Transcription provided by Huntsville AI Transcribe
Welcome.
Um, so, uh, general stuff just to cover before we get going. Uh, we are, uh, as part of the AI Huntsville Task Force, we’re going to do a kind of a general public theme in September. Uh, I am the one signed up to get up and talking front of people. So yeah, um, I have a lot of practice at it at least.
Um, we’re going to do a basic walkthrough of AI tools that some people heard of.
Some people haven’t just a general, Hey, here’s, you want to know some strategy stuff out of whatever.
Here’s, here’s a tool that does that, you know, just kind of walking through, um, hopefully dropping some QR code based links to other tools.
So folks in the audience can kind of walk along as they’re, Hey, here hit the QR code, blow it up on your phone, kind of see how that works and stuff.
Um, the overall use case we’re thinking about is a fictional retail establishment.
Let’s say they want to do some marketing campaign for some upcoming sale. Everything from strategy, market analysis, using some of the research tools that are available, some of the text generation, image generation, video generation, a lot of that kinds of stuff. Um, just a quick band within an hour, and just kind of show, Hey, here’s what’s possible. You know, just kind of get people thinking about things.
Um, so there’s that also looking at July 9th.
We’re looking at a social, thinking about there’s a private room in Yellowhammer.
We’re probably going to try to grab if it’s available to be inside in July instead of, instead of the stovehouse. So the stovehouse turns into a stove house in July. Uh, so that’s it. I’m guessing we’re getting something on this court.
I’ll check in a second. I’m going to hand it over to Tom, who is going to talk about stuff. Not just, but a lot of things. Let me make you the presenter.
All right, so I can make you the host and then you can share screen. Yeah, let me share now. No, it’s not. No, I’m fine. I’ll try again. All right, I’m now the host. You’re now the host. Yeah, sure.
All right. Okay, let’s max the screen here.
Yeah, I’m trying to get it. I’ll be right back. All right. Yeah, he’s out of the room. I want to say thank you for inviting me. I used to run a meetup here in Huntsville from 2010 to 2016. I used to run the big data meetup.
And that’s why I first met Jay because he used to come to my meetup.
But eventually the meetup fees grew too much and I got too busy with work stuff. So I had to stop. Tonight though, I’m going to talk about passing certification exams related to artificial intelligence. And a lot of what I’m going to talk about is going to be generally applicable to all certification exams. But I will have a lot of specific comments on the NVIDIA exams since I just went to their exam program and took their exams. And one thing I will say about their exams is a lot probably depending on the exam, anywhere between 60 and 80% of the material is industry general information. Only about 20 to 40% of these exams is NVIDIA specific. And the other thing I’ll say is you don’t really have to get hands on dirty NVIDIA tools to pass these exams.
If you study the information on their website and other publicly available stuff or take a look at some of the prep classes, you can probably pass these exams without having to spend a lot of time doing hands on stuff. Because they’re general high level or understanding comprehension type of exams as opposed to this button does that exam. So the three I’m going to talk about are their GNI exam, their multimodal exam, and their infrastructure and operations exam, which you can think of as a chip hardware platform exam, basically a hardware exam to LLM exams. It’s really what it comes down to.
A little bit about me, this is the first time I’ve presented at this meetup.
Everything is my own personal opinion. Don’t blame my employer or anyone else and I don’t work for NVIDIA. So give you a heads up on that. I passed seven I certification exams recently. Three NVIDIA ones, three Oracle ones, one Arquitura one.
I’ve done a fair amount of AI stuff over the years.
I’ve also done a lot of other things. I’ve created AI applications in Java, Python, and some other stuff. I wrote a, I did a 1990s project on intelligent agents doing search that within published as an article back in 2000. I’ve also filed patent applications on AI topics, computer vision, all networks, self-directed applications, things like that. And I’ve got some AI presentations on YouTube. Yes, I’m a consultant for Oracle, but you can’t blame Oracle for anything I say today. I’m also an attorney.
I’m in the Virginia State Bar and the Pat Bar. And I taught graduate computer science classes for Virginia Tech and Harrisburg University for about 25 years. We’ve got a bunch of degrees. JD, a master of science, computer science, master of science, blockchain, Bitcoin is a passion of mine. And I got a couple graduate certificates from Stanford and I’m taking some AI and data science classes from the University of Colorado Board. So, all right. So let’s talk about the exams. Enough about me.
Let’s talk exams.
All right. So these are one-hour exams. They’re about 50 to 60 multiple choice questions, typical ABCD format.
Most of the questions are single answer only, but there are a few questions that are multiple answers and you got to identify.
Typically, it’s two. So you got to decide which two choices.
They don’t necessarily tell you it’s two. I say 50 to 60.
When I took these exams, it was 50 questions on the infrastructure exam and 50 questions on the Gen and ILM exam.
And I had 60 on the multimole.
I don’t know if that would be your experience.
I think they may do the experimental question thing. Sometimes added a few extra questions.
