A Hardware Wonk's Guide to Specifying the Best 3D and BIM Workstations—2016 Edition

MATT STACHONI: All right, so welcome to this class. Thank you very much for attending. I know that there's a lot of classes out there, a lot of competition for seats. So it was really nice to have everyone here this morning. So what's this class about? Well, if you had a chance to look at the handout, it's all about specifying hardware for BIM and visualization, basically. Mostly AEC applications, although if you're in the civil stuff with the dirt and all that other stuff, the hardware recommendations that I passed through are going to be basically the same. But I'm focusing mostly on Revit performance, 3ds Max, but also other applications like Rhino, Lumion, on V-Ray, stuff like that.

So a couple of key learning objectives here. Understand how to specify workstation components. OK, so if you go out to Dell or HP or Box or any place like that and you have different lineups of different machines, what are the guts of the machine? What's really going on behind the hood? And a lot of people, when they don't know enough about hardware, don't know a lot about hardware they just sort of pick a box and just hit the send button. They buy it, it ends up at their desk, and it doesn't have half the ram they need, it's the wrong processor, it's not optimized for what they want to do. So this class is about what do you need to know to specify the right parts?

And also understand where the sweet spots are. Because you can buy any machine that has a lowly processor or a very, very fast processor, but your spending curve goes like this as you go towards the faster stuff that gets insanely more expensive. So the sweet spot is that point where yeah, I'm willing to spend this much money for this much performance. But if I go one tick higher, the price just jumps off the scale. So understanding where that best value is is important.

And also understanding how to buy a workstation or specify a workstation for specific users. I know a lot of people-- a lot of IT guyss-- they don't really think about what the user is doing. So they just sort of-- I got to buy 20 boxes for 20 people, and that's it. Whereas you might have one guy who's spending most of his time in visualization, they don't get a machine that's powerful enough for that. So you want to make sure that you know what users you have, what their capabilities are, what their capabilities are going to be in about three years, so they can grow into their system if they need to. But also, don't buy too far down the curve or too high up the curve, depending on what you need. And then also how to specify or buy or how to shop for individual components and systems like that. Where are some good places to look, things like that?

So there's a lot of stuff in this session. We are here for an hour and a half. OK, let me get that right first. Because last year I taught a class, I thought it was an hour and a half, but I only had an hour. So we've got an hour and a half, which is good, because there's a lot of stuff to cover here. At the very end, I'm going to have a couple of system builds that I put together for you so you can sort of compare and contrast different systems for different kinds of classes of people.

So when you're looking at a computer, or looking at a new workstation, your limitation isn't really on how fast the machine is. It's not something that you look at it and go-- one sits there and goes, wow, this thing's just-- I can't do anything as fast as this machine everyone's mind operate faster than their workstation, especially if you're working in Revit. So being able to specify a workstation that works for you, that doesn't keep you just hanging all the time whenever you change a view or render something is important.

So stuff we're going to talk about in depth. We're talking about Central Processing Units or CPUs, which are the brains of the computer, graphics cards, or GPUs, system memory, and mass storage. Those are the four components that you really need to worry about. Nobody uses optical drives anymore, so you don't have to worry about DVD or anything like that. I mean, everything else is pretty much secondary. We'll also talk about peripherals a little bit. I do a lot in the handout about biased opinions on certain little components and stuff like that.

But we're also going to talk about buying guides too. When you're specifying a machine, what does all that stuff mean in that wall of text right there? What stuff matters? What stuff doesn't matter? What's a good base configuration to work for? OK, so I'm going to talk about specific components, specific manufacturers, things like that. These are based upon personal preference.

They are in no way, shape, or form meant to endorse anybody. Nobody pays me, unfortunately. Is anybody here from NVIDIA or Intel? Damn. All right. Just because if you did want me to say something nice about your product, if you wanted to send it to me to evaluate, I would do so. You know, I have standards, but they're really low. So we can go. Darn. Every year I try to ask that question, and every year no one's there.

