Check out www.VFXRates.com
Episode 106 — Redshift Renderer
This is Allan McKay. Welcome to Episode 106! I’m speaking with Nicolas Burtnyk, Co-Founder and CEO of Redshift, and Rob Slater, Co-Founder and Vice President of Engineering.
Let’s dive in!
FIRST THINGS FIRST:
I. [-04:04] I’ve had this brilliant idea of doing 3 Episodes per week. I’m really excited about it! Moving forward, I’m going to start theming the months: Each month will be dedicated to a particular subject — in addition to my regular interview Episodes. We can start going through these together and build a community.
– Every week, you will get two Episodes on a particular subject.
– You will get videos on the same subject.
– You will receive an accompanying guide.
We will tackle everything on:
– How to set up your side business;
– How to make money;
And much more!
If you like this, please let me know. It’s crucial that you communicate that to me, along with possible subjects you would want me to tackle in the future.
II. [-41:24] I am also planning to do a new private Career Intensive, as well as a Portfolio Review. If you want to be a part of it, please make sure you’re on my Insiders’ List. Please sign up for it for free at: www.allanmckay.com/inside/.
III. [-41:03] I have a new Free Training coming out in about a week from now, around on November 28th. If you want to have access to it, please sign up for the Insiders’ List as well.
INTERVIEW WITH REDSHIFT DEVELOPERS
Redshift is the world’s first fully GPU-accelerated biased renderer. It is built to meet the specific demands of contemporary high-end production rendering. It provides the most features and flexibility among all GPU renderers.
In this Podcast, Allan McKay talks with Nicolas Burtnyk, Co-Founder and CEO, as well as Rob Slater, Co-Founder and VP of Engineering. They discuss the benefits of GPU vs CPU rendering, the software Redshift currently supports and this renderer’s future direction.
Redshift Website: https://www.redshift3d.com
Redshift About Page: https://www.redshift3d.com/about
Redshift on Twitter: @redshift3d
Redshift on Facebook: www.facebook.com/redshift3d/
Schirkoa Short Film (Created with Maya and Redshift): https://vimeo.com/170445870
[-40:11] Allan: Do you mind quickly introducing yourselves?
Nicolas: My name is Nicolas Burtnyk. I’m one of the Co-Founders of Redshift. That’s me!
Rob: I’m Rob Slater. I’m also a Co-Founder as well as a VP of Engineering at Redshift.
[-39:41] Allan: Do you want to explain in a nutshell what Redshift is?
Rob: Redshift is a biased production quality render. That means that we have the same feature sets as V-Ray and Arnold but we use GPU to accelerate ray tracing, shading, texturing, all the heavy lifting; so consequently we’re about 10 times faster than CPU renderers.
[-39:08] Allan: That’s awesome! What studios are using Redshift at the moment? I know Digital Domain is diving into it.
Rob: Probably the most well-known is Blizzard Entertainment. They used Redshift for their overwatch cinematics, really nice scenes that push Redshift really hard. I believe they’re also using Redshift for their hearthstone cinematics. You can see all of these on YouTube.
[-38:38] Allan: Awesome! I guess it’s been convenient to have them right down the street from you, guys.
Rob: It helps us expand our features. If a big company comes knocking on our door asking for certain features, then probably other companies will need those as well.
[-38:14] Allan: Whenever I’ve found myself on a production that has collaborated with a software company, it’s been really great: for production to have the attention of developers to implement the features that we need and the software companies get exponential growth from that. Recently, I’ve interviewed Cebas (www.allanmckay.com/82/.) There is a few movies where we collaborated with them. On 2012, nobody had a solution for making big destructions, so Cebas was custom developing everything.
Rob: For us, it’s a very organic process because we’re talking to our customers and our potential customers all the time — and we listen, we monitor the forums. If we see features that are really popular, we develop a strategy.
[-36:53] Allan: That’s really cool. Going back, how do you guys all know each other?
