Next Generation technologies making new workflow paradigms possible

SCOTT HAMILTON: Good afternoon, everybody. My name is Scott Hamilton. I'm with Dell, and this is Himanshu from NVIDIA. We're going to be your presenters this evening. And we'd like to make this interactive, so if you guys have questions during the presentation, feel free to raise your hand and ask a question if you want.

What we're going to cover is next generation workflow. So we obviously both work for hardware companies. And really, our main job, we focus on industry strategy, market development for engineering and manufacturing industry. And we're constantly looking at different trends in the industry and trying to figure out how, as hardware manufacturers, we adapt to those trends and help deliver the technology that's required to enhance and really allow users like you to get the best out of your hardware and your software. And so that's really kind of the subject matter of our talk today.

So with that in mind as we're talking to customers, as we go to events like this, there's a number of trends that we're seeing that I just wanted to talk about. And one of the biggest ones we've seen in the last few years is this advancements in the simulation and analysis tools out there. More and more of the tools are becoming actual replacements for older technology like crashing prototype cars into walls to figure out how they're going to work or not work. You can simulate all that on a computer now.

And so it's opening all these doors. And as hardware manufacturers, we're trying to help make that process be more efficient. So as we deliver products to market, we develop these products that have very sophisticated compute resources, whether that be CPU or GPU, to enable those simulations to happen in a more realistic way and also much, much faster. We're getting to the point where we're able to do some of that in real time, which is incredible. And we'll get into more details on that.

The next one is optimization. This one I find is really, really interesting. I think Autodesk calls it generative design. But I think the general area is called optimization. At least when I was in engineering school, as an aircraft engineer, it was called optimization. It was all about how do you optimize your design to get the best performance out of it.

And when I was in school, it was like you could do three variables at a time and that was about it. Now you can do you thousands and thousands of those. But when you do that, you're basically running a simulation thousands of times.

So the compute resources required to do this optimization work is just off the charts. And that's another area that we closely look at. How do we deliver the technology to enable that optimization process?

Multi-user collaboration is another big one that's happened. COVID has just really ramped that up quite a bit. We saw in our products, our mobile products were growing faster than our fixed or tower products, if you will, before the pandemic. But when the pandemic happened, our mobile product sales just went through the roof because everybody had to be at home because they couldn't work in the office.

So that trend really got pushed to the limit during the COVID. But I think now, even as people are starting to get back into the office, I don't ever see it going back to-- not everybody is going to be back in the office all the time. It's going to be this hybrid working environment, where sometimes you're at work, sometimes you're at home.

And now, rather than just doing one thing at a time, you're doing three things at a time. And you're also on a Zoom meeting at the same time. And it's just putting all these extra demands on the hardware that you use. So again, that's another trend we're seeing.

AR and VR, that kind of took a dip for COVID. I remember you used to go to every trade show of Mind True, and everybody had a VR headset in their booth. And you could go check out all this stuff. And then nobody wanted to touch that for the last few years for obvious reasons. But I hope we're going to get back to that.

And the technology is going to become more popular because-- how many people in the room have tried a VR headset? I assume most people have. Yeah, most people have.

Well, if you haven't, I would highly encourage you to try it out. I don't know. I've seen a couple on the show floor, not a lot.

HIMANSHU IYER: Quite a few, yeah.

SCOTT HAMILTON: So if you haven't done it, I would really encourage you to do it. Because I can't explain the experience to you. You just have to try it. Because it's just amazing how real it is and how incredible the experience is to visualize a product or experience something in VR. In many ways, it seems very real.

And then integrated design, this is where we're-- it used to be years and years ago when I first started being an engineer, was on the cusp of this. But it used to be a lot of things were either mechanical or electrical or they were software. And now almost every product now is mechanical, electrical, and software altogether.

You look at a car nowadays. What do they have, 50 computers in them? I have a Tesla. And it's all about the software now.

