London's Glass Canvas produces some of the world's most beautiful architectural visualisations, from the interiors of Grade 1 listed town houses to the exteriors of contemporary high-rises. As well as stills and animations, the studio has recently been expanding into VR work, on projects ranging from health and safety applications to presenting major urban developments.
Below, Technical Director Bill Nuttall discusses the studio's restless creative spirit, the kind of hardware required to render a 25,000-pixel trompe l'oeil hoarding – and the art of creating virtual reality experiences so immersive they could kill.
AMD: How did Glass Canvas get started?
Bill Nuttall: The studio was founded in 2001 by Andrew Goodeve. Prior to setting up, Andrew was working as a visualiser at Foster + Partners, and after three years wanted to strike out on his own. At the time, there weren't as many other architectural visualisation firms in London as there are today.
I joined full-time in 2005. I was originally an architect, but preferred the more creative stages of the architectural process, in particular experimenting with the software and 3D modelling. I did a few freelance jobs for Glass Canvas, Andrew offered me a job, and I've been here ever since. Andrew and I are now co-owners.
AMD: How has the company changed during its lifetime?
BN: When I first joined, we were in an office in Shad Thames [a historic riverside street next to Tower Bridge in London] with eight or nine people. We've moved around a few times, and have been around during many of the big changes to the industry. The real killer was the recession in 2008. There was just no work. But we managed to survive that through some prudent budgeting, and since then, we've nearly tripled in size, slowly adding people.
Two years ago, we set up a sister company called Pixel & Sons, run by creative director Andrew Li, which creates CG and animation for commercials, products and brands. It’s positioned away from our core Glass Canvas brand so that we can expand our reach into other markets, away from the built environment.
Then last year, we bought into a small visualisation studio in Sheffield run by Glen Austin. He'd been in visualisation for many years, and was looking to grow his services. We were looking to reach the resurging Northern Powerhouse cities, so it was a great moment to combine. Now Glen runs Glass Canvas Sheffield and manages clients and the team there. The shared resources and infrastructure means we can load up on projects without necessarily recruiting freelancers.
AMD: How would you summarise the kinds of jobs Glass Canvas does?
BN: It really is pretty much everything across architectural visualisation. Last week, I was working on a masterplan in Casablanca, producing a two-and-a-half-minute animation, and this week, I'm working on some planning and marketing images of a mixed-use scheme in East London.
Sometimes, it's a loose creative brief where we are required to create the design almost in shorthand for a vision image or masterplan, sometimes it's fully verified planning work, and at other times, it can be really conceptual animation. At any time, we typically have 30-40 projects going across the entire Glass Canvas Group.
AMD: What software do you use in production?
BN: We use Autodesk® 3ds Max® for modelling, along with plugins like Forest Pack and RailClone. Our renderer is Chaos Group’s V-Ray. Post-production is done in Adobe Photoshop. And we use the Adobe suite for animation – After Effects, Premiere Pro and Audition – plus SynthEyes for 3D camera tracking. For virtual reality work, we use Epic Games' Unreal Engine, plus Autodesk® ReCap® for turning photographs into models for use in VR.
AMD: What about hardware?
BN: I assemble all our hardware; I have done for about 10 years. Everything in our office is put together by me by hand, so it's all hand-picked components, right up to our 30TB file server.
The last four workstations I've put together have used AMD's 16-core Ryzen® Threadripper® CPU – the 1950X – and 64GB of RAM. My graphics card of choice is currently AMD's Radeon™ Pro WX 9100. It offers the highest possible viewport quality in 3ds Max and plenty of HBM2 framebuffer memory to boot.
Rendering for us is CPU-based, and the Ryzen Threadripper machines are just fantastic for that. The value for money is incomparable to anything that Intel® makes. I wouldn't want to make a render farm out of the Ryzen Threadripper chips, as I like to put in as many CPUs as possible, but they're definitely comparable to the Intel® Xeons that are going to farms when it comes to power.
AMD: What is your internal render set-up?
BN: We have a rack with around 400 cores in total. During the day, the artists can use them as an extension of their own workstations, then at night, they're either rendering animation frames or large single-frame images: typically, 5,000 pixels by 3,750 pixels.
I haven't built any render nodes for several years, but the new ones would be made with AMD's EPYC processors, given the opportunity. They're cheaper than Intel's offerings, and will offer the same or better levels of rendering power. Like Threadripper, they offer superior bang for your buck.
AMD: What are the benefits of using the Radeon™ Pro WX 9100 GPU?
BN: You don't have to worry too much about running out of VRAM on large architectural scenes. The current industry standard for GPU memory is GDDR5, but this can be a bottleneck because of the way the memory is physically placed on the card. The 16GB of High Bandwidth Memory (HBM2) used in the Radeon Pro WX 9100 gets around the bottleneck by being stackable on the chip die. This increased bandwidth allows us to have a GPU that draws the viewports faster, with higher-quality images. All of the other [non-AMD] cards now feel puny in comparison.
