Video ID: fcGD7kHgxqE
YouTube URL: https://www.youtube.com/watch?v=fcGD7kHgxqE
Added At: 13-06-25 21:17:15
Processed: No
Sentiment: Neutral
Categories: Tech
Tags: AI development, Nvidia Jetson Ora Nano, PS3 emulation, ARM device, robotics
Summary
• Nvidia Jetson Ora Nano super development kit review • AI development and basic desktop use capabilities • PS3 emulation on ARM device
Transcript
in this video we're going to be taking a look at the allnew Nvidia Jetson Ora Nano super development kit it's definitely a mouthful but what we've got here is a super small form factor development board that's really putting down some amazing performance for AI development and even just basic desktop use and really when it comes down to it that's what I love these boards for but this is marketed as an AI development kit and the original Aura Nano might be familiar to some out there but we've got a super on the end of this and it does make sense because they've been able to up the TDP and the memory bandwidth on this unit making it even more powerful than the original or a nano but the price point is much lower instead of the $550 price tag of the original Ora Nano this one is coming in at $2.49 making it a lot more accessible to people out there who've been wanting to get their hands on something like this and with that boosted performance this thing might just be worth it in this video we're going to be taking a look at the whole unit we're also going to be testing an llm on this a large language model just to see how it forms but what I really want to get into is just some basic desktop usage with Linux and I want to show you that this thing can actually emulate PS3 so out of the box a lot of people are going to be running the operating system on this from a Micro SD card but down here at the bottom you can see we've actually got two m.2 slots 2280 and a 2230 so if you did want to up that storage speed instead of using a Micro SD you definitely could with this unit when it comes to basic I/O up front here we've got our power input and this does come with a 35 wat power supply fullsize display port four USB 3.2 ports gigabit Ethernet and USB type-c and we've actually got more IO here like two mippy CSI 22 pin camera connectors so we can add dual cameras on this unit if you want to go that CSI Route 40 pin expansion header 12 pin button header four pin fan header so yeah I mean we've got more iio on this unit and when it comes to the overall specs the super version is putting down better performance than the original or a nano but both of these are powered by nvidia's ampe architecture so with the built-in GPU here we get 1,24 cudic cor 32 tensor cores and this will clock up to 1,20 MHz in this as opposed to the original 635 MHz in turn this offers up to 67 tops of AI performance we've also got a six core arm cortex a78 AE up to 1.7 gahz 8 GB of 128bit low power ddr5 and this is much faster than the original oron also this one's coming in at 102 GB per second and the original was coming in at 68 and we get three power mode so the super is going to be that 25 watt mode but we've also got that 15 and 7 watt mode also with the super configuration so jumping right into Linux here with the Orin super first thing I wanted to do was make sure I was going to get the maximum performance out of this machine right here up at the top soon as we're updated with the latest version got got our power mode and on the older version we had a 7 watt we had a 15 watt but now we've got that Max in so it's up to a 25 watt TDP that's what makes this super along with faster memory uh I've noticed that this will clock up close to 1.8 on all these cores got our GPU here and we'll take a look at this when we're testing some AI on this unit because after all I mean that's what it's marketed for but there's really no reason you couldn't jump in here and use this as a fun little desktop setup and that's exactly what I've been doing now uh I've got some stuff that I want to test here and one recent new breakthrough for emulation was the fact that rpcs3 is now available for arm devices and I got it to work here I am kind of Blown Away by the performance now it's not going to run every single game here but we will test some out and I'll give you a look at that because it's pretty impressive given that we're running this on an arm chip and it's a native ARM version but before we get into that I did want to demo at least one local AI workload here and what I'm doing right now is installing oama well actually llama 3.2 it's a large language model and oama allows us to easily run this natively we don't have to be online and as soon as everything's installed we will disable Wi-Fi here so I can show you that it's running natively on the Orin super and this is actually really quick so the Raspberry Pi can run something like this but it's really slow but given that we've got those Cuda cor here with the Orin super this thing is really quick I mean it actually handles this just fine and now that this is completely installed what we're going to do is totally disable Wi-Fi I don't have ethernet plugged in and this is going to be running in the Chromium browser just a local host and you'll see exactly what I'm talking about here so everything's running natively on the Orin super so just head up here disable Wi-Fi completely we'll just turn this off and now what I'm going to do is open up chromium we're going to run the Local Host so it's basically just going to connect to itself and I could run this in terminal but this does look a little better in the browser and yeah this is a very sophisticated large language model it's basically a chat bot the way we've got it set up right now and I can choose the language model that I wanted to use which is llama 3.2 and now all we need to do is ask it something we can basically ask it anything we can ask it to summarize a movie but what I want to do is have this thing tell me a story so we'll do a story about a dragon who turned into a robot something like that now this will also handle Vision works really well and I did demo that on the original Orin basically the difference between that and the this is we've got that super mode up to 25 watts and of course we've got a little faster Ram but a much lower price point and a lot of people would be using this for robotics but of course these large language models are all the rage nowadays so I figured I'd go ahead and demo this with llama 3.2 and I'll tell you the first time I ran llama 3.2 on this just asking it a random question took about 20 seconds the first time but the next time and actually every other time that I went back to it it was almost instantaneously answering me so I'm not sure what was going on with that it might just be some caching going on but another thing to keep in mind is all of this is running from a Micro SD card with the Orin super you can add