So the way that we design has completely changed. Because now you're not just worrying about mechanical or electrical or software. They're all together. And being able to do that in an integrated way becomes a very, very important process. And this is where Himanshu is going to talk a little bit about digital twins and how you can enable that.

And then the last one, artificial intelligence. This is changing the world and allowing us to do things like generative design. And being able to do, for example, photorealistic rendering now, you can speed that up by rendering partially and then using artificial intelligence to do denoising of an image to make it look crisper before the rendering is even really finished. And that accelerates the process.

So there's a lot of exciting trends. And that's why I love coming to these shows because there's all this cool technology. And we all get to learn about it.

So just kind of looking at historically, this is how we look at the evolution. Wave 1 was what we used to call destination computing, maybe 20 years ago. Everybody went to the office, you sat at a big workstation, and you did your work. Because that's the only place you could do it.

And then we started getting into the realm of Wave 2, which is portable. Probably realistically, about 10, 12 years ago, it really became practical, where you see a lot of people could use a mobile computer to do professional engineering or design work. And that's only gotten better and better. And I think most people use more mobile technology than they probably do fixed or tower technology.

But now, we're getting into this Wave 3, which I call immersive. And this is when talking about things like AR and VR, real time simulation where you can see the results immediately, or digital twins, where you can simulate how a part looks, how it functions, how people would react to it, how the part reacts to its environment, is really bringing us into this immersive design paradigm. And with that, I believe I'm going to hand it over to Himanshu.

HIMANSHU IYER: Thank you, Scott. So my name is Himanshu Iyer, as Scott mentioned. I am the Industry Manager for Manufacturing at NVIDIA. And we are here with Dell. Thank you all for coming to the session.

Before I start with this, I just wanted to get a feel here that people, attendance here, industry-wise what do you identify with? Are you primarily from manufacturing industry? Or if you can get a raise of hand, manufacturing product development, who does that as their regular workflows? Few people.

What about AEC? Are there more AEC folks here? And any media and entertainment industry folks? No. So mainly manufacturing and AEC.

My presentation is primarily focused on the manufacturing industry. But I will cover and touch upon some AEC points as well. So Scott mentioned, Scott covered some of these earlier, how work is changing, how COVID has played a big role in changing our work and such. But if you look at the complexity, the data sets that we are working with, they are increasingly getting bigger.

And they are getting more complicated. It's not just mechanical anymore. It's electrical, it's software, it's electronics. And all of that is coming together in products. And when it comes to AEC and such, piping becomes very critical. HVAC type of systems become very critical. And all of these need to be modeled. Our customers are expecting these to be modeled in photorealistic way, ray-traced, realistic feel of these data sets.

So the data sets are growing. And same thing on the M&E side. So if you see the characters, the CGI that is being done in movies, it's really mind blowing how high fidelity and true to life this is.

When it comes to multi-application workflows, it's very rare that you are using just one application on your computer. You have multiple applications for design, maybe simulation if you are doing rendering, along with all the other apps that you are using, email, PowerPoint, Team, or Zoom, or whatever.

So we are using these multiple applications all the time. And all of these need computational resources. And it taxes your CPU and GPU.

And lastly, collaboration, as Scott mentioned. Last 2, 2 and 1/2 years, this has really accelerated a lot. Because we are working from home most of the time. We are working remotely. We are collaborating with our team members within the country internationally. So collaboration is really critical.

And all of these areas have an impact on the resources, the graphics, the GPUs that your computers have or make your workflows smoother. So this is the reason why I asked about trees of hands earlier. But across all these industry, media and entertainment, manufacturing, AEC, and scientific visualization, data science, and computing, NVIDIA has software and hardware solutions powering all these industries.

So if I just take an example of M&E, an interesting factoid-- last 14 years, every movie that has won an Oscar for visual effects has used NVIDIA GPUs and platform for all the animation and graphics that are generated that go into these movies. Same thing for manufacturing and product design-- real-time ray tracing, immersive models, real-time simulation, AI-driven workflows, these are all driven by advances in GPU technology.

