All blog entries, ordered from most recent. Entry count: 1178.
# D3d12info app and online GPU database
Tue
29
Apr 2025
This post is about D3d12info open-source project that I'm involved in. The project is in continuous development, while I noticed I didn't blog about it since I first announced it in 2022. Here, I describe the story behind it and the current state of it. The post may be interesting to you if you are a programmer coding graphics for Windows using DirectX 12.
Various GPUs (discrete graphics cards, processor integrated graphics chips) from various vendors (AMD, Intel, Nvidia, …) have various capabilities. Even when a GPU supports a specific API (OpenGL, DirectX 11, DirectX 12, Vulkan), some of the features may not be supported. These features span from the big ones that even non-programmers recognize, like ray tracing, to the most obscure, like the lengthy D3D12_FEATURE_DATA_D3D12_OPTIONS::VPAndRTArrayIndexFromAnyShaderFeedingRasterizerSupportedWithoutGSEmulation
and even lengthier Vulkan VkPhysicalDeviceShaderIntegerDotProductProperties::integerDotProductAccumulatingSaturating4x8BitPackedMixedSignednessAccelerated
🙂
Before using any of these features in our apps, we need to query if the feature is supported on the current GPU. Checking it programmatically is relatively simple. Graphics APIs offer functions for that purpose, like ID3D12Device::CheckFeatureSupport
and vkGetPhysicalDeviceProperties2
. When the feature is not supported, the app should either fall back to some other implementation (e.g. using screen-space reflections instead of ray-traced reflections) or display an error telling that the GPU doesn’t meet our minimum hardware requirements.
However, when we plan using some optional feature of the API and we think about testing it on a variety of platforms and eventually shipping it to end users, we may ask:
For Vulkan, answers to these questions are: yes & yes. For querying the capabilities of the local GPU, Vulkan SDK comes with a small command-line program called “vulkaninfo”. After calling it, we can see all the extensions, properties, features, and limits of the GPU, in a human-readable text format. Alternatively, JSON and HTML format is also available.
For the database of GPUs, Sascha Willems maintains Vulkan Hardware Database and an accompanying GUI app Vulkan Hardware Capability Viewer that presents the capabilities of the local GPU and also allows submitting this report to the database.
For Direct3D 12, however, I wasn’t aware of any such application or database. Windows SDK comes with a GUI app that can be found in "c:\Program Files (x86)\Windows Kits\10\bin\*\x64\dxcapsviewer.exe". It presents some features of DirectDraw, Direct3D9, 11, DXGI, also some options of Direct3D12, but doesn’t seem to be complete in terms of all the latest options available. There is no updated version of it distributed with DirectX 12 Agility SDK. There is also no way to use it from command line. At least Microsoft open sourced it: DxCapsViewer @ GitHub.
This is why I decided to develop D3d12info, to become a DX12 equivalent of vulkaninfo. Written in C++, this small Windows console app prints all the capabilities of DX12 to the standard output, in text format. The project is open source, under MIT license, but you can also download precompiled binaries by picking the latest release.
JSON is also available as the output format, which makes the app suitable for automated processing as part of some larger pipeline.
I published first “draft” version of D3d12info in 2018, but it wasn’t until July 2022 that I released first version I considered complete and marked as 1.0.0. The app had many releases since then. I update it as Microsoft ships new versions of the Agility SDK to fetch newly added capabilities (including ones from “preview” version of the SDK).
There are some other information fetched and printed by the app apart from DX12 features. The ones I consider most important are:
However, I try to limit the scope of the project to avoid feature creep, so I refuse some feature requests. For example, I decided not to include capabilities queried from DirectX Video or WDDM.
D3d12info would stay only a command-line tool without Dmytro Bulatov “Devaniti” - a Ukrainian developer working at ZibraAI, who joined the project and developed D3d12infoGUI. This app is a convenient overlay that unpacks the command-line D3d12info, launches it, converts its output into a nicely looking HTML page, which is then saved to a temporary file and opened in a web browser. This allows browsing capabilities of the current GPU in a convenient way. Dmytro also contributed significantly to the code of my D3d12info project.