All right, so how many people here are using something in the Building Design Suite or in the AEC collections? Pretty much everybody. Who here is on the AEC side or the architecture side, I guess I should say. How many people are on the Civil side? How many people are using-- I'm assuming most people are using Revit on the architectural side. Good. Navisworks? 3ds Max? OK, a couple of people. Good. Lumion? OK, a couple of people here. Good. I got to work with Lumion quite a bit in my last firm. It was a gas to use. Really easy. Really expensive, but really powerful. Rhino? Couple of people. V-Ray? Good.

How many people are creating renderings on pretty much a daily basis kind of thing? OK, good. How about animations? A couple of people there. Good. How many people are running hardware that's like two or three years old, and you're wondering what I should do to make everything faster? OK, good. That's what I wanted to see. How many people don't know what to buy? Obviously most people here, because you're here.

OK, so if you're looking to buy a new system, the first thing you have to interesting is what's happening in the industry right now that's causing-- where you need to be. So on the AEC side, basically you've got everyone modeling everything all the time. And the record number of people using BIM tools, whether it's on the design side or on the contractor side, especially, it's increasing at a really high rate. So it's pretty much evenly diffused throughout the entire industry. If you're not on Revit or on a BIM platform, you're going to be within two years, or you're not going to really be doing a whole lot of business.

So we have everybody using Revit now. We have obviously new hardware capabilities coming down the pike every year that allow for some heavier problem solving. It also allows you to do things where you're doing more of what they call parallel processing, such as rendering, modeling, and things like that, where you're doing more than one thing at one time.

We also have distributed computing with the cloud. So we have cloud rendering. Everyone's probably used that at some point. It's fantastic. It's basically changed a lot about how you think about buying a new workstation. If you're going to be looking for our heavy rendering machine, maybe you're not going to look at that anymore, because of the fact that you can do a lot of your rendering in the cloud. Saves you a lot of time and a lot of effort and a lot of money at the same time.

How many people play video games? How many people do not play video games? OK, I'm sure half of you are lying, because everyone plays video games. But nice thing about games is the technology that is driving the hardware side from a market standpoint-- faster processor, faster graphics, more memory, things like that-- those are really largely being driven by the gaming industry, because games are getting more and more cinematic, more and more crazy, more and more realistic. So all of us on the design side are directly benefiting from that advancement there.

A lot of people are running more than one application at one time, right? You're going to have Revit open, you're going to have Max open, you're going to have Photoshop. All that stuff consumes a lot of memory, disk space, processing power as it's running. And also Windows itself is a little bit of a memory hog, because you've got services running in the background, antivirus, all that kind of stuff. And more and more of these applications are multi-threaded so that you can take advantage of more than one core in your processor.

The problem, though, is that we are getting to a point where every year the hardware increases are getting smaller and smaller. So in 2015, CPUs went down to what they call a 14 nanometer process size. And I'll explain what that means in a little bit. But that's essentially getting down to the point where you can't really make the chips any smaller. You can't shrink the size of the transistors. We're just getting down to the point where it's getting really, really crazy on the transistor side, and they have to come up with more and more exotic technologies to get us down to get the rates down to zero.

In 2016, we also had on graphics side of things, GPUs now are down to a 16 or 14 nanometer process too. They we're stuck at 28 nanometers for a while. And I'll explain what this means. Basically it means that GPUs and CPUs are smaller, which means that you get more transistors on the die, which means that they can do more stuff. They're running cooler. The smaller the process, the cooler the chip. Usually the faster it'll run, because if you've got a cooler chip, you can ramp it up so that it's faster and keep a thermal load that is going to be acceptable.

We also have the rise of what I call GPGPU or graphic computing doing computing on the graphics card. And we'll talk about that a lot when we get into the graphics card section. But a lot of things happening on the gaming engine side with Unreal Engine 4 and Stingray, Unity, things like that are directly affecting the AEC side, because now you can use those applications or those engines to basically drive all your visualizations. So it's a pretty exciting time on that side.