AMD: What is the key benefit of the AMD CPUs you use?
BN: The price-to-performance ratio. It's currently almost double that of the equivalent Intel product. The nearest thing that Intel do is one of those high-core-count Core i9s, but those things cost almost two grand, and you can get a 16-core Threadripper for around £800.
I'm very glad that AMD is now back on the scene with the Ryzen Threadripper and EPYC CPUs, and that there is some competition in the market.
AMD: What's the most technically challenging job that Glass Canvas has worked on?
BN: The City of Wolverhampton Commercial Gateway Masterplan was pretty intensive. It's a large collection of buildings, then lots of Forest Pack objects. That really chews through memory, even though we try to make as much of the geometry as we can procedural. In the aerial view, the roofs are hand-modelled, but the facades are all RailClone objects. We had a team of four working on that one for over six weeks.
With every new project we like to deliver something more, be it detail, scope, quality or realism. Although typical job deadlines haven’t changed much in the last 15 years, client expectations have, which puts us in a continual cycle of upgrading our software, hardware and workflows.
AMD: What's the strangest job you've worked on?
BN: The viewing platform at Lewis Cubitt Park was quite bizarre. Having dumped a couple of steel shipping containers next to their building site to create a public viewing platform, the client decided they needed prettying up in some way. Our solution was to produce two enormous prints that they could drape around the structure, one on the front, and one down the side. The renders are 25,000px wide, and the detail on them is quite amazing. They burned through every available byte of RAM on our system. That was one of the most intensive bits of rendering we've had to do.
AMD: What was your most memorable virtual reality project?
BN: We did a health and safety application for a construction company, which involved taking photos of their equipment – really massive steamrollers where the front bit is as big as you – and turning them into animated models. It took one of our artists two days to turn around 100 photographs into a refined 3D model to use in Unreal Engine, via ReCap and with clean-up in 3ds Max. We were demoing the result to the client, and this poor person put the headset on, turned around, saw a steamroller coming in their direction, shouted out in panic, and ran across the room and into a wall.
AMD: You're not meant to kill the client during the demo, you know.
BN: I think the client was actually quite impressed. That really is immersive.
AMD: What's the secret of a good visualisation image?
BN: I don't think it's massively different to what you would see in a classical art gallery. One of the things you often see, especially with Renaissance art, is the use of two dominant colors: very often contrasting colors, like blues and oranges, or close harmonics, like purples and blues. And then a strong composition: the rule of thirds is still as relevant today as it's always been.
It’s also important to study the works of great photographers, cinematographers and film-makers. Look for patterns and themes in their compositions, and make note of the best camera movements and edits. You know greatness when you see it: figuring out what creates it is the key.
AMD: What's a typical working day at Glass Canvas?
BN: Very much nine-to-six. We like people to leave the office at 6pm so the job doesn't become their life. There are lots of visualisation studios that work silly hours, but we've tried to set ourselves apart from them by having a normal working day.
Maybe I'm biased, but I think we have a great atmosphere in the office. Generally, when people come to work for us, they stay for many years. We have a low turnover of staff.
The office itself is a nice big airy high-ceilinged space on the ground floor of a new development near Kings Cross. [Part of North London, served by a major rail station linking the northern part of the UK to the city.] We've been here for five years, and hopefully, we'll be here another five at least.
AMD: What do you see as the next big thing in the industry?
BN: It's not exactly a stretch to say real-time ray tracing, is it? Current algorithms are inefficient and have unrealistic hardware demands: the beautiful Star Wars storm trooper demo was famously done on $60,000 of GPUs. And the demos we've been seeing recently aren't even remotely on the scale of a full-fat Monte Carlo path tracer: we're talking at least a thousand times fewer rays being shot – maybe even a million times fewer. But it's a start, and in ten years the code will have improved, and a GPU with the equivalent power to $60,000 in today's hardware will probably cost $200.
AMD: What impact will VR have?
BN: Virtual reality is still very much trying to find its feet within architectural visualisation. Our last five VR projects have all been very different from each other. Our clients are going through the process of discovering what works best for them, so we’re trying a bit of everything, from room-scale setups to smartphones in cardboard goggles.
I don't see much of an application for VR in 3D modelling for a while. Sitting there modelling in 3D space sounds amazing, but can you imagine having to wear a headset all day? It's bad enough looking at a screen all day long without having it five centimetres away from your eyeballs.
But my predictions can be notoriously inaccurate. When I was at school, back in the 1980s, one of my predictions was that mountain bikes were just a fad. That tells you all you need to know.
Glass Canvas is a valued customer of AMD. The views of Glass Canvas are their own opinions and may not represent AMD’s positions, strategies or opinions. Any benchmarks or performance statements from Glass Canvas have not been verified by AMD. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied. GD-5
©2018 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, Ryzen, EPYC, Threadripper, Radeon and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies. GD-28