Adam Sawicki

Adam Sawicki

@asawicki.info

Independent graphics programmer. Previously at AMD, Intel, Microsoft, gamedev, television... Developer of VMA, D3D12MA, D3d12info projects.

1K followers494 following116 posts

The more examples of DLSS 5 I see, the more I feel convinced that this is the way to fill the uncanny valley. Rendered characters are so far from photorealism, while the advancements in meshes, textures, lighting, subsurface scattering etc. provide diminishing returns. ⬇️

Apparently GPT-5.6 comes with so many dimensions to adjust: - Model version: 5.6 vs 5.5, ... - Model size: sol, terra, luna - Reasoning effort: none...max - Reasoning mode = pro It will be challenging to find good practices about when to use which ones.

I'm excited to see lore.org, Epic's new version control system. I still need to read the docs, but after experiencing the shortcomings of both Git and Perforce, I'm hoping Lore can gain traction in at least some areas of software development, especially gamedev.

lore.orgLearn about Lore: next-generation open source version controlMaintained by Epic Games, Lore is designed for unprecedented scalability of both data and teams. It’s optimized for projects that combine code with large binary assets.

Vulkan Memory Allocator new release 3.4.0 with many additions, improvements, and fixes. If you use code from the "master" branch, you already have them, but if you rely on the official Releases, you should upgrade. This one gathers >1 year of development. github.com/GPUOpen-Libr...

github.comRelease v3.4.0 · GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocatorAdded file AGENTS.md for agentic AI. Added member VmaAllocationCreateInfo::minAlignment (#523). Remember to always fully initialize structures with zeros and don't rely on their specific sizeof to...

I implemented multiple fixes & improvements in Vulkan/D3D12 Memory Allocator libraries, on my fork for now. If you use any of these, please give it a try so we validate they work fine before I turn them into official releases. github.com/sawickiap/Vu... github.com/sawickiap/D3...

github.comGitHub - sawickiap/VulkanMemoryAllocator: Easy to integrate Vulkan memory allocation libraryEasy to integrate Vulkan memory allocation library - sawickiap/VulkanMemoryAllocator

A wacky idea for a game (an indie game or a game jam entry) that makes you feel like like an LLM: You're shown a short chat session with back-and-forth with a "user". Your task is to provide the next word of a response. Just one word. Then, you're switched to a different chat. 🧵

This would simulate how LLMs work: no chance to form a persistent memory of a chat, just general knowledge you have + the current context window = generating next token. The difficulty of this would also demonstrate why these calculations are so expensive.

Not sure if the “user” side should be other players or fixed scenarios. But surely, LLM responses should be scored by the quality/correctness as well as speed.

#UIComplaint PIX on Windows can export binary contents of a buffer, but only if you know where to click. "Export View" button only offers TXT and CSV format. You need to right-click on the data to find the "Save Buffer" command for saving a binary. No button for that 😟

I vibe-coded a TXT to binary converter before I learned this command is available. RenderDoc does this better, offering "Export to CSV" next to "Export to Bytes".

This article is VISIONARY, discussing the past, current state, and preferred future of graphics APIs. Someone please hire him as a CTO to make this new API happen!

Sebastian Aaltonen@sebaaltonen.bsky.social

My "No Graphics API" blog post is live! Please repost :) www.sebastianaaltonen.com/blog/no-grap... I spend 1.5 years doing this. Full rewrite last summer and another partial rewrite last month. As Hemingway said: "First draft of everything is always shit".