Virtualization and virtual desktop infrastructure, or VDI. Anyone using a VDI system now? Basically, there's a server sitting in a closet somewhere, or in a data center, and it just serves up your desktop for you. Your desktop can be one of those little tiny-- Anyone take a lab yet? Anyone have a lab yesterday? If you took a lab, you might see instead of a big box sitting at your feet, you now have this little tiny box about this size.

And the labs are running-- or at least eh lab I was in-- was writing everything off of the Amazon cloud. So the Amazon cloud was serving up your desktop. And me and my partner were walking through a lab, and my partner goes to put a USB thing in and turns it off and reboots the whole system. And we're like, OK, we're screwed. But then the guy came over, reset everything, and we were right back to where we were.

We didn't lose anything, because it's all processing in the cloud. We're just there as a simple conduit to see it. Speaking of the cloud, obviously you've got cloud-based services with Microsoft Azure, Amazon EC2. The cloud is the perfect place to do high demand rendering, because you can put thousands of processors against a rendering problem and not just four or eight or whatever. So it's kind of a waste of money to try to compete with that, because you're never going to be as fast. So as cloud rendering especially becomes more and more prevalent, it really does change the landscape of how things happen.

There we go. Let me back up a second. OK, so I was talking about transistors and how we are kind of quickly reaching a pinnacle of what we can do in traditional transistor and microprocessor design, which is really the reason that drives much of the conversation that we're having today. So there's this thing called Moore's law, which I'm sure everyone's heard of, that basically says the number of transistors doubles every 18 or 24 months.

And it works by shrinking the size of a transistor. And I'll explain what a transistor is in a second. But basically, when you shrink that over time, you can fit more of them on a chip. And you'll notice that this scale runs pretty linearly in terms of the direction of this arrow right here from 1971 up to recently. But the scale on the left hand side is logarithmic. It's an exponential function. It goes zoom like this. So if you double everything every year, pretty much it gets crazy after a while.

The problem is that you shrink these things down so much, and you're literally running out of atoms. At 14 nanometers, I think the size of a transistor is about 70 silicon atoms wide. That's crazy. A transistor works like this. You have a source on one side, you have a drain on the other, and then you have a control gate across the top. And it's an electrically-driven switch. So when you apply a voltage to that thing, it creates a magnetic field, and then the field underneath it, electrons will congregate there and close the switch. So this process size that we talk about right there is the distance between the source and the drain.

So like I said, we're at 14 nanometers currently right now for the high end CPUs, and about 14 to 16 for the GPUs. So the problem is that as you're seeing this trundling on, we were keeping pace with this fairly well until just recently. And to give you a sense of scale of how this works, in 1971 we had about 2,300 transistors on a chip at 10 micrometers or 10,000 nanometers. And then in 2016, we've got over 1.5 billion transistors at 14 nanometers. So it's basically equal to shrinking yourself down to the size of a grain of rice.

Put another way, the Boston Symphony Orchestra holds about just as it happens, 2,300 during pop season. And it's kind of like shrinking everybody down so you could fit the entire population of China, which has 1.3 or 1.4 billion people. You can fit the entire population of China in Boston Symphony Hall. I mean, that's the scale we're talking about in terms of how Moore's law has trundled on.

Now we're expected to go from 14 nanometers now to 10 nanometers in 2017. Whether we get there or not is up for debate. They're having a really, really hard time getting anything smaller than 14 nanometers to scale well. But they've got insanely smart people working on this problem. The problem is that when you get down to certain points, you get to a point where quantum tunneling comes into play.

And quantum tunneling is this weird effect where electrons will just be there. Like the probability of an electron crossing that gate becomes really, really high. And the gate has to get taller. It's kind of like a wall you're going to build between one side and the other. So you're trying to keep the wall there so that the electrons don't go across until you tell them to. The problem is that when you're down that level, they're crossing over anyway, because the probability of those electrons just showing gets to be pretty high. So it's kind of weird. It's not kind of weird, it's very weird.

