How to Build a 4K Editing Computer (More cores are not always better) - Smarter Every Day 202
- Hey, it's me Destin, welcome back to SmarterEveryDay. It's coming up on 1 a.m. I have a problem in my life. It keeps me up at night, keeps me away from my family, which that's the one that really bothers me. It's rendering, look at this. This particular file that I've been working on right now, nine and a half hours worth of rendering on this thing. And I've still got an hour to go. What do you do, man? I love 4k 60 frames per second footage, I love it. Prince Rupert, look at me. The rule of Prince Rupert. (meowing)
This is a Panasonic GH5; it shoots fantastic imagery, but the problem is the camera hardware is outpacing the computer rendering hardware. You put that stuff through that computer right there; it's crunching, like the computer's pegged out, right? And I'm still spending hours and hours of render time. I need a better solution. So I flew to Seattle; there's a guy named Jon about to pick me up. He owns a company called Puget Systems, and they make specialty computers for certain applications. We're gonna go meet with him.
How's it goin'? - Hey, how are you? - Throw stuff in the back? - Yes. - Awesome. This is Jon, he owns a company called Puget Systems, and you're basically the answer to everyone's computer problems, right? - Well, not all the problems. (laughing) We don't really ask what hardware do you want. We ask what are you trying to do? And then we can tell you what hardware do you want. - Oh my gosh. And so you start from there and then build the hardware based on what they say. - Right, right. - What? Why is that so revolutionary? - I don't know; it shouldn't be. - It should be! You're kinda tucked away here. It's kind of a nice little area. - Yeah, we like it.
[Destin] So do you guys personally believe this? Is that the deal? - Personally believe what? - [Destin] This thing that's on the wall behind you. - Oh, well, I mean we have tattoos of it, so -- - [Destin] You have tattoos of it? (laughing) Jon took me into the lab, which is where they test stuff. His brother Matt and their co-worker William spend the bulk of their time testing different combinations of hardware in this room.
What I found hilarious is that when I would ask a very specific yes or no question about performance or hardware to buy, it became very clear that every answer I received was going to come with a caveat. - Sometimes, depends if it's right. - If it's set up right, in supported applications. - Yeah, for that very specific task. - Depends on exactly what you are doing, and the balance between everything.
[Destin] The caveat there answers because they've actually collected the data. They perform a suite of benchmark tests, which are a standardized task you can use to see how fast or efficient a hardware configuration is. Over and over they'll run a software test, swap out a component, and do it all over again. Here you can see Matt's computer is automatically running an Adobe Lightroom benchmark in the background; thousands of files are being manipulated automatically. They'll do this for hours, then they'll change out one component, in this case a processor, and they'll do it all over again.
The results of these tests are a graph like this. In this very specific case, you can see, testing Photoshop, the data shows that this Intel i9, whatever, however you say this 8 core processor is the best you can currently buy. Which is strange because it's nearly a quarter of the price of the most expensive processor they sell. So the question is, how can an eight core processor run faster than a 12 core processor? Because I always thought that more cores meant it's gonna be faster.
First of all, what does a core mean, in a CPU? - It's the computer unit that can execute tasks within your operating system. And so back before 2005, it was all single core. And in order to make your computer faster, you just make that chip faster. But in around 2005, we were getting to the point where Moore's Law was starting to taper off, and we had to find other ways to make computers faster. And so now we're more under this new thing, Amdahl's Law. That talks about, okay, if you have multiple cores, multiple processors, that all have to work together, it's measurement of the efficiency of your software in being able to do that.
[Destin] What Jon is saying here is that even if you have more cores in your processor, it only makes your computer faster if the program you're running is more parallelizable. That's a word, I looked it up. - One size fits all computing, those days are over. It's tailored hardware for the application. - In the interest of not making this feel like a commercial for your company, who else is doing this in the space? 'Cause I asked you if I could come talk about all this stuff, and you agreed to build a computer for me. So let's just be clear about that. There's an exchange happening here. But who else is doing this? Or is this just you?
