Nvidia DGX Spark. It's pretty cool. Nvidia sent me this. So, thank you Nvidia. So, before I'll show you how I built the automatic mansplainer, let's briefly dive into the Spark. It's pretty neat. It's really only this big and it packs a punch. It has 120 GB unified RAM. That means the RAM is for both the system and the GPU. So, you get like almost 120 gigs of GPU VRAM, which is fantastic, right? That's more than an H100. So, you can run pretty seriously big models on this thing right here. >> So, it's recording. This is Matis. He's an engineer at our company and he has genuinely no clue what to expect of this. In front of me is a laptop. When I press the space bar, it records audio and sends that to the

DGX Spark to run the whole setup. When I get back the result, it plays it out loud. You'll see at the same time how awesome this all is. But then also that if you chain these models together in a sub-optimal way like I'm doing here, there's quite a bit of lag involved which uh leaves some room to be optimized. But I think it makes for a good time. >> Tell me uh what you're doing in your free time. What do you like to do? Uh, in my free time, I really like to play around with uh, basketballs and throw them into hoops and uh, you know, like do a lot of like kind of sport gym stuff. Um, I really like that. >> Oh, that's excellent. Excellent. It it >> aha, charmingly vague. You mean basketball, not basketballs, unless you're practicing with multiples like a true prodigy and sport gym stuff. Do you mean exercise or perhaps work out? Let's aim for specificity, shall we? >> So, could you be a bit more specific, please?

>> By basketball, I mean I I like to like, you know, play with friends, like get a kind of like a hoop, you know, and really throw it in, try to score some points, like uh I'm really kind of like that. The German accent makes it so good. Oh, how charmingly folksy. You mean basketball, the sport invented by Dr. James Naymith in 1891. Not just playing with friends and throwing it in. How delightfully. >> Tell me an interesting fact about food. >> Did you know that uh Sriracha sauce comes from Thailand? It comes from a like a town called Sriracha. It's near Bangkok. It was made there. And Red Bull is also Thai. >> How quaintly uninformed. Sriracha sauce, you see, is indeed named after the coastal city of Sasha Sria in Thailand.

But it's not from Thailand in the way you're implying. The sauce we know today was popular. >> No. >> Is it because I spoke? I'm >> the sauce you know today was popularized by a Vietnamese refugee in California. >> Really? >> Yes. >> I genuinely did not know that. I thought I had a cool fact. I've been I've been saying this to everyone for years. This guy really sounds like I have a German flatmate that yesterday was explaining me that I need to swipe the shower the glass shower thing after I shower. Uh he was really explaining to me and it I'm getting flashbacks from this. >> Yeah, this is running completely on the DGX Spark. Yes. Yes. >> The voice sounds super realistic. >> It also corrects itself. >> Exactly. The way he said like Sriracha, sorry, Sriracha like that is pretty crazy. Sounds super >> ah the delightful cultural exchange of bathroom etiquette. I'm sure your German flatmate was merely enlightening you on the nuanced art of glass shower maintenance. Some of us just naturally understand the important

>> it has a context as you can see. >> The voice is super cool. I really How do you Is there like a list of voices you can kind of >> I don't know this was the default. I don't know why they make the default to be a super annoying German guy. It needs to become a bit quicker to make it really conversational, but >> really sick. >> Well, thank you. >> Good annoying roommate. Thanks, Yanik. Thank you. Thank you. >> The shoe, this must be shown in the video. I was recorded by a shoe. Must be >> This is what was recording me. I had to be serious with this thing and the shoe recording. >> The whole thing is actually just three models in sequence. So there's a whisper model that takes the voice and translates it to text. Then there's a mistral medium that takes the text and translates it to the man's explanation. And then there is vibe voice which is a project by Microsoft that takes the text and uh generates sound from it. And yes, the uh annoying German guy voice is just

the thing that came with Vibe voice. And I think it's just so fitting for this project. All of these models fit on the DGX Spark at the same time. Now, Nvidia has not just sent me one, but Nvidia is actually raffling out a DGX Spark. And later in the video, I'm going to tell you how you can win that DGX Spark. Stay tuned. The box itself is pretty cool, but what's also cool is what's on the box. I am the kind of person who buys an Android phone with as little stuff on it as possible. I'm usually the person that goes full bare metal plane because I do not want all the vendor crap on it. Nvidia did a excellent job of giving me just enough tooling to make me still use all the underlying things but make my life a lot easier when it comes to building things on top. So, let me show you what I mean. So, first of all, network connectivity is a breeze. you actually plug in the Spark regular keyboard monitor and connect it to a network and immediately it not only gets an IP but it also gets a local host name