If you scroll down the report, you can see a table with texture formats and the capabilities they support. Many of them are mandatory for every GPU supporting feature level 12_0, which are marked by a hollow check mark. However, as you can see below, my GPU supports some additional formats as “UAV Typed Load”:
The web page with the report also offers a large green button near the top that submits it to the online database. Here comes the last part of the ecosystem: D3d12infoDB. This is something I was dreaming about for years, but I couldn’t make it since I am not a proficient web developer. Now, Dmytro along with other contributors from the open source community developed a website that gathers reports about various GPUs, offering multiple ways of browsing, searching, and filtering them.
One great feature they’ve added recently is Feature Table. It gathers DX12 capabilities as rows, while columns are subsequent generations of the GPUs from AMD, Nvidia, Intel, and Qualcomm. This way, we can easily see which features are supported by older GPU generations to make a better decision about minimum feature set required by the game we develop. For example, we can see that ray tracing (DXR 1.1) and mesh shaders are supported by Nvidia since Turning architecture (GeForce RTX 2000 series, released in 2018), but support from AMD is more recent, since RDNA2 architecture (Radeon RX 6000 series, released in 2020).
As I mentioned above, I keep the D3d12info tool updated to the latest DirectX 12 Agility SDK, to fetch and print newly added capabilities. This also includes major features like DirectSR or metacommands. D3d12infoGUI app and D3d12infoDB website are also updated frequently.
I want to avoid expanding my app too much. One major feature I consider adding is a separate executable for x86 32-bit, x86 64-bit, and ARM architecture, as I heard there are differences in DX12 capabilities supported between them, while some graphics programmers (e.g. on the demoscene) still target 32 bits. Please let me know if it would be useful to you!
Finally, here is my call to action! You can help the project by submitting your GPU to the online database. Every submission counts. Even having a different version of the graphics driver constitutes a separate entry. Please download the latest D3d12infoGUI release, launch it, and when the local web page opens, press that large green button to submit your report.
Only if you are one of those developers working for a GPU vendor and you use some prototype future GPU hardware or an internal unreleased build of the graphics driver, then, please don’t do it. We don’t want to leak any confidential information through this website. If you accidentally submitted such report, please contact us and we will remove it.
Comments | #productions #directx Share
# Fixing Godot 4.3 Hang on ASUS TUF Gaming Laptop
Wed
26
Mar 2025
In January 2025, I participated in PolyJam - a Global Game Jam site in Warsaw, Poland. I shared my experiences in a blog post: Global Game Jam 2025 and First Impressions from Godot. This post focuses on a specific issue I encountered during the jam: Godot 4.3 frequently hanging on my ASUS TUF Gaming laptop. If you're in a hurry, you can SCROLL DOWN to skip straight to the solution that worked for me.
The laptop I used was an ASUS TUF Gaming FX505DY. Interestingly, it has two different AMD GPUs onboard - a detail that becomes important later:
The game we developed wasn’t particularly complex or demanding - it was a 2D pixel art project. Yet, the Godot editor kept freezing frequently, even without running the game. The hangs occurred at random moments, often while simply navigating the editor UI. Each time, I had to force-close and restart the process. I was using Godot 4.3 Stable at the time.
I needed a quick solution. My first step was verifying that both Godot 4.3 and my AMD graphics drivers were up to date (they were). Then, I launched Godot via "Godot_v4.3-stable_win64_console.exe", which displays a console window with debug logs alongside the editor. That’s when I noticed an error message appearing every time the hang occurred:
ERROR: Condition "err != VK_SUCCESS" is true. Returning: FAILED
at: command_queue_execute_and_present (drivers/vulkan/rendering_device_driver_vulkan.cpp:2266)
This suggested the issue might be GPU-related, specifically involving the Vulkan API. However, I wasn’t entirely sure - the same error message occasionally appeared even when the engine wasn’t hanging, so it wasn’t a definitive indicator.
To investigate further, I decided to enable the Vulkan validation layer, hoping it would reveal more detailed error messages about what the engine was doing wrong. Having Vulkan SDK installed in my system, I launched the Vulkan Configurator app that comes with it ("Bin\vkconfig.exe"), I selected Vulkan Layers Management = Layers Controlled by the Vulkan Configurator, and selected Validation.