So don’t rely on it. Assume you’ll have at least 50 questions and you might get 60.
You can take it online for your own home.
It will be a live proctor. My guess is the proctor is only there for the first five minutes and then they just have the camera on you the rest of the time. But, you know, the proctor will ask you to show the camera around your room. Typically, they just want your desk to be clear. So I had a whole bunch of stuff on my desk. Moved it all off and they were happy. Fair amount overlap between the certification exams. The three that I’m going to talk about.
So I’ll talk about that in a minute. Here’s what the Gen and ILM exam looks like at a high level.
We are going to dive into all these exams in much more detail as we go on.
So I’m just giving you a high level first.
So machine learning is about 30%.
But again, it’s not hands on coding machine learning.
So you don’t actually have to write Python code or read Python code as part of this exam.
But it might be good to know the difference between NumPy and, for example, some other APIs like TF or something like that.
Have an idea as to what the libraries are used for might come up in a question.
But actually having to do anything or look at code is not going to talk about a question. Data analysis 14%, experimentation 22%, software, and I’ll dive into what these mean actually in a minute. But these are the five topics and all these numbers are straight from their website.
If you go through their website, you’ll see them. Now, the multimodal exam is almost the same except it’s got two extra topics.
So basically they took some percentages off of each of these and threw them into the two extra topics.
The two extra topics they added were a multimodal data, which makes obviously a lot of sense when you’re talking multimodal. And then they added a section on performance optimization. I am not sure what was going on, why this section is not also on the previous exam. Maybe they’re experimenting here and they’re eventually going to add it to the other exam. Because it would make sense, you know, performance optimization in both of them. But anyways, that’s the big difference between the two exams. I’ll talk more about the difference, but there’s a high degree of overlap. I actually took those two exams on the same day and I saw a lot of the same questions from these other topics on both exams. So it wasn’t complete overlap between the two. So I would say, you know, they probably have a test set of like 300 questions or something like that.
And so I got, you know, maybe 10 questions were on both exams, you know, just your random overlap.
AI infrastructure and operations is actually quite a bit different how they wrote that exam.
Its topics are all different, but really essentially I acknowledge you can treat that as like 38% of what’s on the other exams. But maybe a little simplified version of it, because they’re aiming this at hardware engineers. They’re not really aiming it at a software engineer. What is the other exams are more software engineer, data scientist type of thing. And then the AI infrastructure that’s completely towards hardware and networking, you know, it fit a band, all that kind of stuff, network cards. And then AI operations is really, you know, how do you manage and control all this stuff. It’s more the process of running the data center type of thing where you’ve got AI workloads. So that’s the three exams at a high level.
Now, before I go any further, questions, what do you want to know about this before I go further?
What did it cost?
Yes. Okay, so I believe that their actual prices, if you looked at the website, are $130 to $150.
However, they’re promoting these things, so they do offer discounts.
Okay. I got all mine at half off. It’s pretty typical for them to offer it half off if you attend one of their conferences. Okay. So if you, and you can get free passes to their conferences, if you just sign up and, you know, get the free virtual pass and you get a 50% discount to the conference, the certification exams.
I have seen them give the exams for free if you actually take the exam in person at a conference.
So I’ve seen that as well. And there may be other discounts that I’m just not aware of.
And I don’t work there, but I have gone to some of their conferences.
Yeah. There are a couple of admitted reps associated with Maxwell AI peripherally. So if we did want to go try to ask, hey, are there discounts available for our group?
That’s an avenue we can take if people are interested in trying to do the same thing.
Yeah. The federal rep for NVIDIA, the person who’s in charge of NVIDIA federal lives here in Huntsville. Right. And she also used to work with me at Oracle and also worked with me at IBM. They’re around, yeah. How much time did you spend, you know, learning and everything?
How much time did I spend on what? Yeah, how much level of effort did you? Yeah, that’s an interesting question. So I started thinking about taking these back in September when I first got an email from NVIDIA saying, hey, 50% off. I was like, oh, 50% off, I’ll think about that. And then I never got around to doing it. So in March, they had their annual GTC conference, which is their main hardware conference that’s in San Diego. I think it’s in March. It might be in February. But anyways, somewhere around there. I don’t know if you’re on able to establish a secure connection to Zoom. All right. Do you think I should switch my internet? I can do that. Here’s the guest.
If you want to try.
Yes, it’s dropped down twice. It’s dropped me once.
So I could switch over to my cell phone.
If you want to try that. Stream your phone. It’s not the first time I’ve. I think it was the last time we were here. I might have done something like that.
Yeah, no, it definitely dropped me here. All right.
But anyways, while we’re talking about that, we’re talking about level of effort. So yeah. So GTC, I think was either end of February and March somewhere around there. And I figured, oh, well, they’re going to give me a 50% discount again.
Kind of virtual pass. And I was like, yeah, they’ll probably do it. So I decided what I would do is that time I decided I would go ahead and plan ahead. So there’s the three exams.