AECO-- if you are presenting something to customers, this could be a room, a whole floor, a whole building, or a whole block on a city or larger data sets and models. So these are becoming more and more involved processes. And these solutions can drive those type of workflows.

Just to give you an understanding of the breadth of the platform, these are all the GPUs that are available in Dell systems that Scott will talk a little bit later about, in terms of desktops, laptops, going all the way to data center and the Cloud. So there are multiple options available. And ideally, these will be based on your workflows, whether you're using Fusion 360 or Revit or Alias or Red, whichever applications.

Working with Dell and working with Autodesk, we create kind of good, better, best recommendations and suggestions that are available from the Dell website. So based on your workflows, if you're really graphics-intensive workflows, you are creating really high end AR, VR type of experiences, , you want to go for the highest graphics cards which is the A6000 with 48 GB memory, which is really the NVIDIA flagship graphics card.

And there is no competition to that card in the market. It really is capable of driving very large workflows, powering AI type of workflows, and also increasingly simulation. And then based on your workflows, whether you're using desktops, laptops, data centers, you have multiple options available.

So when it comes to manufacturing and product development, these are really the five key areas in the product development, process and the manufacturing process, where NVIDIA tools and technologies play a significant role. Rendering and visualization, will talk about that a little bit later as well. But really creating these immersive renderings, creating these photorealistic rendering, it's really driven by this GPU and the memory that you have.

Artificial intelligence-- when you're rendering these massive data sets, there are technologies such as denoising, which can accelerate this process of creating these large-scale rendering. Supersampling, which kind of upscales the resolution of the images. These-- and I'll share some numbers later on. But these are really, really cutting down the time from days to hours to minutes and seconds of creating these large-scale renderings and visualization.

Simulation-- when people think about GPU, most of the time what comes to mind is rendering and visualization. But more and more, simulations, especially if you are doing any type of fluid dynamics simulations, these are being run natively on GPU. In fact, 2022 has been a significant year, because the two major CFD solution providers, Ansys and Siemens both announced a GPU version of their CFD codes.

So as compared to CPU, these have shown 5, 10, 20 times performance gains running these large CFD simulations. So rendering has been the more traditional workflow. But more and more simulations are taking advantage of GPUs, the massive parallel processing that is available.

Same thing on the structural side-- if you are running any Abacus or similar solvers and such, they too benefit from GPUs for running these workflows on workstations. Virtualization-- again, this ties back to how we are working, remote working collaboration. So you may have a really powerful desktop workstation, which has high end graphics in the office, but at home, you may be using a slightly lower powered mobile workstation. But by connecting to your desktop, you can virtualize the GPU, meaning your GPU is on your desktop. But you can get all the graphics on your laptop through the virtualization tools that are available from NVIDIA and Dell.

And lastly, Scott covered at this point as well. Customers are expecting they are demanding more immersive workflows. Or even if you have to do design reviews with your teams, collaboration or design approval and sign-offs, AR, VR is really enabling all of that to happen. You get a real true-to-life scale understanding of the data sets that you're working with. It could be mechanical products, or it could be whole buildings and those type of renderings that can be done in AR, VR.

No, I'm OK. That's fine. So just to give you a couple of examples-- Alstom is a French company that makes trains, and other public transport systems, metros, railway stations, and platforms and such. And they use Autodesk Red for real-time rendering on NVIDIA RTX GPUs.

And some of the benefits that they have seen-- I'll share some of the numbers on the next slide. But they have seen up to 40 times performance gains when they are rendering these huge, huge, really massive data sets. It's not just the train, but the whole platform the whole structure that is surrounding it. And this is all a real-time rate race. So depending on the day of the time, where the sun is in the sky, what type of shadows it is creating, all that is captured in those renderings.

So I mentioned earlier about the performance gains. So when they were doing this on a CPU, when they were doing these VRED renderings on a CPU, it was taking them hours, 5.8 hours as you see here, with just one GPU rendering. So any VRED users here? No?