Rob: Well, we all used to work for the same games company. So Panas [Zompolas], also a Co-Founder, and us, we used to work together in the U.K. for many years. I moved here and we met Nicolas. We had the idea of Redshift because a friend of ours, who is an artist, had a question. We were doing realtime rendering for video games and it looked pretty good. [Our friend asked us,] “Why is nobody trying to accelerate ray tracing for film, for final rendering”. Because it’s really hard! We liked the challenge and tried to see if we could knock out a prototype. Very early on, we found out it was actually possible. That encouraged us to take it seriously and that’s how Redshift started.
[-35:38] Allan: I’m sure there were some beers involved somewhere along the way. For you, guys, to do that: Was it something you decided to go all balls in, or were you dabbling on the side until you figured out you had something cool.
Rob: Well, it was dabbling on the side, basically, just to see if it was even feasible as an idea. Our first prototype was super basic: 640 X 480. We are programers, we are not artists. We had ray trace shadows that were textured, and that was running at 30 frames per second. In my eyes that was quite impressive at the time. We thought we could have something pretty special in our hands. At that point, we went balls deep, as you said.
Nicolas: Just to put some perspective on it, that was 7-8 years ago. The technology improved drastically since then. We’re talking about early generation GPU’s, that kind of performance was pretty impressive.
Rob: Basically, pre-CUDA.
[-33:25] Allan: I want to get into the tech side pretty soon. What were some of the hurdles you had to overcome? So many people were stuck with CPU and not thinking about pioneering ways to approach it.
Rob: There was a ton. For a start, we had to think about parallel programing. GPU does so many operations at once, but all the algorithms that you use for ray tracing, we had to rethink them so they would fit on the GPU — and we had to do it right. Because if you don’t, you’ll get a worse performance than you would with a really good CPU renderer. So we had to think long and hard, to make sure we got it right. We also had to think about the data footprint: How can we take a scene and make it fit? Back then, 500 megs of memory probably was the maximum. To make decent size scenes fit into that memory, with textures, geometry. Texture compression was something we had to think about really early on.
– Basically Redshift was written without the core rendering in mind. You could take any size scene — and it didn’t matter how much memory the GPU actually had physically — how could you render that scene without the GPU crashing? That was a challenge.
– The flexibility: How can we get enough features, like shader graphs, tons of stuff that CPU renderer already supported? We had to cramp all that stuff on the GPU and make it blazingly fast.
[-31:09] Allan: When was it that you realized you had something? And what was that like in the beginning? It takes a lot of work to get established, to get up and running, to make a splash in the market. Going back 5-8 years ago, the market was really competitive.
Nicolas: One of the challenges was that we were experts in a sort of a different space. We were working in games. While the team had the expertise regarding the GPU — knowing the GPU and how to take advantage of its power — we didn’t necessarily completely understand the space of off-line rendering or final frame rendering in detail. You think you know what people are going to want. But when you actually release your product, all kinds of things comes up. Rob has touched on the insane number of features in a program like Maya that you need to support. And then, there are all kinds of obscure pipelines and workflows that could be unique to a particular studio. Wrapping our minds around those, learning those, distilling them down to a workable set of tasks was a challenge.
In terms of how we approached going to market, we knew we had something special and unique. We didn’t necessarily have the business development resources to go and install ourselves in a particular studio. We actually released the product by announcing ourselves on forums like CG Talk and the 3d Mailing List, and the Softimage Mailing List. Our first version was only integrated on Softimage. In the beginning, it helped because it was a very focused community. That community was really technical too. They were small shops and freelancers [that] were willing to take on a new product and test it out. It’s not only about the performance but about getting the job done. We were able to get in with those customers and work with them into developing it into a production renderer.
[-27:55] Allan: I agree that seeing a single focused packages is great at the beginning. A lot of renderers came from 3DS Max. I didn’t catch on until Alf Lovvold (www.allanmckay.com/39/) told me what he’d been doing: One show, on one computer, multiple cards. It’s a huge mind shift. When I used to work at Blur, you’d have the entire render farm at your disposal and it still wasn’t enough. Here it is, one filmmaker — one computer — and you’re able to do whatever you want. It was a mind shift on what limitation we’d have in the future. It was really inspiring to see that capability. What was your impression seeing Alf pop up on the radar?
Rob: It was great. It was during the early days. We were making the renderer and art work was starting to come in from Alfa testers and Beta testers. So it was exciting seeing what people were creating. And when Alf came along, it was the maddest thing we’d ever seen. It was fantastic!