Well, I mean, I can't say that there's nobody else doing this, but the reason that we focus so hard on this is that we see there's such a void. You'll see a lot of places that’ll do gaming machines and consumer machines, but this world of workstations is like where server meets consumer, and it's kinda this blend in the middle, and we don't see a lot of people going after it, and that's why we are. I could take you out to the warehouse and you could see like how we actually put things together, do quality control, and image computers and whatnot.
So this is the actual box, I spec-ed out with William, being built right now. - Yeah. - Okay, and so this is the housing. - Nice little dolly you got going on here. Anyone can assemble a computer. Just like Legos, the hardware is built to click together so you can do it yourself. But there's a huge difference in assembling a computer and integrating a well-running system. According to systems engineers, all physical components in the world interact with four different interfaces.
These four interfaces are spatial, material, information, and energy. A spatial interaction is how they physically connect together, like Legos clicking together, or like with screws or something. A material interface would be some type of mass that's being moved back and forth between different sub-components, like moving hot air, or fluid through a system. They interact with information by, well, I mean this is obvious, it's a computer, data, moving around, right?
One of the more interesting interfaces in a computer system is energy. You have electricity and heat transfer you have to deal with. I just realized that that's what you are. You're computer systems engineers. Obviously, your name says systems, but systems engineering is about integrating all these different components and seeing how they work together. And if you change one thing, it affects everything else -- - Oh yeah, oh yeah.
Are you familiar with the field of systems engineering? - Well, I mean I'm empathizing with what you're saying right now. Like some example of that, almost all modern computer hardware does thermal throttling. So you're probably not gonna break something if you don't cool it adequately. But you're not gonna get the full performance out of it. And so it isn't enough just for us to see is this computer stable because it might be stable but not running at full speed. So that's why it's so important that we actually understand what the computer's gonna be used for. And that we benchmark it, and we know that it's providing the performance as expected.
[Destin] Oh, so that's why the benchmarks last six hours because you have to hit steady state. - [Jon] We have to let it all heat up, and yeah. - [Destin] Oh, so it's not just like processor steady state, power draw; it's thermal steady state. If you wanna check out how they test these interactions, go see the behind-the-scenes video on the second channel.
What's going on here? - This is our -- - [Destin] You have a laser engraver. - Yeah, we have a couple. This one we use for etching logos onto things. - [Destin] What the heck, dude? Are you about to do mine? - [Jon] Yep. - [Destin] Really? - Uh-huh. - [Destin] So you planned all this. - Yes, this was planned. - [Destin] This was planned. Look at you, being all fancy schmancy.
In summary, I have personally fallen into the trap a time or two where I went to a website and bought the most expensive machine I could afford at the time, thinking that that would buy me time. Do not do this. In my case, I need to edit, and I wanna sleep. (whimsical music) Can you put this in the... - Ooh. - So does this mean you're gonna get to eat dinner with us a lot more? - Heh, yeah. You have the honor; who's gonna do it? - [Boy] What? - [Destin] Who's gonna hit the button? I can't even hear it, can you? That's the license agreement. You're supposed to read that entire thing. - C'mon, I don't have time.
- We're getting everything ready for you, Dad. - I just like looking at the back; there's just so many boards. (laughing) - Okay, we have now optimized a computer for rendering capabilities; I'll tell you the benchmark numbers here in a second. So you heard my son talk about dinner earlier. I'm not gonna lie; that kinda hurt. So I'm very happy to say that after doing this little benchmark, that 11-hour render is now down to two and a half hours. Now I know that's an old computer; this is a newer computer, but still. That is a lot of my life back, and I'm very happy about that.
Just get robes and wizard hats and be done with this. Wizard hat, just put a wizard hat on top of your head, wear robes in here, and like, have a dragon statue or something. Oh dude, golly that feels good.