Unfortunately, when I launched Godot again, no new error messages appeared in the console. (Looking back, I’m not even sure if that console window actually captured the process’s standard output.) For a brief moment, I thought enabling the Vulkan validation layer had fixed the hangs - but they soon returned. Maybe they were less frequent, or perhaps it was just wishful thinking.
Next, I considered forcing Godot to use the integrated GPU (Radeon Vega 8) instead of the more powerful discrete GPU (RX 560X). To test this, I adjusted Windows power settings to prioritize power saving over maximum performance. However, this didn’t work - Godot still reported using the Radeon RX 560X.
THE SOLUTION: What finally worked was forcing Godot to use the integrated GPU by launching it with a specific command-line parameter. Instead of running the editor normally, I used:
Godot_v4.3-stable_win64_console.exe --verbose --gpu-index 1
This made Godot use the second GPU (index 1) - the slower Radeon Vega 8 - instead of the default RX 560X. The result? No more hangs. While the integrated GPU is less powerful, it was more than enough for our 2D pixel art game.
I am not sure why it helped, considering that both GPUs on my laptop are from AMD and they are supported by one driver. I also didn't check whether the new Godot 4.4 that was released since then has this bug fixed. I am just leaving this story here, in case someone stumbles upon the same problem in the future.
# DirectX 12 Agility SDK 1.716.0-preview Explained
Sun
02
Feb 2025
On January 30th 2025 Microsoft released a new version of DirectX 12 Agility SDK: 1.615.0 (D3D12SDKVersion = 615) and 1.716.0-preview (D3D12SDKVersion = 716). The main article announcing this release is: AgilitySDK 1.716.0-preview and 1.615-retail. Files are available to download from DirectX 12 Agility SDK Downloads, as always, in form of .nupkg files (which are really ZIP archives).
I can see several interesting additions in the new SDK, so in this article I am going to describe them and delve into details of some of them. This way, I aim to consolidate information that is scattered across multiple Microsoft pages and provide links to all of them. The article is intended for advanced programmers who use DirectX 12 and are interested in the latest developments of the API and its surrounding ecosystem, including features that are currently in preview mode and will be included in future retail versions.
Comments | #rendering #directx Share
# Global Game Jam 2025 and First Impressions from Godot
Wed
29
Jan 2025
Last weekend, 24-26 January 2025 I participated in Global Game Jam, and, more specifically, PolyJam 2025 - a site in Warsaw, Poland. In this post I'll share the game we've made (including the full source code) and describe my first impressions of the Godot Engine, which we used for development.
We made a simple 2D pixel-art game with mechanics similar to Overcooked. It was designed for 2 to 4 players in co-op mode, using keyboard and gamepads.
Entry of the game at globalgamejam.org
GitHub repository with the source code
A side note: The theme of GGJ 2025 was "Bubble". Many teams created games about bubbles in water, while others interpreted it more creatively. For example, the game Startup Panic: The Grind Never Stops featured minigames like drawing graphs or typing buzzwords such as "Machine Learning" to convince investors to fund your startup – an obvious bubble 🙂 Our game, on the other hand, focused on taking care of babies and fulfilling their needs so they could grow up successfully. In Polish, the word for "bubbles" is "bąbelki", but it’s also informally used to refer to babies. Deliberately misspelled as "bombelki", it is a wordplay that makes sense and fits the theme in Polish.
My previous game jam was exactly two years ago. Before that jam, I had learned a bit of the Cocos Creator and used it to develop my game, mainly to try something new. I described my impressions in this post: Impressions After Global Game Jam 2023. This time, I took a similar approach and I started learning Godot engine about three weeks before the jam. Having some experience with Unity and Unreal Engine, my first impressions of Godot have been very positive. Despite being an open-source project, it doesn’t have that typical "open-source feeling" of being buggy, unfinished, or inconvenient to use. Quite the opposite! Here are the things I especially like about the engine:
I like that it’s small, lightweight, and easy to set up. All you need to do is download a 55 MB archive, unpack it, and you’re ready to start developing. This is because it’s a portable executable that doesn’t require any installation. The only time you need to download additional files (over 1 GB) is when you’re preparing to create a build for a specific platform.
I also appreciate how simple the core ideas of the engine are:
I’m not sure if this approach is optimal in terms of performance or whether it’s as well-optimized as the full Entity Component System (ECS) that some other engines use. However, I believe a good engine should be designed like this one – with a simple and intuitive interface, while handling performance optimizations seamlessly under the hood.