The infrastructure and operations exam, which was a hardware one actually has, and I’ll get into the study guides in a little bit, but in the study guide, it actually lists a course.
It’s supposed to map into the topics.
So I took that course.
And that course was free at the time. I don’t know if it’s still free, but it’s probably like 50 bucks if it’s not free. And that course was, I don’t know, 15 hours or something of really boring lecture. But I went through the course, I took the exam, passed the exam, but realized the course didn’t really prepare me for the exam. And I’ll talk about the good and the bad parts of that course and what would help prepare you more than that course did when we get to that topic. But basically, so for that exam, I probably put in 15 hours or so watching the course and read a few references.
The other two exams, the general and the multimodal, I came to a couple conclusions in the study guide. I’ll talk about the study guides in a little bit, but they have a number of references.
And then a friend of mine I was also studying with, my friend is a big fan of Udemy. So we took a look out on Udemy and there’s a lot of people put up their own little prep courses for like 10 bucks or 20 bucks. And so I took a couple of the courses that were out there. I think I spent about 40 bucks. One of the courses was like 14 or 15 hours. I did not spend 15 hours. I maybe spent 10 on it. The other one was like five and I probably spent about three on that one, three hours. I’ll talk about them in a little bit. But basically, I probably put about 15 hours into those two courses, studying for the two LM things and 15 hours or so into the infrastructure one.
You know, I did reasonable on the infrastructure one and I got 95% on the LM ones.
So I think I overkilled the amount of study I had to do. You might be able to get by with less. It just depends. And I actually took, I took both the LM and the multimodal the same day. I did take the infrastructure one in March. The other thing is when you sign up, you’ve got like two months from when you sign up to actually take it. So I had to like pay for the exams in March during the conference, but I actually took the two LM exams in May. Two months later, the very last day was due. I was like, fine, I’ll sit down and take them. You’ll procrastinate. All right. What are you doing here?
We’re trying to connect to you.
Sometimes you just go to the website, act like you’re looking for something. And then within a day, you’ll get an email with like half price or a $10 course or whatever.
That’s $9.99 for every video course I want to take.
I was like, gee, okay.
So pro tip, don’t buy the thing the first day you see it. They know you were there. They’ll send you an email. Check back. All right. I am connected into my home cell phone. I was going back in the same year. I wonder how yellow hemorrhage Wi-Fi is. If they have.
So what are you doing?
You know, I always use my own phone.
I don’t actually use it. I don’t. So.
It’s.
Short. Because it would break the balance. It’s a different kind of. Any seven instance.
It’s.
It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s. It’s gotta be. Oh.
It’s.
It’s. It’s gonna chew it. It’s going to stop. So that means. I got you this thing.
It’s.
It’s. It’s. It’s. It’s. I’ve been. A thousand dollars. There. Time.
A lot of.
Try to work. You can. You can do stuff right now. All right. So students can come in and sit down and chat and talk about what you have in your mind. See that kind of basic thing around your head. Light it or not. So yeah. I would like to get the guy to work with you. Keep you saying it a lot though. That’s a cure for like L1A stuff. Using it on your meetings. That’s kind of interesting. I just want to get you. Graphical.
So it’s all like that.
On this project as well.
There’s a… What do you use?
What do you use? What are your skills? What does the… what are your social media platforms? I hope anything sits down. It’s… Right.
It’s… I don’t know what it is. Yeah. I’d like to… How can you keep that salad head on?
There’s all this PlayStation.
A lot of great stuff.
some way to go back out and clotting to the rhythm and to develop something, right? Okay, I don’t want to enjoy this. Sorry, so we answered the question about level backwards. Probably… are we listening again? No, I’m still here.
You have the detector. This is your… Zoom is not being in front of you today. So we answered the question about level of effort. Probably I would say 10 to 15 hours, for example, although those last to the LAM, regular LAM, you could combine them together. My answer would vary as to how much effort, and how much you already know this stuff and don’t. Other questions? Talk about price, like I said, it’s like 130 to 150, but I’ve got it for that price.
If you just go to one of their conferences, or even a virtual attendee, you’re probably going to have price discount. Then they may not email it to everyone who attends, but they usually do these certification little prep things, and if you go to the certification prep thing, or you watch the video, they’ll give out the code during the video. And pro tip, they’re actually doing one of those tomorrow, for everybody globally. So if you watch that one, I’m sure they’re going to give the discount tip on that one tomorrow. So they knew I was doing this session here today, and they follow us, I’m sure.
They call their boss top. Your boss. Just like Udemy, they follow Udemy approach. All right, so what other questions do you have before I go further? So for the AI operations and infrastructure, what’s the kind of scale that they targeted, low end, high end, as far as like operation?