So when we're talking about other hardware trends or other trends that are affecting your hardware purchasing decisions, the idea of multi-processing, multi-threaded stuff is coming into play. We talked about how you can have a single program execute many threads of execution. So for example, Revit will pop out a thread for doing wall clean up. It'll pop out a thread for doing rendering. It'll pop out a thread for doing a view generation, things like that. Those threads can be put on different cores of your CPU, so they can all run concurrently.

You can also have problems when you're doing that. You have issues where you have to keep all that straight, and that's one of the problems that Autodesk and other software developers have is that this is hard to program. So they have to keep all of this stuff straight, and it just consumes a lot of time on the development end. And so when people talk about how multithreaded Revit is how multithreaded AutoCAD is whatever, you have to understand that they're looking at certain aspects, certain performance targets they're trying to hit, and certain aspects that are low hanging fruit that they can easily sort of spin off as a thread of execution, and other things aren't quite so easy to do.

You can see multithreaded stuff happening in mental ray. If anyone has ever done a rendering in mental ray, you'll see these little white squares that will appear. Those are called buckets. And the number of cores you have in your machine indicates the number of buckets you'll see, because each one represents basically a core working on that particular square. And then each of those gets processed independently of the others, and then it creates your final image. The more cores you have, the more buckets you have, the faster your rendering is going to happen.

Like I said, gaming engines are becoming a huge deal. This is a scene from a rendering-- or actually an animation-- a walkthrough. Not really an animation, it's kind of like a walkthrough game in Unreal Engine 4. Has anyone ever heard of that? Unreal Engine 4? It's basically the big mother of all gaming engines right now. The cool thing about Unreal Engine 4 is first of all, it's developed in an incredibly aggressive pace. If you want to look at it, it's very easy to download and install. It's also free.

For visualization artists, you can download this, you can run it, you can go ahead and build your interactive animations in this thing for free, no charge whatsoever. Epic, who publishes Unreal Engine, charges gaming developers to use their code when they publish a game. So that's how they make their money off the engine. But for visualization artists, no problem. Obviously with Stingray and LIVE, which is basically the connector from Revit to Stingray, that's also available. It's also fairly inexpensive. I think it's $30 a month or something like that.

We have Lumion, which is kind of the granddaddy of this kind of stuff. It does use a gaming engine, but not a commercial gaming engine. They just created it for high end visualization. And so we have that kind of thing that's available to us. So we don't have to sit there and wait for a rendering for four or five hours. That's really 2010 kind of stuff.

And so obviously, the cloud has a big effect on everything, because again, if you can do all this stuff in the cloud in some way, shape, or form-- even building a rendering farm on EC2 is fairly cheap. So you don't have to put a rendering farm in your company, which ties up machines that somebody else might want to use. You can put all that stuff in the cloud fairly quickly, fairly inexpensively.

From a pricing standpoint, we also have what I call pricing compression, where you get hardware every year, and essentially the prices of things become so close for double the performance or double the capability. So like for example, a hard drive. A one terabyte hard drive I think is about $70, four terabytes is 140. Why not just spend the extra money and get four times the amount of space for basically double the cost? So when you look at it from that side, it just makes sense to go take the system one step up to do what you need.

So if we look at the AEC applications that most people use-- this is using the Building Design Suite-- what I did was gave you a chart right here that basically says for any of these applications, what does it stress? And for most things, a score of like 7 can be handled by any machine you buy pretty much off the shelf. If you went down to Best Buy and just sort of picked up a machine that was a medium-priced thing that was sitting on a shelf, it would handle anything up to about a 7 here. Anything above that, you really need to start shopping to optimize what's going on.

So for example, Revit, it really requires a fast processor. It also requires a lot of RAM. It wants a lot of hard disk speed, because the files are very large. Same thing for the 3ds Max Design. Here for the graphics card capabilities, a decent graphics card will work OK, but once you start using iRay for rendering, then you have to jack the card up. Because iRay works on the graphics card, so it stresses that. And this is in no way, shape, or form scientific. This is really just my personal impressions of what gets stressed on each application. But I think it's fairly accurate. I think it basically tells a story of the landscape.