[-25:11] Allan: Did he ever give you a bottle of that Norwegian alcohol?
Rob: Of course, he did!
[-24:55] Allan: It was disgusting.
Rob: You need to spend more time in the colder parts of the world.
[-24:31] Allan: Did you have other people popping up on the radar, doing amazing things with [Redshift]?
Rob: Yeah, loads!
Nicolas: There is the Schirkoa short film that’s winning all kinds of awards: https://vimeo.com/170445870. It’s done in a painterly style, also [with] a very small team.
Rob: They did it on their own time.
Nicolas: It’s taken [creator Ishan Shukla] 2 years to finish that but he’s killing it in the festival circuit.
[-23:36] Allan: For someone who isn’t too familiar, what are the advantages these days with GPU over CPU?
Nicolas: It mostly comes down to speed but speed doesn’t only mean to get renders out of the farm quicker and save money. That’s one big thing: you save money! The faster it is, the less hardware you need to do the same amount of work. A more subtle but more important side effect of that is it improves iteration time. The faster your renders come out, the more artistic you can be. You can ultimately end up with better results because you were able to iterate on your work. Those are the main two advantages.
Rob: For small shops and freelancers, they now have the power of a small render farm at their fingertips. Now they can compete with larger shops for less money. All you have to do is get a Titan.
[-22:21] Allan: I worked a lot with Atomic Fiction that was the poster boy for cloud rendering. You’re a victim of your own success. If you deliver the work, studios will give you more and more work. The solution is to get more artists. But you’re not thinking about the overhead: electricity, air conditioning, software licenses, etc. What do you do afterwards when you’ve gotten more power? Nothing is more expensive than man power. That real time feedback is worth its weight in gold!
Rob: One of the cool things about GPU, it’s cranking out new architectures every year. You don’t necessarily need to grow your render farm. You just get better hardware.
[-20:26] Allan: That’s one thing at ILM I found interesting. A few IT guys were discussing the move to GPU. When you buy super fast CPU’s, they become redundant quite quickly because new stuff comes in and it gets old really quickly. With GPU, you buy it one time and then at some point you’ll update to a new video card. That video card can be handed down to someone in a different department. Suddenly, it becomes an efficient way of recycling.
Nicolas: And the other thing, in general, it’s a lot easier to replace a GPU. When you’re replacing a CPU, you’re replacing the whole machine. The socket will be different, you’ll need a new motherboard, and so on. Whereas with a GPU, you have a drop-in replacement. You keep the same case and power supplies.
[-19:04] Allan: With some of the external GPU accelerator boxes — where you can buy the external video cards these days — have you found it’s a beneficial feature?
Nicolas: It’s a hit and miss. The hardware isn’t always super high quality.
Rob: What was the one we tested that was pretty good?
Nicolas: The Cubix. That one did perform extremely well. The price is pretty high, so it’s not clear if it’s worth it — instead of getting a second machine, for example. With these external boxes, it’s hit and miss. Where it helps a lot is with systems that don’t necessarily have a CUDA-enabled GPU onboard, like laptops or Mac computers.
[-17:30] Allan: When CUDA got implemented, how much did that change things for you, guys? Has it been a pretty big breakthrough in terms of how you work with GPU?
Rob: Yeah, totally. Our very first prototype was using DirectX 9, which is really old. It had so many limitation. With CUDA, it opened so many opportunities on how we could write this thing. Just how we can access memory, for example. That was the number one thing that was so important! Given so many options, and how it can synchronize better, the tricks you could do. CUDA changed everything. To be honest, without CUDA, we couldn’t have made Redshift. It would’ve have been a very different product.
[-16:29] Allan: Feel free to go in depth, by the way. Most listeners of the Podcast are technical. If you want to elaborate on what CUDA is and how it’s benefited you.
Rob: Mostly, it was about how you could access memory, in non-coalesced fashion; being able to read and write from anywhere, anytime. Just using that flexibility. It was also [about being] able to read data through texture cache that’s nice and fast. Synchronization primitives for doing sneaky tricks. CUDA is a great language, really flexible.