Yes, I got questions on everything from your personal computer with single GPU to H100s in their big NVIDIA Super X, whatever they call it, data platform, whatever. Yes, the whole scale of NVIDIA products, really. Now that being said, they’re primarily focused on AI workloads. So they didn’t really ask anything about, if you saw an answer that was a gaming card related, it was the wrong answer. So what’s the most useful thing you learned and all that? The most useful thing I learned in going through this process was that I really knew a lot of this stuff already. I guess I could.
So it’s not to be scared of it.
And that, you know, I mean, a lot of it is… Now that being said, yeah, obviously AI is a huge space and there’s more stuff being added all the time, more buzzwords being created, more platforms being created. There’s obviously, they’re going to ask some questions to which you won’t know the answer. And that was when I went through the infrastructure operations, I was actually surprised at how many questions they had that I didn’t know the answer. I still passed, but I was surprised on that one. And so I did spend a little more time studying and correcting for the other two before I took them. Another thing is that I figured out the pattern in their exams, which I’ll talk about in a minute. So other questions.
All right, so I think that’s the end of my prepared slides. So what I want to do is now move over to NVIDIA’s materials and walk us through. Yeah, before I get to the NVIDIA’s materials, let me talk about what you can use as resources for studying for the exams. So next up is I’m going to talk about these study guides, because they do have study guides for each exam on their website. So I’ll talk about the study guides. But there are obviously a lot of other places you can go to study for the exams.
You know, the NVIDIA website’s got a ton of information.
NVIDIA has GitHub sites out there that support all their different APIs and so forth and products. There’s blog posts, they’ve got YouTube channels. Third parties have YouTube channels about using this tech. There’s a lot of places out there where you can learn about this stuff.
And again, generally, eye sources are going to help you.
You know, attention is all you need, the transformer, paper, all these other things that we’ve been talking about over the weeks, and this beat-up and other beat-ups. It’s all good stuff. And then there’s obviously third-party study sites. People have put blog posts about studying for these exams. There’s the UDME courses and so forth. And I’ll actually show you a couple of courses I took.
I do have internet for a moment.
I’m a long kid, too. Here’s what you get for nine bucks. Now, I’ll give you a hint.
Both of these courses were not originally written for NVIDIA exams.
It looked like the people who wrote them were AWS instructors who repurposed a lot of their AWS machine learning materials or AWS GNI materials and called it an NVIDIA course.
And in fact, you’ll even hear them say that during one of the courses. Oh, yeah. And as part of your AWS machine learning prep, the other course guy was a little better. He said, oh, it was part of your NVIDIA stuff. But he only does that every so often. So you know he’s cutting past everything else. So, all right.
So let’s take a look at the study guys.
All right.
I am going to start with, I think, the GNI LM1. All right. This is one of the study guys looks like.
It’s a basic, you know, eight page PDF or something like that. It talks about the weights. We already talked about, you know, in this case, it’s five topics.
We already talked about the weights. It gives you a page job description, which I’m going to skip over. It gives you a, then it dives into each of those five topics.
It has a page on each one.
Now, the other multimodal exam will follow the same approach, but then it’s got seven topics.
It’s got a page on each one.
The infrastructure and operations only has three topics.
So you get the idea.
It’s got a page on each one. All right. So what do they got on here?
So they got some ideas of things you should have, familiarity, and so forth. And then they’ve got some objectives and a list actually mapping this into NVIDIA’s courses.
I did not take any of these LM courses from NVIDIA, but that’s how they map in if you’re interested in taking their courses.
And they have both, you know, like, look, all the course providers, they have both online and person and all the different options, whatever you want to pay for basically. And then down below, they have suggested readings. So for example, the first one they put there is attention is all you need, the transformer paper, but they also got stuff like, you know, end to end for NVIDIA based PCs, Gen AI, what is it not as a word, blah, blah, blah. I don’t understand that. Model, Laura, low rank adaptation, diffusion based models, and so on.
All right.
So here’s our first pro tip on what I noticed after taking these three exams.
So if I was going to suggest how they built their certification exam is they gave the person who was creating the certification exam a list of all these suggested readings and told them put in a question from every reading. Because I literally, like there’s, you know, between these five topics, there’s probably about 30, 40 references.
And I think I saw a question from every single reading in those 50 questions.
So that’s not all 50 questions.
That’s 30 to 40 questions are covered in the readings.
And usually they didn’t go too deep into the reading.
So we’re not talking about page 17 and the transformer attention is all you need. We’re talking page one or page two. So that’s first pro tip. Read the first page or two for all these little references.
And there’s only 30 of them.
So actually on the day of the exam, I glanced through for like half an hour at the references right before taking the LM exam.
And, you know, I was just thinking, well, this is what I’ll do in my last half hour of breath.
And then I literally saw every single reference popping up in the questions.
So I was taking a multimodal exam in the afternoon.
I took the LM in the morning. So I said, okay, I’m going to spend the next hour before I take the afternoon reading through all the multimodal references.
And again, I saw the same thing.
Now, I did not see that pattern of the infrastructure operations exam. Maybe that’s not the case.
