Ryan Vance

@DS2LightingMod When you say 'previous denoised data', do you mean using the previous frame's reprojected radiance for the tail if the hit point from the first bounce is on screen and visible instead of sharc and then falling back to sharc/full trace otherwise?

LE does per pixel path tracing , multi-bounce is supported using Nvidia SHARC and previous denoised data (diffuse/spec seperate). 1st is no SHARC/No Previous. 2nd is with Previous, 3rd is with SHARC and Previous. If all cache misses it computes the lighting at the point.

@iveytron I thought this was a CD player boom box.

Love that my new rice cooker has stickers and labels that make it look like a 1998 PC purchased from Walmart

(If not, you’re welcome whoever gets to it first)

Has anyone made a Warhammer 401k joke or is it just sitting there waiting for the right home?

@EricLengyel @C4Engine Any paper/talk/code reference suggestions for this?

Doing some actual graphics development! Just finished a filmic tone mapping upgrade in the @C4Engine with histogram analysis and adaptive exposure.

@maxliani @savage309 @lauriewired Could you give me an example of 'live state packing'?

@savage309 @rgba8 @lauriewired Strongly agree. Some offline renderers architectures may require modularity for plugins architecture in the renderer. That’s very hard to deploy (without implementing JIT compilers).

Pixar doesn't use GPUs (much). Their render farm compute is mostly CPUs with a ton of cores + memory, using AVX-512 and SSE 4.2 optimizations when appropriate to optimize render time. Why? GPU render compute doesn't speed things up as much as you might expect. Yes, in certain instances, you can get a 3-4x benefit, however you have to fit all the polys in GPU memory. Paging back out to system RAM brings you back down to CPU speeds. Scenes can often contain > 150,000,000 polys. Only very recently have gpu memory sizes gotten big enough to be "worth it" over CPU rendering, and even then, it's often less flexible/tolerant of mixing older architectures in the cluster.

@maxliani @lauriewired I'd love to hear more about your approach, the results are fantastic. SER helps with shading divergence, but a reasonable feature set tends to bloat the payload which made wavefront look desirable. Curious what you mean by hybrid technique.

@rgba8 @lauriewired In my timeline you see tests with 1000 ray bounces… those are very unpractical with a wavefront renderer.

@maxliani @lauriewired Appreciate the reply! Is a simpler, more performant renderer still wavefront, just with less kernels? Avoiding death by too many launches and death by too large a kernel by reducing feature set or something else?

@rgba8 @lauriewired My direction, which got abandoned, was to create a simpler, much more performant renderer, that was running near realtime, and retire a list of features that were determined to hit performance too hard. Or at least do so until an high performance alternative was discovered.

@maxliani @lauriewired What sort of architecture design do you think would perform better? I'm aware of the pitfalls of mega kernels. What are the issues with a wavefront/ extremely fine grained micro kernels?

Pixar decided to implement the GPU renderer in a way to make it as feature-compatible as possible to the existing CPU renderer. That poses design and performance limitations on the GPU. Going down the path of a wavefront architecture with extremely fine grained micro kernels. I know for a fact that it doesn’t have to be that way… and much higher performance can be achieved with an architecture design that accounts more deeply for the hardware.

Twitter hive mind.. Anyone have an implementation of 'Ray Tracing Animated Scenes using Coherent Grid Traversal' sci.utah.edu/~wald/Publicat… It's straight forward, but it'd be helpful to compare with code that works. 2006 was in the before times when code wasn't posted with papers.

@Pjbomb2 @turanszkij github.com/knarkowicz/GPU…

@turanszkij How are you compressing bc) in real time? I wasn’t aware of a good gpu compressor that existed yet for bc6

Added better directional irradiance to surfel GI, and it's compressed to BC6 in real time. The bottom shows a debug view:

@adam_orth Rookie numbers.

@KostasAAA @SebAaltonen I’d suggest pbrt.org too for gfx. It’s good to have a baseline for what you’re aspiring to and having an implementation along with the theory was really helpful for me.

@SebAaltonen Game dev or gfx programming specifically? If the latter buy Real Time Rendering, get the simplest sample from the DirectX SDK (to avoid all the boilerplate) and start coding. Reach out to fora and Twitter for questions. APIs are more complex but there is much more help nowadays.

What are the best sources for learning game dev? I read code listings from a local magazine and learned how to enable 320x200 pixel VGA model + hook keyboard interrupts in MS-DOS from a post in a local BBS system, which I called using my father's landline connection.

@penberg My suggestion: fgiesen.wordpress.com/2011/07/09/a-t…

I have some really basic understanding of GPU programming with shaders and CUDA, but am mostly clueless of actual GPU architecture and how code gets executed on it. What books or blog posts would you recommend me to read to catch up?

New adventure!

@cmuratori q3dm17

Doom E1M5

@reduzio I've found 'death by 1000 cuts' to more prevalent with memory than compute.

@_benui Skippable *AND* support pausing. Skipping and pausing should be different input actions which are labeled.

@_benui voidtools.com

@ephtracy link.springer.com/content/pdf/10… Though you’re probably not tracing triangles

did some tricky work to reduce the artifacts of ray-surface self intersection. maybe one day will find a more robust solution...

I've been on here since 2010 and haven't really posted anything, the imposter syndrome is strong. Graphics twitter is amazing though, so I thought I'd finally try contributing. Working on a side project that took a hardware path tracing detour... #screenshotsaturday