Tim Sweeney

speaker
640 appearances 1 recordings 1 series first heard Apr 2025 last heard Apr 2025

Tim Sweeney’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months
No recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.

Appearances

newest first · ▶ plays the moment
His seventh or eighth try after he'd done something time and time again, tried it, found a better approach, thrown out the old one, built it again, and continually rewrote his code until he found the absolute best solution to a problem. And I think that stands as a lesson for every programmer to pick up on.
When something is really, really important, its performance is absolutely critical to the product, or its quality, or its capabilities, just Iterate on it until you've achieved perfection, and don't settle for the first or second solution is good enough.
It's been an astonishing experience. Nobody 30 years ago had anticipated that we'd see the performance gains in hardware that we've actually seen in that timeframe. It's something like 100,000 times higher CPU performance between multiple cores and higher clock rates and more parallelism. If we had that in aviation, then we'd be taking a trip to neighboring stars. Alpha Centauri, yeah. Exactly.
And in graphics, it's been even more so. It's something like literally 10 million times more net usable GPU performance than we had back running on a Pentium 90 CPU, all in 30 years. And it's really made me appreciate that over the generations, some areas of our engine development have absolutely kept up with technology.
And the rendering team that works on Unreal Engine are the real miracle workers there. Just about every generation of Unreal, we've replaced most of the rendering code. And the different leaders in different points in time and the different luminaries have built systems that were absolutely rethought and optimized for the latest generation of hardware.
Unreal Engine 1 was built for software rendering, and then the Voodoo 1 came along late in the cycle. And we had support for it, but it wasn't fully fully capable and utilized. Unreal Engine 2 was about bringing all the latest GPU hardware acceleration features to the Engine and keeping forward and building some new features like vehicles and a few other capabilities.
And this was in the early GPU era, before GPUs had really broken out of everybody's expectations of Moore's Law. But that breakout occurred with DirectX 9 and the capabilities of programmable shaders.
Once you had control of writing code running on the GPU that could color every pixel on the screen, and that GPU code was literally a factor of 100 times faster than the equivalent code I wrote a few years earlier on the Pentium 90. And so that DirectX 9 generation was a godsend.
And Andrew Scheiderker, a longtime Epic luminary, wrote the core of the Unreal Engine 3 render around real-time pixel shading, real-time lighting, being able to do dynamic shadows using several different techniques, and multi-thread the render to support bits of the early dual-core CPUs that were starting to show up at the time. And it was a massive, massive graphical upgrade.
Unreal Engine 4 made a number of improvements and just continued to add features to give artists more and more options for lighting and for geometry that created realism.
But then I think probably our biggest single level of leap came with Unreal Engine 5 with Nanite micropolygon geometry solution and with Lumen, the global illumination lighting solution, which I think really bridged the gap from game-ish computer graphics to... total observable photorealism for artists who wanted to create that. And so that's been the evolution.
And the progress on the graphics side is absolutely astonishing, as it is on the audio side in a number of other areas. But parts of the engine also haven't changed all that much since the version I wrote and shipped in 1998. You know, the file management system has been optimized a number of times, but it hasn't been completely rethought.
And the networking system, the ways that clients and servers talk together and negotiate game state is still an evolution of the thing I wrote. And, you know, it's feeling kind of dated now. You still see networking bugs in Fortnite where, for some reason, when you're spectating, you're not seeing some parameters update. Well, that's because of the lossful nature of that networking model.
And the biggest limitation that's built up over time is the single-threaded nature of game simulation in Unreal Engine. We run a single-threaded simulation. If you have a 16-core CPU, we're using one core for game simulation and running the complicated game logic because single-thread programming is orders of magnitude easier than multi-thread programming.
And we didn't want to burden either ourselves or our partners or the community with the complications of multi-threading. And over time, that becomes an increasing limitation. So we're really thinking about and working on the next generation of technology and being Unreal Engine 6.
And that's the generation where we're actually going to go and address a number of the really core limitations that have been with us over the history of Unreal Engine and and get those on a better foundation that the modern world deserves, given everything that's been learned in the field of computing in that timeframe.
There's a lot happening here on screen. The real hero of this image isn't Epic. It's the artists and technical artists who worked together to build this environment. The reason we showed it at GDC was it went way, way beyond what we realized the system was capable of doing, largely because of their brilliance. This is the magic of computer graphics. There's not one feature that makes this cool.
There's a dozen technical features that each interplay. Because of the ways that they... interplay with each other. It's hard to actually identify the individual components of it. One thing that's happening here that's really critical, oh yeah, now we're seeing it being turned off, is the lighting happening.
The lumen lighting system that's powering the scene is doing different kinds of lighting calculations at different scales. This was the work of Daniel Wright following a decade of moving the state of the art of lighting forward. But his theory, which was rather controversial at the time, was that if you have enough levels of lighting calculation, then you can get everything...
global illumination working everywhere from the absolute highest levels of a scene, you know, that buildings are casting cracked shadows all the way down to details like you see on the dirt here, all working in concert and without distinguishable boundaries. So there is a good decade of foundational work there to make the lighting work. In particular, when you see that
Showing 201–220 of 640 · page 11 of 32 ← Previous Next →