So in VRED you get this option-- CPU rendering or GPU rendering. So as soon as they switch to the GPU rendering mode and on one NVIDIA RTX A6000, the rendering time dropped from 5.8 hours to 8.2 minutes, so a massive, massive gain in performance. And then these cards, these A6000s can be linked together because they scale linearly.

So going from one card to eight cards, the simulation time-- sorry, the rendering time went from 8 minutes to 1.1 minute. So that is the type of performance gains. It's 20x, 30x, 40x types of performance gains that you will see. So this is an example of rendering. But we have seen similar performance gains on the simulation side as well.

How are we doing on time? Am I talking too much?

SCOTT HAMILTON: We're 20 minutes in, so we're fine.

HIMANSHU IYER: I'll go through the section a little bit faster. So the previous slides, I talked mainly about hardware, all the graphics cards we have-- desktop, laptop, data center, all the way to Cloud. Next few slides, I'll introduce you to Omniverse, which is a software solution platform from NVIDIA. Anybody here heard about Omniverse? Familiar with it at all? No?

So Omniverse is really a platform to launch your products, to launch your designs, to launch your concepts into the metaverse. It is a platform for creating metaverse. And we'll see how.

[VIDEO PLAYBACK]

[MUSIC PLAYING]

- As an animator, designer, or engineer, you can bring the power of collaboration between applications and people to your workflow with NVIDIA Omniverse. Before Omniverse, working on a project using multiple applications required exporting and importing from one format to another, often causing creators to lose information and compatibility along the way. And adding more people into a workflow was a nightmare of updates and file mergers.

The Omniverse platform enables true real time collaboration for individuals and teams, working between multiple apps and working across geographies. Omniverse brings the ability to seamlessly sync file changes between the creative tools you know and love, maintaining their integrity throughout your pipeline. Whether working alone or together, Omniverse creates an infinite number of connections from app to app, person to person, or office to office. Get Omniverse now and experience the future of collaboration for yourself.

[MUSIC PLAYING]

[END VIDEO PLAYBACK]

HIMANSHU IYER: So Omniverse is really a platform that enables collaboration between design teams, between engineering teams. And this. It does this by using this file format, standard file format called USD. USD started out of the M&E industry. It was developed by Pixar. But it is an open ecosystem.

So there is a lot of companies that are working towards integrating it into their tools and applications. And this, the USD ecosystem is constantly being fed by all of these companies that are working on it since it is an open and flexible platform. So it can start with any application that can export the USD file. And then essentially, Omniverse is where you can collaborate on these different systems, different files. And we'll see how.

So like I mentioned, it could be Revit that you are working with. It could be Maya. It could be 3ds Max. Doesn't matter, all of these can save the file as USD.

So once the US file is created, it comes into something called as Omniverse Nucleus, which is really the hub of collaboration and syncing all the different data and file formats that are coming into Omniverse. And once the data comes into Nucleus, other NVIDIA software tools and toolkits and libraries in terms of AI, physics, robotics, rendering, all that can be used on this data for creating simulations, for creating renderings, for creating robotic, training, and inferencing type of use cases.

And then finally, it can be sent to different portals. It can be sent to a standard laptop or a desktop. Or you can use it in an AR, VR format. So whichever way the data needs to be consumed, it can be easily sent out to that format. So I mentioned the main way of interacting with Omniverse is through these connectors. Connectors are bidirectional connectors, meaning data can go from the application to Omniverse and back. And this is all happening in real time.

Let's say something changes in Revit. The person who is using Omniverse-- and Omniverse will get a notification that things have changed. And if the person made some changes on this side, it will automatically be communicated to Revit as well. At the same time, like I mentioned, the ecosystem is growing. At this point, there are about 40 ISPs-- Autodesk, Unreal, Unity, Siemens, a lot of these have these connectors.

But at the same time, to make it more flexible, you can also use import export to standard file formats through Omniverse. It's not the most efficient way. But it is there. And this is just a small subset. There are over 20 file formats that Omniverse can work with.