an m.2 if you wanted to and you'll get faster boot times faster load times and it might even help out with something like this but most of the responses that I've got here from llama 3.2 have been very sophisticated uh coming up pretty quickly here and all of this is running natively on the board so we're not connected online it's not hitting up a bigger database it's all internally on this SD card running on this arm-based SBC and messing around with these llms on this board have been pretty fun but I'd actually like to install Vision Works on this and kind of come back to it just see what kind of a boost we have with that super mode up to 25 watts we've got the Jetson version of Ubuntu 20.04 we've got our software Center here it's just going to make it easy to get in here and download new applications so if you wanted to go through download some games and emulators without hitting up terminal you could definitely do it there's actually a lot of great stuff in here but keep in mind some of this is kind of mixed in with the x86 stuff so for instance if I was to try to install Steam it just doesn't meet the dependencies because this is an arm-based operating system on an arm-based board this does have built-in Wi-Fi 6 and you could always use ethernet if you want to but yeah I mean with the newer updates here this thing is really Snappy so if we wanted to go over to the uh Jetson sight just head right over here and everything loads up really quickly check this out robotic development and yeah that's really where these boards are going to shine but uh again you know if you're not into that and you just wanted to get some usage out of this we're just going to head over to YouTube real quick I wanted to show you some video playback I want to jump into a little bit of photo editing here and of course we will test out some emulation with PS3 I know for a fact that these chips would handle 4K video playback just fine at 60 and I'm also sure it could probably handle 4K 120 with the power this thing puts out but the software isn't quite there again it's not made for this but at 10860 I mean you could have a pretty good time with something like this if you wanted to do some video playback natively or from your favorite website like YouTube Netflix HBO videos will be buttery smooth at that 1080 other software I love to use on this device is it's just an image editing software it's open source I mean think of Photoshop but totally free we'll just open up this right here and yeah I mean this is real time let's go select by color we'll do this white and we're going to actually replace it so we'll get those edges there we'll do a little more over here we'll just go with pure white over there and we can just export this directly to let's say the desktop lossless yeah we'll do lossless and we'll export so I'm not going to save the changes here and we should have it right on our desktop there we go so nothing special just replace a little bit here and there but yeah I mean full photo editing on a machine like this is totally possible and again from that software Center tons of different applications that we can download very easily without even using terminal but uh one thing that I was kind of excited about here was PS3 emulation on an ARM device like this so I've got rpcs3 right here this is the native ARM version of rpcs3 for Linux and it works pretty well on this machine I was pretty surprised by the performance I'm seeing now it's not going to run every single one of these games at full speed but there are a couple here that I tested that run great and I'll show you real quick if we go to config I've taken a lot of these settings down but under GPU you can see the Tegra Orin we do have access to openg and Vulcan but I'm going to be using Vulcan and my Default Resolution taking it down to 480 along with Shader quality to low vsync so I have gone through and tweaked a few things here and there to make this run better on this arm chip but we'll get right into Tekken 6 and I've already compiled the shaders so it just kind of needs to uh load the PPU modules but this should load right up into a game and I've got a controller connected this is Tekken 6 it's not a super hard PS3 game to run but to see it running so well on an arm chip in my opinion is pretty impressive when loading into a match or a stage for the first time uh it will compile shaders and you'll see it pop up down in the lower left hand corner kind of falls on its face initially but you know going through it kind of evens itself out set it that lower resolution and I've got basically all the settings at relax so it's not perfect emulation by any means and I'll tell you I even tried to go lower than this so basically anywhere from like 320 up to 720 is around the same kind of performance because we've got that Tegra GPU which does a decent job but this emulator does rely more on that CPU we've got lower clocks here than other boards on the market but even though they're lower clocks they are putting out some really great performance even something like ninja Guiden Sigma runs really well here but I did have a lot of latency with this game My Controller was kind of all over the place and by the way I did try an Xbox controller but since I'm on Linux I had to resort to using a different controller I usually just go with that X input but this is sdl input I'm not sure what's going on here but yeah I mean it's actually kind of hard to to play this game because the latency is pretty bad and I am using a wired connection so both of those games do run better than I thought they would but they're easier to run on basically any system I did want to take it up a bit and as you can see with skate three this is not doing a great job at all and it really comes down to CPU performance with this one if you take a look over on the le- hand side all of our cores are maxed out and I wasn't expecting it to run these games at full speed all the way but we did get a few that ran really well overall with the Nvidia jet Nano or in super I think this thing's putting out some really great performance especially at a decent price tag coming in at $249 and the original Orin I believe the price on that was like 550 might have been $4.99 either way this is coming in at a much lower price tag and we're getting better performance out of it so yeah this could definitely make sense to some people out there but that's going to wrap it up for my first look video I definitely want to mess around with this a little more and I will be running more games on this if I can get steam installed and have some real PC games I've been running on this I'll definitely make another video but until then I really appreciate you watching and if you're interested in learning a little more maybe pick one of these up I'll leave links in the description but that's it for this one like always thanks for watching