So like I said, Revit, when we concentrate on that, that's the main bread and butter application that most people are using. It's stressing everything about your system in one way, shape, or form. It's very CPU-dependent, because it is a database application. So database applications stress the CPU a little bit differently from most others, because they typically prefer internal caches, like memory caches, like the L1, L2, and L3 caches, much more than other applications do. And Revit files consume a lot of system memory, obviously.

It is a parametric change engine, which means that when you make a change, it propagates through the entire database to make sure that change is OK. Whenever you do a synchronize with central operation wherever Revit's opening that file on the server, pushing your changes to it, making sure those changes are OK, and then pulling the changes down that everybody else has made since you did a last synchronization, that's all processor-intensive, it's network-intensive, it's pretty much everything-intensive.

Because it's all about creating relationships. There's no shortcuts you have in Revit. Revit has to dot every I and cross every T to make sure that everything's going to work OK. Otherwise your building's going to fall apart. If you have an unconstrained or something that's constrained that's conflicting with another constraint, all this stuff has to be double checked.

| it's computationally expensive. So that's why Revit is basically the hog of the bunch there. When we talk about specific CPU considerations and things like that, there is a Revit 2017 Performance Technical Note that you can download. I have a link to it in the handout. I'll probably post it to the AU site before the end of the week, so you have that available. There's a couple of supplemental documents I have too that I didn't post yet, but I'll post those before the end of the week. And like I said, Revit is somewhat multithreaded. It's getting better every year. As Audodesk finds places that they can go, oh, we can optimize this, or we can optimize that, they'll make it multithreaded. So things like vector printing and the new art renderer is completely multi-threaded. File opens and saves, stuff like that.

When it comes to views and graphics, views are live reports of the project database. So most people have several of these open at one time. Each view has its own display properties. A lot of these things consume a lot of memory. So a lot of the performance that you get out of Revit is driven by the hardware to a certain extent, but it's also driven by how you use Revit. Because the way you use Revit is a big variable for a lot of people.

Some people model everything. Some people model very little. Some people can use the efficiencies of the graphics system to do things like have families that have a lot of masking regions and detail, or symbolic line work so you don't have to show 3D stuff in a hidden line view, which consumes a lot of time on the graphics card to hide those elements. So the more optimized you're Revit approach is, the better optimized Revit will be on your system. So a good 600 megabyte Revit model will probably perform better than a bad 200 megabyte Revit model. So understanding how the model or the optimizations you can make when you're modeling and documenting things is really important. And a lot of that stuff is covered in the technical note.

Shadows specifically are a tough one, because they are actually now computed on the graphics card, which is nice. So the faster your graphics card, the faster your shadows will be. And they get better every year too, because when we talk about the specifications of your graphics card and the capabilities of that graphics card, a lot of stuff again from the gaming industry is moving onto the graphics card. So we can do things like shadows on the graphics card. It doesn't have to do it on the CPU.

So when we have ambient shadows where Revit is calculating kind of like the dark area where you have two surfaces meet in the corners, that's actually a technology that was developed by Industrial Light & Magic for the crappy film Pearl Harbor. They won an award for that. But basically, that has drilled down again to our design applications, where ambient occlusion just happens. It's pretty effective. It really makes an animation or a rendering pop. But again, this is all stuff that's done on the GPU as well. So when you combine the two effects, you get some really nice views.

Like I said, it stresses RAM. It stresses the storage. it stresses the network connections, as you suck down a 600 megabyte central model every morning when you open it up and create a new local. That's why solid state drives are the big deal with Revit, especially. And the move to 10 gigabit networking is becoming prevalent now too. Most people operate on a gigabit network. So you get 1,000 megabits per second speed between you and the server on a good day. That's the theoretical limit. The practical limit is much lower than that. But now you have 10 gigabit networking is coming into play that still uses the same cabling, which is nice. But it also boosts the transmission speed by 10. So now machines are coming with 10 gigabit networking ports, and you can get 10 gigabit switches and things like that for about the same price. So it's coming down.