I also appreciate the idea that the editor is built using the same GUI controls available for game development. This approach provides access to a wide range of advanced controls: not just buttons and labels, but also movable splitters, multi-line text editors, tree views, and more. They can all be skinned with custom colors and textures.
Similarly, files saved by the engine are text files in the INI-like format with sections like [SectionName]
and key-value pairs like Name = Value
. Unlike binary files, XML, or JSON, these files are very convenient to merge when conflicts arise after two developers modify the same file. The same format is also available and recommended for use in games, such as for saving settings.
Then, there is GDScript - a custom scripting language. While Godot also offers a separate version that supports C# and even has a binding to Rust, GDScript is the native way of implementing game logic. I like it a lot. Some people compare it to Python, but it’s not a fork or extension of Python; it’s a completely separate language. The syntax shares similarities with Python, such as using indentation instead of braces {}
to define scopes. However, GDScript includes many features that Python lacks, specifically tailored for convenient and efficient game development.
One such feature is an interesting mix of dynamic and static typing. By default, variables can have dynamic types (referred to as "variant"), but there are ways to define a static type for a variable. In such cases, assigning a value of a different type results in an error – a feature that Python lacks.
var a = 0.1
a = "Text" # OK - dynamic type.
var b: float
b = 0.1
b = "Text" # Error! b must be a number.
var c := 0.1
c = "Text" # Error! c must be a number.
Another great feature is the inclusion of vector types for 2D, 3D, or 4D vectors of floats or integers. These types are both convenient and intuitive to use – they are passed by value (creating an implicit copy) and are mutable, meaning you can modify individual xyzw
components. This is something that Python cannot easily replicate: in Python, tuples are immutable, while lists and custom classes are passed by reference. As a result, assigning or passing them as function parameters in Python makes the new variable refer to the original object. In GDScript, on the other hand:
var a := Vector2(1.0, 2.0)
var b := a # Made a copy.
b.x = 3.0 # Can modify a single component.
print(a) # Prints (1, 2).
I really appreciate the extra language features that are clearly designed for game development. For example, the @export
attribute before a variable exposes it to the Inspector as a property of a specific type, making it available for visual editing. The $NodeName
syntax allows you to reference other nodes in the scene, and it supports file system-like paths, such as using /
to navigate down the hierarchy and ..
to go up. For instance, you can write something like $../AudioPlayers/HitAudioPlayer.play()
.
I also like how easy it is to animate any property of any object using paths like the one shown above. This can be done using a dedicated AnimationPlayer
node, which provides a full sequencer experience with a timeline. Alternatively, you can dynamically change properties over time using a temporary Tween
object. For example, the following code changes the font color of a label to a transparent color over 0.5 seconds, using a specific easing function selected from the many available options (check out the Godot tweening cheat sheet for more details):
create_tween().tween_property(addition_label.label_settings, ^":font_color", transparent_color, 0.5).set_trans(Tween.TRANS_CUBIC).set_ease(Tween.EASE_IN)
I really appreciate the documentation. All core language features, as well as the classes and functions available in the standard library, seem to be well-documented. The documentation is not only available online but also integrated into the editor (just press F1), allowing you to open documentation tabs alongside your script code tabs.
I also like the debugger. Being able to debug the code I write is incredibly important to me, and Godot delivers a full debugging experience. It allows you to pause the game (automatically pausing when an error occurs), inspect the call stack, view variable values, explore the current scene tree, and more.
That said, I’m sure Godot isn’t perfect. For me, it was just a one-month adventure, so I’ve only described my first impressions. There must be reasons why AAA games aren’t commonly made in this engine. It likely has some rough edges and missing features. I only worked with 2D graphics, but I can see it supports 3D graphics with a Forward+ renderer and PBR materials. While it could potentially be used for 3D projects, I’m certain it’s not as powerful as Unreal Engine in that regard. I also encountered some serious technical issues with the engine during the game jam, but I’ll describe those in separate blog posts to make them easier to find for anyone searching the Internet for a solution.