Maybe I just wasn’t paying enough attention at the time. So, but definitely this just in readings are really what I would focus on in terms of studying. And none of the courses I took told me to do that. So you’re getting here first.
Those are the free parts.
And these are all free too. Yeah. Well, I mean, some of these you might have to, if you don’t have a research thing, you know, some of them might be in one of those archive sites where you have to go in through using your research, but most of them are all free.
You know, the tension is all you need is all over the place. And most of the other ones are free. So if we go into like data analysis, it’s pretty much not, it doesn’t have quite as many references.
So some of them only have a couple of references mentioned.
Some of them are the same records.
So earlier.
So it’s not 30 or 40 references for the exam.
It’s probably like 20.
But you can see now we’re starting to see more of the NVIDIA specific stuff.
Rapids is an NVIDIA specific API.
CUML is a specific NVIDIA flavor machine learning API.
You know, and they get another article on using rapids for accelerating data science. But again, it’s a general SK learn package.
So again, it just depends.
You know, a lot of this stuff is general machine learning or general LL information. Here is the reading list for the next topic.
You know, how to conduct AB testing, very general stuff, not specific to NVIDIA.
Inference optimization, zero shot testing, again, very specific.
And zero shots, very specific to LLMs. But you know, again, it’s not NVIDIA specific.
In fact, everything on here is not specific to NVIDIA.
Billion transformer based natural language, cross validation, machine learning, RAG.
So a fair amount of questions on RAG. Which people are made a generation?
How many of the questions did you run into that were more on the evaluation side of knowing whether a model was working correctly?
I saw a couple of references above and a lot of the courses guide you right through building something, but then they stop.
They don’t actually tell you how to evaluate whether the thing you build actually is working for.
Yeah. So I had a fair number of questions around accuracy and precision and recall and F1 scores. So that’s sort of evaluation. Okay.
That’s in there as part of the machine learning stuff. So a few other questions about, like, let’s say you’ve already decided it’s not working. How would you change your architecture to get a better result? Okay.
So that sort of problem statement, yeah, here’s the situation. What can we do to improve the situation? Okay. There’s a fair amount of that.
That’s good.
Here we got TensorRT, which is, again, this is it.
We’ll get back to some NVIDIA products, NVIDIA NEMO, customized NLMs and NVIDIA GPUs, TensorRT again.
Fair number of questions on TensorRT.
That was probably the NVIDIA product that came up the post often.
Stuff about distributed deep learning, deep learning scaling. Burr came up a fair amount of time. Again, that’s not NVIDIA specific.
The trustworthy AI is really their ethics section. And these questions are pretty straightforward, like most ethics exams are. So just think ethically and you’ll get 10% right.
These are your easiest 10% on the exam. And it’s 5% of the multimodal exam.
Basically, you only got three references.
What is trustworthy AI?
What is RAG?
And trustworthy AI for the federal world.
Really, just what is the trust? How do we ethical?
You’ll get some questions like, you know, if you’re building an AI system for a government customer, blah, blah, blah. So that was the AILM study guide. I’m going to pop up another study guide now.
Let’s take a look at the GNI multimodal exam.
And I’m going to skip over all of it, except for the two topics that are new here.
So now that being said, there are probably some multimodal questions buried in those other five topics where they’ve added a few multimodal specific questions in there. But those resources you study are pretty much the same. So performance optimization, they got precision training, fixed precision. There was a fair number of questions on that. Quantization in machine learning, that came up. Neural network pruning, energy efficiency. Again, like I said, every single topic, there is going to be at least one question of the references. So that’s your best place to study in terms of, you know, maximizing your benefit for a minimum amount of time.
And then let’s see where to call it.
Here it is.
The multimodal data clip is an API for connecting texts and images.
Again, it’s not NVIDIA specific, but it’s a big part of this exam on the multimodal data side. A number of questions on it, really.
Data fusion was the other place.
So a number of different questions dealing with different aspects of data fusion.
And a little bit of the monotony of text portal encoders and multimodal data fusion.
Those were the main topics I saw clip and data fusion. I saw a number of questions on both of those topics on this particular multimodal exam.
All right, let’s go take a look at the hardware exam.
The infrastructure exam study guide.
See, job description is a little bit different, but really the exam is about the same.
All right, now the exam does break it out a little bit. So we got systems and networks. We got 17%.
So here are the references are, you know, GPU software, InfiniBan, command line interface with NGC, cables, NVCC, the QMCC, power driver, some system management interface, and system management user guide.
Now for this exam, I did not do all the readings.
I took this one first. I hadn’t gotten the pattern yet. So that may be why I was upset with my score on this exam, even though I passed, you know, it just wasn’t 95%. I was like, eh, what’s going on? So definitely I saw questions on InfiniBan.
In fact, I saw questions on all those topics.
I know it. So the references are key here.
The troubleshooting and optimizations.
Yeah, again, we get rapids, pandas, acceleration, back decks, DCGM, GPU manager, H100, H200 there.
We’ll field DPUs in the data processing units.