We're talking about 3ds Max. When we talk about polygons and materials-- You know what? Hang on a second. Yeah, I'm sorry. I had a slide out of place. I thought I was in the wrong presentation. Sorry. So we're talking about polygons, right? Millions and millions of polygons in a scene. 3ds Max can handle that. When you take a file from Revit and throw it into Max, a lot of times those polygons are optimized. You're having a lot of polygons that are naturally going to get hidden, because they're concurrent with another face and something like that. So again, the modeling aspect of that is higher when you bring something in from Revit, because the model is automatically less optimized than if you just model every single polygon yourself.

And working with that in a view port is also graphics-intensive. That's all handled by the graphics card. It also has to handle materials. Materials can be very processor-intensive as well, because they have physical properties about them. They record the UV mapping and the different maps that happen, the fuse and roughness and all that other stuff. And then you also have procedural materials with stuff like substance designer, where the material is not using a bitmap at all, it's basically all computed pretty much on the fly. And that's also processor-intensive.

We work with lighting, where we're dealing with photometric physical lighting, calculating how that light actually works with the materials. So as light bounces off of something like this, where it's kind of diffuse, you're not getting a lot of shiny stuff. But if you bounce it off of a metal object, it is very shiny. Everything's working together here. And then we also have rendering. So that's where everything's coming together, and it has to process this stuff. And this is also very, very computationally expensive as well.

Most renderers, like the art render, mental ray, the new Arnold Renderer that's now in 3ds Max, those are all CPU-bound. They do not use the graphics card whatsoever. The renderers that do use the graphics card are iRay and V-Ray RT. Not just normal V-Ray, but V-Ray RT. Those will use the graphics capabilities on a particular graphics card. We'll talk about that when we get there.

Get to Navisworks. How many people are using Navisworks? OK, good. It's really come into its own. Navisworks is much lighter in terms of hardware demands than Max or Revit, so anything you have that can handle either of those two applications is going to handle Navisworks just fine. Essentially what happens, when you bring a file into Navisworks, whether it's CAD or Revit, it turns it into an NWC, which is just geometry. But it's also a very, very simplified geometry. It's only the outside faces of that thing. So the geometry becomes very light. It's basically the geometry and then the data that's behind the BIM elements.

Any normal computer nowadays will handle these things just fine, like that the Clash Detective and TimeLiner produce some very interesting results, but they're also very lightweight in terms of how they affect the overall system. And Navisworks now includes cloud rendering too. So if you're into rendering in Navisworks, you can do so pretty inexpensively.

And when using ReCap point clouds, point clouds are really tough on a machine. It's tough on the CPU. It's tough on the graphics card. It's stuff on memory, because these files are huge. It's tough on disk space for the same reason. So you're talking about billions of points. You're talking about files that are gigabytes, terabytes sometimes amounts of data. So being of a process those things effectively requires you to basically just max out pretty much everything on your system, especially storage.

So let's look at the hardware trends that are happening in 2016. Now, I'm going to be talking about Intel CPUs. Right now this year, everyone's like, well what's happening on the AMD side? The answer is, nothing's happening on the AMD side. Anyone here have an AMD-based computer? Probably not. You probably haven't had one for at least 10 years. AMD is coming out with a new architecture in 2017 called Zen which is targeting basically decent sized Intel chips. So there might be competition next year in terms of what's happening on the CPU side. Right now, Intel owns everything from the low end to the high end. So that's why when I talk about-- oh, he's an Intel fanboy. I'm like, I don't have any choice. It's Intel or nothing to get the performance that we need.

What happened in 2015 was we had a a die shrink. And I'll talk about what this means, but basically we have a new architecture called Skylake. When we talk about architectures and the models of chips and things like that, we talk about things in terms of code names. So we have things like Skylake and Broadwell and Haswell from 2014 stuff like that. So when I talk about those things and as you read the handout, you'll understand what I mean by those things.