that just makes it that much easier just to like SSH into it or something like this. So if you want you can use this as a regular device on the network that you can SSH into. Let me do this. So I just tried sshing into Spark here. It's disconnect. I'm going to plug it in. So, as I said, you can just SSH into this thing. Again, this is a regular Linux box with Ubuntu. And you can see it has 120 gigs of memory, has an ARM CPU. It has 3.4 terab of disk. And the crown jewel here is obviously the Nvidia GB10 GPU that's built inside. And you can see here again memory usage is not supported. Meaning there's no memory at the GPU directly. The memory is the same memory that the system uses. And that's pretty neat because you can now shift this memory around. If you're using a lot of memory for some process, you can do that. But

if you're using a lot of memory for running a model, you can do that as well. So Nvidia has really nice tools. They have this Nvidia sync utility where you can basically connect to the Spark. But the cool thing is to open this AI workbench right here. You can see I already have this open in the background here. And what you do in the workbench is very interesting. So this is divided into projects. You can open a new project. Once you do that, you get to choose a environment. Nvidia makes it very easy to work with containers on the spark. So let me create one of these pietorrch containers. So if one of your projects needs CUDA 12.6 and another one of your projects needs CUDA 13. If one of your projects needs Python 3.11 with PyTorch 2.6.5 and another one of your projects you want to be at leading edge, you can do that with containers. It even suggests to me a few things right away. So it installs a Jupyter lab for me. I can just start the Jupyter lab. I can wait for the container to start here. And then once I open that, I now have a Jupyter lab available. A bunch of

suggestions here. And if I do that, then I get torch. It's available here. This makes it super easy. It also means you can share these projects with others and publish them to Git. But let's make it even easier for people. Nvidia has a web page. They help you from doing basic things like setting up network access for your Spark, installing VS Code or code server on the Spark. So besides the starter packs here, we have all of these playbooks and these are pretty neat because they show you what you can do with the Spark. Here's a playbook that instructs you how to achieve your own local VIP coding setup running GPTOSS 120B completely local. Here is how to build a full enterprise rad application. Here is how to achieve NVFP4 quantization for Blackwell directly on the Spark. Here is a recipe how to fine-tune models with PyTorch. Now, I'm not going through all of these playbooks here because you would shatter out of

sheer awesomeness. That is me following instructions that are on a page. I did this one here. Live VLM web UI. You get the basic idea and you get a few instructions right here. It realistically takes about 2 minutes. And here is what that looks like. There's me. Let's see whether it recognizes me. Bald. Thank you. Bald man wearing sunglasses holding a microphone. Let's go here. Yeah. So this is a GMO 34B doing the recognition right here. What is this? A pair of headphones? No. Small silver cylindrical object. So I hope you can see how easy it is to get started here and to get up and running. Who is this for? I think there's two user groups that might be particularly interested in this. The first one is the group that really values privacy and autonomy. This thing is you saw like a single one of these is able to run GPT OSS 12B and that's a big model. So you're easily able to run the world's largest openw weight models with one or maybe two of these boxes that you stitch

together. The second group that is really interested in this is tinkerers. Having a local device means that you can play around. You cannot just consume models but you can fine-tune these models. You can experiment. you can go deep into the weeds and you can change things that you could never change if you had to deploy something in the cloud or you consume a cloud API. So if you fall into one of these two user groups valuing privacy or being super tinkery, this is a good deal. And to me, coming back to the box itself, the form factor is pretty cool because it means small enough to put it away somewhere and just SSH into it, but it's also neat enough to take it with you to somewhere, plug it in, be able to have your local entrance wherever you go, basically. Let's have a last go at a man's explanation. I really want to travel to Paris. Such a nice city. Oh, how original Paris. Yes. Charming, full of baguettes and Roman. As if the the whole city is a postcard. If you actually want culture, maybe learn a few phrases, skip

the tourist traps, and try real cafes, not chains. All right, so now I'm going to tell you how to actually enter into the raffle for the DGX Spark. Again, super thanks to Nvidia for doing this. Uh, in order to win the DGX Spark, you have to attend GTC. GTC is a conference that Nvidia is putting on every year and is super awesome. uh they have a lot of good speakers, they have tutorials, they have workshops, they have new releases, all kinds of things. And what you have to do is you have to register through my form. There's a link in the description. You have to attend GTC and you have to attend a session other than the keynote. So attend the keynote, but then also attend the session other than the keynote. Uh this is for people in AMIA specifically. So Europe, Middle East and Africa. If you are of a different region, you cannot win the spark. I'm very sorry. Um, but uh, register anyway. It's a super cool conference and I'm

sure it will be entertaining. All right, all the details are in the link in the description or you go to yculture.com/gtc and that will get you everything. Again, many thanks to Nvidia uh, for having me do this. This is super cool. The spark is super cool and I wish you tons of fun if you also have one or win one. Bye-bye.