And the videos. Yeah. So you’re going to have questions on all these topics, I think, on the infrastructure operations exam. Let’s see what else. They may not mention it here, but you’ll also have some questions on the chips themselves. So, you know, even though there’s not a reference in here, if you will want to read about the chips for the hardware exam.
All right.
So here we’ve got a bunch of things for systems and servers.
We’ve got CUDA compiler drivers again, and that’s the same reference from the previous page.
So, DGX software utilities, DPU, user guide again, a lot of the stuff on power and cooling, the A100 system, H100 system. They also have the Blackwell now.
So, but I don’t think that’s on the exam yet. DGX super pod, NVV and ME, then app design guide and confidential computing that was on the exam as well.
All right. Yes. How small of a system are they?
Looks like you’d actually have an event knowledge.
So again, how small of a system?
So are we just like like three or four servers that are actually expected to be connected in a band?
For the exam, they’re not really, you know, it’s general exam questions.
So they’re not really asking you, like, the architecture system or anything like that. They’re asking you questions like, you know, what would be the best way to handle this workload? And they’ll propose you could run it on your PC. You could run it on a Blackwell, you know, system from Nvidia. And you get to decide.
They’re not super hard questions in terms of, now you might, on the hardware exam, you might get a question asking like, does this network card support eight systems or four systems?
It is possible you could get that question.
Well, you can, I think, pass these exams if you get 60%, right?
So even if you get one wrong here or there, it’s not a big deal.
Because you don’t remember whether it’s the card supports eight or four or 16. And certainly I saw a few questions like that. I was like, really, you expect me to remember that?
I looked that up if I need that in the workplace.
So, let’s see.
What else are we going to do?
All right, physical layer management.
We’ve got AI Enterprise, a lot of AI Enterprise stuff, data center drivers, multi-instance GPU users.
So, you know, you’re virtualizing the GPU to let other people use it. You know, what are the best practices for that?
Deep learning frameworks, Ethernet interface, number of questions about Ethernet versus Venom Man, various user guides, CLI installation, number of questions about CLI.
All right, so that was the study guide for the hardware and infrastructure. Yeah, I went through these pretty fast, just to give you kind of an idea of what the exams were like.
Again, I think the key, if you’re going to take one of these, is primary, is I would focus on reading those references.
I think they’ll be your best preparation.
You could take the courses from NVIDIA.
Again, they do offer sales from time to time.
When I took their hardware infrastructure course, it was free.
I think it’s 50 bucks right now, so it’s not too expensive. That being said, I thought it was done by an AI instructor. I was not happy with it. Yeah, it was that guy who’s got the monotone voice and just wants to put you to sleep. That may also be why I didn’t do as well. Let me talk about the identity classes.
Let me see if I can log in here.
Where did they go?
Try this.
While you’re doing that, I realize you didn’t fail any of the tests, but if you did, was there a retake option or a new fee? You know, I don’t know. I don’t know if they asked me for money or not if I’m trying to retake. That’s a good question. I read up on it, but I think the only way you can retake is if they kick you out versus if they kick you out. If they kick you out, then they have a little clause saying you might not be able to retake if you’re trying to cheat on the exam. Yeah, but they might charge you to take it again. They might charge you another 60 bucks or whatever, but you know, so… Let’s see. Okay. Now you say this, but I really look… I value you 60 dollars.
All right. So this is what Udemy looks like. And in particular, I’ve got one of my two exams I took. Again, both of these exams were written by guys who are an AWS instructor, primarily AWS machine learning related types of classes, who decided to branch out to do an NVIDIA for more money. All right, so what this instructor did is he’s got over here like six hours of machine learning fundamentals and three hours of deep learning fundamentals. That’s all of his AWS machine learning stuff. He just reprinted and worked over here. So that’s 10 hours of AWS machine learning.
It’s more detailed than you actually need for this exam, but if you want to review machine learning, he’s actually pretty good at teaching machine learning.
I’ve seen university professors who are not as good as him. So I will say that.
And you can tell some of these I’ve actually taken, some of these I haven’t taken, like I didn’t take his NLP or his large language model stuff, but he did actually go through most of his deep learning and most of his machine learning. Now, the only problem with this particular exam, on this particular course, it’s NVIDIA certified associate JLMS. I’ll send out an email later with the link to this one.
But the only problem with this one is that, yeah, it’s got a decent number of ratings over here.
5,000 students took it, 4.5 ratings. I don’t know if that’s all for this course or if that’s for other courses.
He’s got a lot of courses on NVIDIA.
But again, this is very not specific to NVIDIA. This is more general machine learning.
So if you’re going to do this, I’d also definitely read out to two that are more specific to the questions.
He does have a few comments down here about the NVIDIA exam, but he’s only got a few comments. It’s almost all the AWS machine learning class. Now, the other class I took, let’s see if I can bring that one up.
Go to my learning.
Yeah, so my learning is these two courses. This is the one we just looked at from Manifold AI Learning.