So just touching upon this very briefly-- 3D data is involved in all of these stages in manufacturing and product development, and also when it comes to AEC. 3D data is involved throughout the process. And Omniverse can play a part in each and every stage of it, whether it is design, simulation, actually manufacturing the product on the factory floor, or when it comes to the sale and marketing of it.

So just to give you an example, we saw about collaborative 3D workflows, how Omniverse can enable collaboration between distributed teams using different applications. But the crux really is this full fidelity design visualization. All this data can be rendered in real time-- global illumination, ray tracing, really giving you lifelike experiences of whatever data that you're working with, whether it is a product like a car or a blender or vacuum cleaner. Or it could be a whole building or a cityscape or a factory or a warehouse that can be rendered in real time in its global illumination.

Also, virtual prototyping-- again, when it comes to manufacturing, this is a step that most manufacturers want to implement to cut down on the cost of the product and accelerate product development cycle. So doing this in full scale, doing this with the data in its true working environment is really what is going to provide the biggest benefits. So I talked about product level on digital twins. It could be a digital twin of a car, it could be digital twin offer of a cell phone that we all use. But then, it really allows you to scale, going from a product level to a much bigger factory level or a warehouse level or a city level type of a digital twin, where all these tools.

If you think about a factory, the factory itself is going to have an assembly line. It is going to have a lot of these individual machines that are doing the actual manufacturing. There is going to be a conveyor that is going to be feeding these machines. They are probably going to be robots that are moving along.

So all these data sets, how the actual product is manufactured-- if it is a BMW car, some of it might be done in Red. Some of it might be that in CATIA. But the actual machine might come from Siemens. It might have some other controllers and such.

The whole factory floor, the planning, the design might be done in Revit. So you already have three different data sets there which don't talk to each other. I know from experience, I used to work for Siemens. Siemens and CATIA do not talk to each other at all. But Omniverse gives you the capability to bringing all of these multiple data sets together into one environment, and then creating those digital twins, large-scale digital twins.

Just finally to leave you with a couple of examples-- a lot of these major automotive companies already have Omniverse in production. For example, the BMW case that I mentioned earlier, or they are doing some testing POCs and such. And Volvo is one of the early adopters. Again, they use multiple tools for developing these cars and SUVs. And it has allowed them, given them an opportunity to collaborate on these data sets between multiple teams, multiple people.

I think I've talked about most of this already. So anyway, I will hand it over to Scott, but happy to answer any questions after as we come to the end of this session. So Scott, over to you. Want me to drive the-- I can drive the slides.

SCOTT HAMILTON: So I want to talk getting back to the hardware side of it, how we look at computer systems. And I hope you'll find this helpful. Because one of the most common questions we get is, hey I'm running Autodesk Inventor, I'm running Revit, I'm running VRED, what hardware do I need to get the best performance? And the answer is always, well, it depends. There's a lot of variables.

And what we've tried to do is break it down to look at it in two major sections here, what we call interactive work and computational work. And the difference between those two, the way it exercises your computer or uses the components is very different. And if we look at the interactive example, this is when you're sitting in front of your computer and you're modeling something in Inventor or Revit. You're inputting data, you're thinking, and then you're inputting more data, that's what we call interactive.

And that type of operation tends to be very bursty, obviously, because you're starting and stopping doing things in that application. But also, applications in that mode tend to be what we call single threaded. That means they really use one CPU core at a time. And they also-- and that's the top one, that CPU turbo frequency. That becomes the most important part to give you performance, as well as GPU rendering performance because you've got that model that you're working on and you're spinning it around and looking at it from different angles, walking through a building, whatever it may be, that's putting a lot of load on your computer in those two areas.

And then you have sometimes GPU memory speed and then sequential storage performance. When you load or save a model, and we save the model all the time. So if something goes wrong, we don't lose the data. Those are the really important factors when you're talking about doing interactive work.