I also don’t know much about Godot’s performance. The game we made was very simple. If we had thousands of objects on the scene to render and complex logic to calculate every frame, performance would become a critical factor. Doing some work in every object every frame using _process
function is surely an anti-pattern and it runs serially on a single thread. However, I can see that GDScript also supports multithreading – another feature that sets it apart from Python.
To summarize, I now believe that Godot is a great engine at least for game jams and fast prototyping.
Comments | #godot #events #competitions #ggj #productions Share
# Thoughts Beyond C++
Mon
30
Dec 2024
Earlier this month, Timothy Lottes published a document on Google Docs called “Fixing the GPU”, where he describes many ideas about how programming compute shaders on the GPU could be improved. It might be an interesting read for those advanced enough to understand it. The document is open for suggestions and comments, and there are few comments there already.
On a different topic, 25 November I attended Code::Dive conference in Wrocław, Poland. It was mostly dedicated to programming in C++ language. I usually attend conferences about game development, so it was an interesting experience for me. Big thanks to Tomasz Łopuszański from Programista magazine for inviting me there! It was great to see Bjarne Stroustrup and Herb Sutter live, among other good speakers. By the way, recordings from the talks are available on YouTube.
Those two events inspired me to write down my thoughts – my personal “wishlist” about programming languages, from the perspective of someone interesting in games and real-time graphics programming. I gathered my opinions about things I like and dislike in C++ and some ideas about how a new, better language could look like. It is less about a specific syntax to propose and more about high-level ideas. You can find it under the following shortened address, but it is really a document at Google Docs. Comments are welcome.
Of course I am aware of Rust, D, Circle, Carbon, and other programming languages that share the same goal of replacing C++. I just wanted to write down my own thoughts about this topic.
Comments | #c++ #compilers #productions Share
# FP8 data type - all values in a table
Tue
24
Sep 2024
Floating-point numbers are a great invention. Thanks to dedicating separate bits to the sign, exponent, and mantissa (also called significand), they can represent a wide range of numbers on a limited number of bits - numbers that are positive or negative, very large or very small (close to zero), integer or fractional.
In programming, we typically use double-precision (64b) or single-precision (32b) numbers. These are the data types available in programming languages (like double
and float
in C/C++) and supported by processors, which can perform calculations on them efficiently. Those of you who deal with graphics programming using graphics APIs like OpenGL, DirectX, or Vulkan, may know that some GPUs also support 16-bit floating-point type, also known as half-float.
Such 16b "half" type obviously has limited precision and range compared to the "single" or "double" version. Because of these limitations, I am reserved in recommending them to use in graphics. I summarized capabilities and limits of these 3 types in a table in my old "Floating-Point Formats Cheatsheet".
Now, as artificial intelligence (AI) / machine learning (ML) is a popular topic, programmers use low precision numbers in this domain. When I learned that floating-point formats based only on 8 bits were proposed, I immediately thought: 256 possible value is little enough that they could be all visualized in a 16x16 table! I developed a script that generates such tables, and so I invite you to take a look at my new article:
"FP8 data type - all values in a table"
Comments | #math #artificial intelligence Share
# How to do a good code review - a new article
Mon
19
Aug 2024
Today I would like to present my new, comprehensive article: "How to do a good code review". I can be helpful to any programmer no matter what programming language they use. Conducting good code reviews is a skill worth mastering. In this article, we will discuss the advantages and disadvantages of this process and explore the types of projects where it is most beneficial. We will consider the best approach to take when reviewing code, how to effectively carry out the process, which aspects of the code to focus on, and finally – how to write comments on the code in a way that benefits the project. The goal is to ensure that the review process serves as an opportunity for fruitful communication among team members rather than a source of conflict.
The article was first published few months ago in Polish in issue 112 (March/April 2024) of the Programista magazine. Now I have a right to show it publicly for free, so I share it in two language versions:
I wasn't active on my blog in the past months because I took some time for vacation, but also because I'm now learning about machine learning. I may be late to the party, but I recognize that machine learning algorithms are useful tools in many applications. As people learn the basics, the often feel an urge to teach others about it. Some good reads authored by game/graphics developers are: "Machine Learning for Game Devs" by Alan Wolfe and "Crash Course in Deep Learning (for Computer Graphics)" by Jakub Boksansky. I don't want to duplicate their work, so I will only blog about it when I have something unique to show.