Oh, he does have a GitHub site, so you can get access to all his AWS machine learning, GitHub, Python code and stuff.
The other one is this guy, Andrew Larkin.
This one is a bit more specific, but again, he still also is pulling some stuff from the AWS. But you can tell he’s a bit more specific to NVIDIA. He has got some really nice videos showing NVIDIA stuff on his slides, whereas the other guy didn’t have any NVIDIA specific slides. At least he went a little further in the packaging, showing the NVIDIA GVU cloud and CUDA libraries and virtual GVU infrastructure and so forth. But again, he also has his generic stuff, like his new computing stuff, which I’m sure he’s used 17 different classes, and his basic L and intro to AI and intro to gen AI is all very specific to every class he’s offering. It’s only really section five on NVIDIA and the multimodal stuff, which is specific to this class. There is actually several different classes on the basic LLM exam, offered by instructors who I think are repurposed to materials. This was the only class I could find on multimodal at the time. Now, it could be there’s more classes now.
I took this like months ago.
They’re both good. Both these instructors, they said, are pretty good. They’re competing in the marketplace, and they’re better than a lot of university professors. Again, university professors are tenure, so they may not figure their students be entertained. University professors may drone on, and I was one for 25 years, so I can say that. So, questions, comments, what would you like to know?
Did you write the courses?
I did not, but if I was going to write the courses, I would start off by saying, read all the readings.
How many did you rate the courses?
I’m not sure if I rated these courses.
At five stars, what would you give them?
Given the fact that I spent nine bucks on this one, and I spent maybe 15 bucks on the other one, I’d probably rate this one at five stars, and the other one at four and a half stars. This one actually has, they’re more entertaining. It’s not perfect, but it’s more entertaining. Also, I’m a pretty easy career.
I usually give students age. But yeah, courses, that’s worth the time. I did not go through every single lecture and their stuff.
I kind of picked and changed. Neither of them are the type of thing I would repeat.
Just to give you a contrast, I took online classes from Stanford for several years, and when I took Stanford’s Bitcoin class, I think I watched the lectures for that class eight times.
None of these are things I would watch one at once.
And we’ll show you for all prep time between reading the classes.
I would say that probably the infrastructure and operations, like I said, probably about 20 hours, and the GNI and the multiple rule was probably about 20 hours.
I read fast though. Not everybody reads as fast as I do.
Also, in those papers, I did not, like I said, I didn’t read all the papers.
So, for example, I read the first couple of pages of paper, and I would skim down the conclusion, but I would ignore all the math in the middle. Something I learned as a computer science professor is too many other computer science professors write math.
They wanted to be a math professor, and the math department didn’t accept them.
Trying to prove they’re better at math than math professors? Don’t put math by computer science. So you’ve done NVIDIA, you’ve done the Oracle ones, and the other ones. What’s your recommendation on which ones to do and why?
All right.
Yeah, so, and actually, you know, since I took these classes from guys who normally teach AWS machine learning, I can probably tell you about AWS machine learning too. I would take an exam, yeah, but I think I’m prepped for it. So I took the NVIDIA ones because a lot of people are looking at NVIDIA as a hardware platform.
Even if they’re not going to use NVIDIA’s software, they’re planning to use OpenAI or somebody else’s model, or they’re using Anthropic or whatever they’re using.
A lot of people are using NVIDIA as their hardware platform.
Not everybody, a lot of people are. So that’s why I took the NVIDIA exams, and I started off with a hardware exam for that reason, but I said, oh, I just took this exam, I may as well take the software ones too.
All right, so that was my reasoning. I took the Oracle exams because I worked there and they were free. I took the Arcatero one because I used to work there and it was free. So my feeling is if a certification exam is free, I should go ahead and take it because then I’m not spending money and it’s something like a resume, blah, blah, blah. And, you know, the reason behind certification exams is if someone already knows you, they’re going to hire you based on knowing you. They’re not going to make an evaluation based on your skill set.
They’re not going to read your resume, they’re just going to hire you. But if they don’t know you, a certification exam is an external attestation that you know this particular skill. And so it doesn’t hurt.
I’ve taken like over 40 of them in my career, probably, maybe more. So many have lost track. And part of that is also because, you know, I taught a lot of classes. And so sometimes I’d talk, you know, I’d talk certification exams related classes. I was teaching to show approved people. Yeah, he actually does know what he’s talking about. So for the Oracle ones, again, we could actually do a presentation on that some other time.
I would say that they are a mix.
There’s somewhat similar to the NVIDIA ones in that, you know, there’s a fair amount of general machine learning, general LLM materials in there.
But the focus on those exams is more about how do you do LLM in the Oracle Cloud?
How do you do, you know, AI workloads in the Oracle Cloud?
How do you use NVIDIA GPUs for workloads in the Oracle Cloud?
That sort of approach. Slightly different spin.
But they’re also very good if you’re interested in doing that.