Now for computational work, it's very different. And what computational work is, is this is kind of the batch process when we're in Inventor and we're going to run some kind of stress analysis. Or in Revit, we're going to run a rendering. Those type of operations tend to be extremely multi-threaded. So they will use lots of cores, whether those are CPU cores or GPU cores, because those processes lend themselves to what's called parallelization. You can do a lot of the same things at the same time.

Take rendering as an example. Depending on the type of render you're using, you can render a whole frame at once. You can render in buckets, which are like chunks of the frame. Or you can render at a pixel level. Or if you're doing an animation, you can render each frame independently of each other.

So that can be highly parallelized. And that requires a different situation. So what's really important there is CPU and GPU core counts or computational performance. And Himanshu showed that example of going from what was it, 8 hours to--

HIMANSHU IYER: 6 hours to 8 minutes.

SCOTT HAMILTON: To 8 minutes to?

HIMANSHU IYER: 1 minute.

SCOTT HAMILTON: 1 minute, that's a perfect example of computational workloads. And Himanshu, if you go to the next slide. So you can see the difference. Those two workloads exercise your computer in a very different way.

So when people come to me at trade shows, and they typically will walk up to the booth and say, hey Scott, my IT gave me, whatever, $2000 budget to buy a computer. I say, oh, wait a minute. I don't even want to know how much money you have. Let's understand number one, what software are you using, what are you doing with, it and what are the computational workloads that it's going to demand. Because that will help you decide what is the best computer to buy based on those needs.

And you may do-- if you're a design engineer, like I used to be at an aircraft company, I did a lot of design work. So it was mainly interactive work. But then one of the guys that I worked with, he was a stress analysis engineer. And he did a lot of computational workloads.

So we have very different needs. We had very different needs in the systems that we would use. And so you have to consider that when you look at that.

Now nowadays, because of all of these new paradigms in the workflow, we're able to do these kind of things together. An engineer now, is really multi-disciplined or an architect. An architect may design something in Revit. But then on the same machine, they may export that into Unreal Engine or Omniverse to do a complete visualization to show to their customer or potential customer.

That's how you win jobs in architecture now. You've got to show this great visualization and what this vision of what the building is going to look like. So these things are becoming-- it's much more common for you to require kind of both interactive and computational if you're doing that type of work.

So I want to talk a little bit about some new technology. One of our other partners of course, is Intel. And how many people have heard of Alder Lake or 12th generation Intel CPUs? A couple?

So this is kind of a big deal. And it's really going to help you guys out. We talked about collaboration. And we talked about OK, you're an engineer, but you're also on Zoom calls all day long. You're multitasking all the time.

And this architecture, it's called the hybrid architecture. And what they've done is they've created a new architecture for the CPU. If you remember, if you're kind of a geek, there was CPUs with cores. And on a mobile system, you could get last year a year ago you could get an eight core mobile computer right workstation, which is great. It was really good at doing interactive and some computational work at the same time.

But now on the same type of computer, you can get 16 cores. And now they've separated these cores into two different types of cores. They call them P-cores and E-cores, performance and efficiency cores. And they have slightly different designs.

The performance cores are for the maximum compute power. And efficiency cores are a balance between computing and power efficiency. And then along with that, there's a technology. It's called Intel Thread Director, that it just runs in the operating system. And it will direct all the processes that are happening within the applications to the P-cores or the E-cores automatically, depending on the workload of the system.

So now, you've essentially doubled the number of cores you have available in a system. And you've got a much more efficient way of distributing these. So a perfect example is, if I'm using Revit or Inventor and I'm doing some modeling, but at the same time, I'm sharing my screen on Zoom with somebody I'm working with, the P-cores will get used to do the main modeling stuff. But the E-cores will be used in the background to help process the video for the Zoom that you're having to stream out to everybody else viewing your screen.

So it really gives a great balance of technology that allows you to get the best of both worlds. And we've seen performance gains, believe it or not, in CAD applications, up to 60% performance gain with this new architecture that's being offered. So why is all this stuff we're talking about so important? I find this slide kind of fascinating because I think everybody inherently understands this, even if you're not in technology. But nobody has really done it well.