So if you're running something that's kind of old but decent like a i7-2600K, which was Sandy Bridge Then we had Ivy Bridge Then we had Haswell. Now we have Skylake. So keeping track of all that stuff is kind of painful sometimes. So like I said, the Skylake architecture came around in 2015. And then since then, they've been improving it with little tweaks here and there. So it's important to pay attention to your CPU specs when you're looking for stuff, because you can get decent performance over something that was just released a year or two ago.

On the graphics side, we have a new Pascal architecture from NVIDIA. And just like I talked about with the Intel CPUs, I'm also going to be talking a lot about NVIDIA GPUs here. Not that AMD doesn't make great graphics cards. But when we talk about certain things that are specific to Autodesk applications, then we have to start talking about NVIDIA stuff. And I'll talk about that in a little bit. But they also had a process or a new architecture that is now in that 14 nanometer range or 16 enemy range too.

On the RAM side, we're looking at a new standard called DDR4. That's the new RAM standard. We'll be talking about that a little bit too. And now on the storage side, basically if you're buying a new solid state drive, 500 gigabytes is pretty much the standard main line minimum that you would want to get. Even one terabyte drives aren't that much more expensive anymore. And they're getting up there in size. So now you find two terabyte drives, stuff like that. So expect the mechanical drive market to sort of disappear over time. Anyone still running a mechanical drive? Like a rotating platter? Don't. Go out and get a solid state drive. Everyone's on solid state drives by now, at least in some way, shape, or form.

Let's talk about processors. Let's get to the meat of the stuff here. The first thing you have to do when you're talking about a processor is you have to talk about the platform. Excuse me for just a second. When you're talking about the platform, what you're talking about basically are the capabilities of the system. So you're not really choosing a CPU first, you're really choosing a platform. And there are three of them.

There's a desktop platform, kind of a run-of-the-mill kind of thing you'd find at Best Buy, there's what we call a high end desktop platform, and we also have a workstation platform. So the desktop platform is basically run on the Skylake series, what we call the sixth generation Skylake series CPU, such as the i7-6700K that I have listed there. It has a particular socket. It has a particular chipset.

Now, the chipset is the glue that basically connects the CPU to the rest of the system. And the chipset has capabilities that really flesh out the rest of the computer. So it determines how many USB ports you can have, and the audio, and connections additional PCI slots, and stuff like that. So the better the chipset capabilities, the better your overall system experience is going to be.

And for a long time, Intel put out really crappy chipsets. They were like, you had these great new shiny processors that did all these crazy good things, but they were hampered by a chipset that didn't have enough USB ports, or had terrible SATA configurations and stuff like that. So Intel finally got religion on that, and now they're putting out decent chipsets now. So we don't have to worry about those being hampered by that too much.

So each platform, each CPU in that platform has its own socket. That socket has a chipset behind it. And we work at it from that side of things. So when you're looking at your users, if you've got what I would call a BIM grunt-- the guy that is just sitting there doing Revit all day long-- a desktop machine is probably going to work just fine for him. The BIM guys, the really good guys that are running Dynamo and probably running Revit in conjunction with maybe Lumion or something like that, that's the next higher up sort of class of people.

Then you have the guys that are the visualization wizards of the VisWizes, as I called them. And they need a machine is completely on a different planet from the other two, because they're running high-end rendering applications, they're just really taking things to the max. But at the same time, a BIM grunt can turn into a VizWiz in fairly short order with all the stuff that's happening with software.

So when you're looking at stuff, if you go to buy a machine yourself or build a machine out yourself, you're looking at stuff like what's on the motherboard, which is basically the big circuit board that everything plugs into. So you're looking at things like the number of PCI slots. You're going to have at least one for the graphics card. But if you want to run more than one graphics card-- which is becoming more and more popular-- you have to look at the capabilities of your motherboard. If you try to put two or three graphics cards in a little tiny form factor box, that's not going to happen. So you have to pay attention to certain aspects of that.

How many people here build their own systems? Maybe for home. How many people go out and they'll go to like Dell or Box or someplace like that, and they have a corporate vendor that they look for all the time. Yeah, most people do that. So if you're in that camp, you're not going to really have any choice over the motherboard. Dell doesn't give you a list of motherboards to choose from.