Comments | #productions #software engineering Share
# Shapes and forms of DX12 root signatures
Tue
14
May 2024
This article is for you if you are a programmer using Direct3D 12. We will talk about a specific part of the API: root signatures. I will provide a comprehensive description of various formats in which they can be specified, stored, and ways to convert between them. The difficulty of this article is intermediate. You are expected to know at least some basics of D3D12. I think that advanced developers will can also learn something new, as some of the topics shown here are not what we typically use in a day-to-day development with D3D12.
I will use C++ as the programming language. Wherever possible, I will also try to use standalone command-line tools instead of writing a custom code. To repeat my experiments demonstrated in this article, you will need two of these:
PATH
environmental variable, so you can open Command Prompt and just type "dxc" to use it.You don't need to know the command-line syntax of these tools to understand the article. I will describe everything step-by-step.
Warning about DXC: If you also have Vulkan SDK installed, very likely your PATH
environmental variable points to "dxc.exe" in that SDK instead of Windows SDK, which can cause problems. To check this, type command: where dxc
. If you find Vulkan SDK listed first, make sure you call "dxc.exe" from Windows SDK, e.g. by explicitly specifying full path to the executable file.
Warning about RGA: If you want to repeat command-line experiments presented here, make sure to use Radeon GPU Analyzer in the latest version, at least 2.9.1. In older versions, the commands I present wouldn't work.
A side note about shader compilation: Native CPU code, like the one we create when compiling our C++ programs, is saved in .exe files. I contains instructions in a common format called x86, which is sent directly to CPU for execution. It works regardless if you have an AMD or Intel processor in your computer, because they comply to the same standard. With programs written for the GPU (which we call shaders), things are different. Every GPU vendor (AMD, Nvidia, Intel) has its own instruction set, necessitating a two-step process for shader compilation:
In Direct3D 12, a root signature is a data structure that describes resource bindings used by a pipeline on all the shader stages. Let's see an example. Let's work with file "Shader1.hlsl": a very simple HLSL code that contains 2 entry points: function VsMain
for vertex shader and function PsMain
for pixel shader:
struct VsInput
{
float3 pos : POSITION;
float2 tex_coord : TEXCOORD;
};
struct VsOutput
{
float4 pos : SV_Position;
float2 tex_coord : TEXCOORD;
};
struct VsConstants
{
float4x4 model_view_proj;
};
ConstantBuffer<VsConstants> vs_constant_buffer : register(b4);
VsOutput VsMain(VsInput i)
{
VsOutput o;
o.pos = mul(float4(i.pos, 1.0), vs_constant_buffer.model_view_proj);
o.tex_coord = i.tex_coord;
return o;
}
Texture2D<float4> color_texture : register(t0);
SamplerState color_sampler : register(s0);
float4 PsMain(VsOutput i) : SV_Target
{
return color_texture.Sample(color_sampler, i.tex_coord);
}
I assume you already know that a shader is a program executed on a GPU that processes a single vertex or pixel with clearly defined inputs and outputs. To perform the work, it can also reach out to video memory to access additional resources, like buffers and textures. In the code shown above:
A root signature is a data structure that describes what I said above - what resources should be bound to the pipeline at individual shader stages. In this specific example, it will be a constant buffer at register b4, a texture at t0, and a sampler at s0. It can also be shown in form of a table:
Root param index | Register | Shader stage |
---|---|---|
0 | b4 | VS |
1 | t0 | PS |
2 | s0 | PS |
I am simplifying things here, because this article is not about teaching you the basics of root signatures. For more information about them, you can check:
To prepare for our experiments, let's compile the shaders shown above using commands:
dxc -T vs_6_0 -E VsMain -Fo Shader1.vs.bin Shader1.hlsl
dxc -T ps_6_0 -E PsMain -Fo Shader1.ps.bin Shader1.hlsl
Note that a single HLSL source file can contain multiple functions (VsMain
, PsMain
). When we compile it, we need to specify one function as an entry point. For example, the first command compiles "Shader1.hlsl" file using VsMain
function as the entry point (-E
parameter) treated as a vertex shader in Shader Model 6.0 (-T
parameter). Similarly, the second command compiles PsMain
function as a pixel shader. Compiled shaders are saved in two separate files: "Shader1.vs.bin" and "Shader1.ps.bin".