I shouldn't say nobody. Some people have done some work and analyzed this stuff. And there's this that's noted here. It's called the Economic Value of Rapid Response Time. It was a research project that was done by some people at IBM. Literally 40 years ago, they did this research.

But essentially what they learn from doing this, is they measured people's productivity relative to the response time of a computer. And essentially, what they found out was that yellow curve is your productivity. And is what you can see this diagram is basically saying is, that as the response time increases from my right to left, your opposite, the productivity goes down dramatically.

And this is all something we inherently understand. It doesn't matter whether using Revit Inventor, VRED, or you're using Microsoft Word, when you do something in the computer and it doesn't happen immediately, what happens? You start getting distracted as you're waiting.

You think about oh, what do I have to do? What am I going to have for lunch? Or I got to go pick up the kids after work, or whatever. And you lose your train of thought. Your productivity dips down tremendously.

What happens if you do something that takes a really long time? Well then, you're in the coffee cup zone, I call it. Oh well, that's going to take a while. Let's go get a cup of coffee.

You come back to your desk and you sit down, and you go, what the hell was I working on? And so your productivity is just terrible. So really what the number-- you can't see the numbers here. But this is one second, this is a half a second. So basically what this diagram is telling you is, anything that takes more and a half a second, your productivity is really dipping.

So that's why this is so important. And remember when I was telling the story about, oh, I don't want to know what your budget is. Because what we're going to talk about is that doesn't matter, believe it or not. So if we go to the next slide and we look at the cost of a workstation, what's the average cost of a workstation? Probably around 2000 to $2,500 is probably the average. Some are less, some are a lot more, depending on what you're doing.

We could sell you a workstation for 30,000 if you want to buy one. And there's reasons to spend that money, believe it or not. But if you look at the cost of the workstation compared to everything else, it's almost insignificant.

And here's why. First of all, if we look at all that stacks on the value side of the scale, labor, number one-- that's the cost of the person sitting at the workstation. Any engineer or architect makes an order of magnitude or orders of magnitude more than the cost of their workstation. So if you just do simple math-- for example, I did some research this was probably eight years ago. The average salary of an engineer in the US was like $75,000.

And then you figure, all the overhead for a company right to employ that person. Most finance people will tell you take the salary and multiply it by 1.5 to 2 times, is about average. That's the true cost to the company when you add in their benefits, the office space, and all the other things that it takes to have that employee on your payroll.

So think about it this way. Let's just make the math easy. Let's just say they're fully loaded cost is $100,000. If we can give that person a workstation that makes them 10% faster or 10% more productive, how much money does that save? And the math is easy, 10% of 100,000, $10,000.

What was the cost of that workstation? 2,500. So if I spend an extra $500 on a workstation to give you that 10% productivity, you just saved $10,000. Or well, you spent extra 500, so let's say $9,500. That's how much money you save, just in labor.

And by the way, that's just the first year. The second year you save all that money over again, and the third year, over again. Now that's just a small part of it.

Prototypes-- I was at another conference. And one of our biggest customers was presenting. It was an automotive company. And they were telling us that in the past, they used to have to build five physical prototypes of every single car model that they bring to market and crash those cars into walls to make sure that they would meet all the safety standards. Guess how much each car cost to make?

They told me it was $1,000,000 a car. And these are because they're all prototypes. They didn't have a factory to build these yet or assembly line. These were all pre-production prototypes they had to able to do this testing at $1,000,000 a pop.

And they said that with computer technology and the simulation capabilities of today, they were able to eliminate three out of every five tests. Because they could simulate it in a computer now and submit those results to the safety organizations to get approval. So that was $3 million. How many workstations could you buy for $3 million? A lot.

Other things, time to market-- if you can design faster, you could deliver your product faster to market, beat your competition. That could save you millions and millions of dollars. Quality is another great one. There was another story I heard.

I actually heard at Autodesk University, it was probably 10, maybe even more than 10 years ago. Another company was saying they had designed a product, and they released it to market. And within a few months, they had to recall the product because they realized they had a cooling issue. And the product was overheating.