[-15:26] Allan: What are you views on V-Ray and few other renderers that are starting to look into hybrid GPU and CPU together?
Rob: Panos and I actually saw a presentation. Our first question was: What’s the point? I guess they’re doing it because they want to take advantage of the spare CPU cycle while the GPU is cranking. I don’t think we have any plans to do that, if that’s what you’re asking.
[-14:51] Allan: Have you found that a lot of cloud rendering services are catching with GPU?
Nicolas: There is definitely a bit of catch-up still happening but we’re getting requests at least once a week from some new farm. It ranges from full-service farms — or they have some plug-in — to the ones that are trying to rent you the software or rent cloud machines that are preconfigured with software that you could use within your pipeline. All those different services are popping up pretty quickly. And I assume it’s because they’re getting interest from their customer base.
[-13:39] Allan: That’s really cool! I’m going to throw in some basic questions. What are all the packages you currently support with Redshift?
Nicolas: So, Maya, 3DS Max, Softimage, Cinema 4D, Houdini and Katana.
[13:21] Allan: I’ve spoken to you before about FumeFX, have there been any improvements?
Nicolas: There hasn’t been any direct progress there, but the conversation is still happening. We got an email this morning asking if we were interested in supporting their cache files directly. And of course, we are. So that will happen at some point.
[-12:34] Allan: Once Kresimir hands that over, he’s opening all the doors. That’s the last thing that’s holding me from going GPU completely. There is no reason to look back.
Rob: Actually, we have potentially very large customers who initially were interested in Redshift because we render volumes really fast.
[-11:20] Allan: I was hanging out with some guys at Digital Domain. They’re just finished a commercial in V-Ray. And just for shits and giggles, they decided to redo it in Redshift. They were able to get pixel perfect comparison between the two; but the difference was one V-Ray was 8 hours per frame, whereas Redshift was 3 minutes per frame. This is just as I was discovering that. I think a lot of people are starting to rethink their approach.
Nicolas: We hear of those super impressive numbers. I wouldn’t say that’s what you can expect from every scene. It depends on which hardware you’re using, which GPU vs which CPU.
[-09:57] Allan: I think it’s so cool what you’re doing. I want to use this as opportunity to shed light on benefits of Redshift. For anyone who isn’t aware, what are the key features that set it aside / the things that you can publicly talk about?
Rob: One of the things we’re about to come out with is custom AOV’s. The large film studios need to have that power. We were locked before. Custom AOV’s changes the game. Right from the beginning, people were asking about shders because they wanted to write their own procedural shaders or BRDF’s. Of course, we have our own internal shader language. We’re going to clean it up. Some of the new features: trace sets (some of the large studios have been asking for it).
Nicolas: There is a whole host of individual features that we need to support. For example, in Maya, there is XGen Interactive Grooming, XGen Scattering. A lot of these third party plug-ins. There is a laundry list.
Rob: One of the features we’re working on, Allan, is ray trace subsurface scattering. We already have a point base, which is really fast. But it’s not good for fine detail. Very encouraged by how that’s coming along! But also directional lighting: If you have an area light, there is a way to focus that light in a specific direction.
Nicolas: And sort of in the immediate term, our big push is on the interactivity. While we have one of the fastest final frame renderers, we haven’t done the best job on time to first pixel. Making that faster and smarter and how it rescans the scene, so it doesn’t have to do full scene exports all the time. Basically, making it so you can interactively create your scene with your render window open. That’s what we’re putting our effort into right now!
[-04:50] Allan: Yeah, that’s huge! Removing the barrier of waiting to see the scene is the Holy Grail for everyone. That’s fascinating to see where it’s going to go! Have you thought about licensing your renderer for video games as well? Is that something you would look into?
Nicolas: We’re always open to those types of discussions. It’s not something we can talk about right now.
[-03:25] Allan: Where would people go to follow you?
Nicolas: Our website is www.redshift3d.com. You can check us out on Facebook, Twitter.
[-02:10] Allan: Thanks, guys! It’s been a lot of fun.
Rob: Nice talking to you!
I hope you found this Episode enjoyable. Please leave a review on iTunes. Please share this Episode around.
If you want to get into the insider circle, please go to sign up for www.allanmckay.com/inside/.