And you do have to do a lot of research to find out if that particular machine will handle some of the things you want to do, because a lot of times, they put out stuff that you get it, and you're like oh, I can't put a second or third graphics card in here, because I don't have the capabilities. So pay attention to look for the technical specifications. Couple new things that are happening down the pike here on the platform side. I'll get into this, but we have a new M.2 slot for new SSD form factors. You know, Wi-Fi onboard, more than one gigabit port, or 10 gigabit ports, things like that.

There we go. When you're looking at a motherboard, especially based upon the platform again, it's going to drive a lot of the capabilities of that motherboard. So the platform's can tell you things like how many RAM slots you have. So a desktop platform is probably going to come with four slots. You get up to the next high-end desktop based on the Broadwell CPU, and you're going to be looking at eight slots for RAM. Same thing on the Xeons as well.

But you also want to look at the PCI expansion slots. Again, if you're looking at multi-GPU kind of stuff, using iRay especially. So when you do that, you have to look at the different kind of CPU you have. Because some CPUs in this class that I'm showing right here have 40 PCI lanes. Some have 28. And the way that those lanes get configured is in terms of the number of slots you have based upon their speed, or the number of lanes going to each slot.

And usually these big slots that you have here are what they call x16 slots, the long ones. And so the more lanes, the faster that particular connection will be. Now, the thing about this is that when we talk about PCI 3.0, this is basically double the bandwidth on a single lane that PCI 2.0 had a couple of years ago. And the thing about this is that we're not constrained on that at all.

Like, the graphics card really doesn't pump that much data across the thing so much that it would choke it. So when you're looking at the slot configurations, like if you put one graphics in here, it's going to run x16, if you put two, it's going around both at x16. If you put three, one's going to be x16, two are going to be x8. Or I'm sorry. Two are x16, one's x8. And then and so on and so forth.

But even here, if you stuffed five graphics cards in a system and they're all running at a x8 speed, that's as fast as a x16 speed on PCI 2.0, which was fine. So you don't have to worry about that. Like x16, x8, it's the same performance. You're not going to see anything happen there. But the idea is that depending on the number of PCI lanes in the CPU, it determines this number here. And it can determined what a manufacturer can do in terms of laying out those PCI slots. When you get into like the four and five slot range, you're looking at extended ATX form factors, bigger motherboards, bigger cases, bulkier stuff.

All right, and so we talked about the desktop strategy or the strategy for what Intel and AMD do on their CPUs, the desktop strategy basically is a four core processor. The high-end desktop or HEDT range has anywhere from six up to 10 cores. And then on the workstation side with the Xeons, you're talking about anywhere from four to up to 18 and beyond. So the number of cores you have is predicated on the platform you decide.

We also have hyperthreading, which essentially doubles the number of logical cores that the operating system sees. So if you have a four core system, Windows is actually seeing it as an eight core system, or eight logical CPUs. And on AMD's side, they've always been performing-- or their performance level right now is what they call a core i3 a core i5, sort of a low to medium level of performance, which is why they're really not competitive on the AEC side. When we're talking about Intel's chips and we're talking about the i7 with four cores plus, that's where we are right now. So AMD really can't touch the core i7 or the desktop range that we have on Intel side.

But like I said, in 2017, we got this new Zen architecture coming out, and the whole thing is going to be-- hopefully blows wide open. This is the reason why prices are so high for CPUs. When you look at a price for a CPU, just the CPU by itself, and you see its $500, $600, $800, over $1,000, basically it's because there's no AMD here to keep them legal, keep them happy, keep them sane. So the prices are going through the roof just because there is no competition happening. So once we get that in there and it performs OK, we're going to be looking at lower prices.

Now, what we've had over the past couple of years is what Intel calls a Tick-Tock development strategy. And what you have is you have to things about this. A tock starts out with a new microarchitecture, an architecture based upon a certain process size-- 45 nanometers, 22 nanometers, 14 nanometers, and so on. The tick is where that process is allowed to shrink using the same microarchitecture.