I won't mention the company name. They were building a gaming product. They had to recall that product. And it cost them a billion dollars. So imagine if you can do more iterations, design a better product, bring it to market faster, do less physical prototypes, all the money that you can save your company.

So the take away from this is don't think about the cost of the hardware. Think about the money you can save by delivering all of those things on the value side of the scale here. And they far outweigh any cost of any hardware.

And what ultimately it comes down to is making you guys more productive, so you can be better engineers, better architects, better designers. Because everybody in this room, you guys, we like to call you, you are the experts at your company. You're the lifeblood of the company. You design the products for the company that you work for.

That is the company. And if you can deliver all of those things faster, better quality, it means a lot of value to your company. And then we'll just talk a little bit about this.

So again, if you look at whether it's-- I think this is true whether we're talking about the architecture world or the construction world or the traditional product manufacturing world, you have a project or a product lifecycle. And in the beginning of the cycle, the cost of making a change is very small. Because you're basically in the digital realm.

If you're modeling something in Revit or Inventor and you have to make a change all digitally, that's pretty easy, usually. But as you get further down the product lifecycle and you start construction or you start setting up a factory floor and assembly line and now you have to make a change, it gets super expensive. Because now you have all these physical things. And those are not easy to change.

And then when you get to the end where you're delivering to customers and you have to make a change, like you have to recall and it cost you a billion dollars, that costs a lot of money. So the ideal area to be is in that beginning of the cycle where you can do tons of iteration and refine your product as much as possible in the digital world. Because it's easy. It doesn't cost very much to do that.

So spend a lot of time there. Do lots of iteration. Take advantage of all the new workflow paradigms and the hardware technology that enables you to stay in that ideal iteration cycle, get everything down, and then make the rest of the product development cycle, product lifecycle go much smoother. So we have to do a little bit of product stuff.

So I have two more slides for you guys. These are precision products. So Dell's a big company. We sell computers to all kinds of customers.

We sell them to consumers. We sell them to gamers like Alienware products. We have servers. And we have precision products.

This is the product line that are designed for professional engineers, architects, media and entertainment professionals, that have very specific demands. And we talked about some of those like that compute versus interactive performance, all that kind of stuff. It's very different from somebody that's sitting there using Microsoft Word. That doesn't really put a lot of demands on a computer.

So it's a very different proposition. And we design these products in our precision product line specifically for Autodesk users, engineers, architects, et cetera. And we really have three different series of products-- 3000, 5000, 7000. And really, the gist is 3000 is what we consider entry level. It's a little less expensive. So people that are price-sensitive or maybe they don't have that many compute demands.

And I think a great example-- who uses AutoCAD? I suppose there's quite a few. Everybody knows AutoCAD is not hugely demanding on your computer. You're mostly doing 2D work.

But then as you get up into Inventor and Revit, you're doing lots of 3D stuff. And I'm not knocking AutoCAD because I know it has some good 3D stuff in it these days. I was surprised. But those applications become more and more demanding.

And then you look at something like Red, that's super demanding on the graphics. Or you look at like 3ds Max or Maya, it has all that interactive stuff. But then you do tons of rendering, which is that computational workload. So that has more demand.

So we have different product lines to give you different performance levels at different price points, and also to give you different levels of expandability. You can buy, for example, a or 3000 series tower, which is a nice little small device that you can have sitting at your desk. And by the way, if any of you guys want to see this stuff, come to our booth, Manufacturing 400. You can buy that small little thing to sit on your desk.

Or you can buy a 7000 series tower. We have one of those that our booth. And we're running Omniverse and VRED on a 7820, are more expandable. It has two NVIDIA A6000 graphics. Remember all that graphics acceleration he was talking about.

It's a dual socket, so two Intel CPUs in there. And you can go up to 512 gigabytes of RAM in that machine. So we like to call it a supercomputer at your desk. And some people need that type of stuff.