I would, and again, I haven’t taken like, you know, the AWS for Microsoft exams, but I would imagine they’re kind of similar to the Oracle.
So in terms of recommendation, I think it all depends on your personal situation, like what you’re trying to achieve and what your time investment is.
I’m kind of a forever student. I’m always taking classes. So even if I’m teaching, I’m taking classes. So I think learning is good and continually learning, you know, technology changes rapidly. So these are your certifications are good for two years or, I think, I know I saw one of their, you know, I started clicking through some links.
Yeah. Yeah, that’s a good question. I think it is two years. I’m not 100% sure.
I’m pretty sure the Oracle ones are two years as well.
Arcaterra is still good after, and I took the Arcaterra exam like six years ago.
And I still have not wanted to be in SA to come take it again.
And I know eventually they won’t want me to take it again, but have nothing to do with it. The NVIDIA exams and the Arcaterra exams are up on Predleaf.
Microsoft and Oracle currently don’t have a good relationship with Predleaf, so you have to kind of do a workaround to get those exams up on Predleaf.
They used to.
I think Predleaf’s charging them too much money. We were complaining earlier today about need of charging money. I think Predleaf charges a lot. Has anybody taken any of the other AI certification exams out there? I’ve taken some courses, but not, especially a lot of the exams will offer up some free stuff leading up to the exam, to the cert. And for me, the certification doesn’t buy me as much, but the material is useful. Oh yeah, material is always useful.
If it’s there and I can learn, I usually hit that. But then usually I’ll bail out for the whole, you know, certification thing. It doesn’t help me that much to have it. If I get into a spot in my career where it would help me, I’m sure I’d go real quick, go get some certs. The infrastructure guys, it’s really important. Yeah. Just that’s the culture of those sorts of folks. Yeah. Microsoft really, you know, in the early days, you know, 90s really hammered on the certification lifestyle.
I may have had a MSDN certification or something. I don’t know if that was the name of the thing. That was one of them, yeah. Back when you got your documentation on a set of CDs. Windows AT and all that. Yeah.
Way back. This was good. I will send out the links to the two courses I checked.
And the other materials.
The recording will be up and I will probably at some point actually do my own version of one of these courses, but I’ll just put it for free on YouTube.
I’ll actually put it on $9 for you Demi. Although those $9 add up. It’s like 5,000 ratings. So, you know, 5,000 times 9 bucks is $45,000. I’ve had some people tell me some of the stuff that I’ve been doing for, you know, I’ve got an intro to AI course.
I have probably gone through 20 to 30 times. You know, I mean, I can throw that on video and go through pretty quick.
I don’t know how useful it is to hear my perspective on something that everybody else already has a course out for.
All right.
I will out myself. I’ll show you my YouTube channel. Hold up here. All right. Let’s see.
It’s right here.
All right. That’s the area.
All right. So this is my YouTube channel. Recently I’ve been doing some machine learning stuff and playing around with recurrent neural networks, and I’ve been doing some work to analyze patents. Is that also patents, right? So it’s looking at a data set of cannabis patents, you know, for cannabis edibles. And another data set for quantum computing. Because I long time ago taught a classic quantum computing as a science professor. So I did quantum computing. The quantum computing one was doing long short term memory, recurrent neural networks. And I also threw up some prep when I was practicing for this. I did a little video for each one.
But shorter, you know, probably like 10 minutes. I need one of these.
So practice on linear regression and stuff.
I actually got 88 views. I was not expecting that. You know, 100 views at some of these. Yeah. And I just threw them up. I’ve got about, what, 1,400 subscribers?
And I put this channel up about five years ago. Now, in the early days, I cheated on getting subscribers.
Students? Yes.
Well, I didn’t tell them they had to subscribe. I just said, hey, by the way, if you miss class, I’ll have a YouTube version of the class.
Watch the YouTube version. You don’t have to be here. If you don’t want to, if you’ve got a conflict, just watch the YouTube version. And if you want to, feel free to subscribe. So what else I’ve got on here is like playlists. So I see, I’ve got to have it as a playlist.
Like I said, I did a lot of cryptocurrency stuff.
So I got a Bitcoin playlist.
I’ve got a crypto wall playlist, an Ethereum playlist, a software architecture, machine learning, and a full list of criminal books.
That’s pretty cool.
All right. That was it. I will read the whole post real quick. Share. Good. Yep.
Just get back to this. Finally to the meeting here. Well, that’s pretty much it. Let’s tell Tom.
Tom, thanks for.
We surely appreciate it. I definitely do.
Are we getting the online question? Did we give any online questions or people?
Let me check.
Thank you for that. Yeah.
I was sharing that. We had one, we had Nate. It was talking about. He also passed the data bricks, analysts, associates test with you to me as well. That was really it. McKayla, Nate, Blake, y’all have any, any other questions? Hey, I don’t have any questions, but thanks Tom. That was great. Okay. Sure.
All right.
Well, let me go ahead and stop recording. Good job. Thank you.