Robert Ryan

Resources are now also moved to disk if the allowed RAM is exceeded. With this addition to the resource mangement system, the limit moved from unobtainable RAM to overpriced NVMEs!

Working on my node-based GPU-accelerated terrain generator, I've implemented a system that moves resources between VRAM and RAM. Using this system, larger graphs are possible even when VRAM is limited. With RAM being unobtainable at the moment, a file-system cache is next up.

I've finally implemented image based lighting. It adds a lot to how the ocean looks. Next up will be some level of detail sampling on the ocean to get rid of specular artefacts!

For quite a while I've been working on a Fast Fourier Transform based ocean simulation in my render engine. Whilst there is still a lot to work on, I'm excited to see the result once I've added IBL to this renderer.

@Acerola_t What parameters are you using for the JONSWAP spectrum? I've found Fetch especially to be quite tricky to find a good value for.

no foam yet but I swapped from phillips spectrum to a dual JONSWAP as proposed by the devs of Atlas and the difference is crazy!!! so much more customizability and so much detail that blinn phong can't handle it and it's killing the visuals

@Acerola_t You can optimize the FFTs even further. Because the outputs you require a real valued, the spectra of those must have Hermitian symmetry. This means you can multiply one spectrum by i, add it to another and do a single complex FFT to get both results at once.

the fft is fast enough that I can calculate 5 1024x1024 maps for a much higher detail simulation and it's still way faster than the dft was when calculating a single 256x256 map, how cool! gonna work on foam next and then make it ocean sized

@Acerola_t @bobacupcake As for the ringing artifact, that seems to be caused by bad noise. In your case the bad noise is causing waves opposite of the wind to be close to equally as strong as waves in the direction of the wind, which means that the result after the FFT you get still seems plausible.

@Acerola_t @bobacupcake The line through the diagonal seems to be caused by the positive directional spreading model that Tessendorf uses. Using some other directional spreading function like the one proposed by Hasselmann or the one from Donelan Banner should provide a better result there.

phillips spectrum weirdness

@SebAaltonen Memory Aliasing is allowed but it has some constraints (#resources-memory-aliasing" target="_blank" rel="nofollow noopener">khronos.org/registry/vulka…). GPU copy from buffer to compressed block works as well.

Our virtual texture implementation on consoles was directly writing RGBA32_UINT data on top of BC blocks. But this wasn't allowed in PC DirectX. BC views are not compatible with uint views. Have to use CopyResource to do GPU->GPU copy. DirectX allows GPU copy between int->BC.

Does Vulkan support directly writing to ASTC texture in compute shader? Can I have a RGBA32_UINT view (128bit) to a ASTC texture? Or do some GPU vendors have different tiling layout between 128bit uint and 128bit ASTC blocks? Does GPU->GPU copy cmd in Vulkan support int->ASTC?

@SebAaltonen According to #choosing-a-present-mode" target="_blank" rel="nofollow noopener">github.com/KhronosGroup/V… there isn't a way to disable vsync on android due to the GPUs not having said capabilities

Is there any way to disable v-sync on Android? VK_PRESENT_MODE_IMMEDIATE_KHR renders at 16.6 ms.

@SebAaltonen Forgot to mention - It's an instance extension ofc.

@SebAaltonen Don't know if available on Android but you can try using VK_EXT_validation_features and enbale VK_VALIDATION_FEATURE_ENABLE_SYNCHRONIZATION_VALIDATION_EXT.

How do I run vkconfig for Android? I want to enable synchronization validation in the validation layer.

@SebAaltonen @dan_dreher_ @AceStapp The tutorial landscape for Vulkan really is dreadful. The only way to learn it really was to look around for any blog-post possible that goes more into detail about whatever you're looking for.

@dan_dreher_ @AceStapp Somebody could write actually good Vulkan tutorials that follow the best practices.

The Vulkan triangle sample I have been editing crashes if frame rate drops below 10 fps :) vkWaitForFences(device.device_, 1, &render.fence_, VK_TRUE, 100000000); 100000000 ns might feel like a big number, but it's only 0.1 seconds :)

Graphics Programming weekly - Issue 219 - January 23, 2022 jendrikillner.com/post/graphics-…

@MartianDays Simply said, you no longer have to create a renderpass if you want the same functionality of a renderpass with a single subpass.

The first SIGGRAPH 2021 Rendering Engine Architecture talk videos are live on YouTube: youtube.com/playlist?list=… @stevemcauley @MichalDrobot @kenpex @SebAaltonen and Tim Cooper and Michael Vance

SIGGRAPH 2021 Advances in Real-Time Rendering in Games course videos: youtube.com/watch?v=4Yxh64… @kechogarcia @thomasdeliot @Lumonix @BrianKaris @Stubbesaurus @gwihlidal @AriSilvennoinen @jasminpatry @DanielMakeGames @staerlor @andreas_brinck @GraphicsRunner @bxs3514 Enjoy!

@mirror2mask @DanBrink91 @siggraph @stevemcauley @kenpex Does this include the talks or only the slides? In the latter case, which SIGGRAPH attendee tier would I need to see the talks / courses?

@DanBrink91 @siggraph @stevemcauley @kenpex Yep, will be on …inearchitecture.realtimerendering.com post conference

This year we're cooking up something new for @siggraph with @stevemcauley @kenpex and Michael Vance from Activision - organizing a brand new (hopefully) series of courses on Rendering Engine Architecture in Games - …inearchitecture.realtimerendering.com