Changelog:

# General 
+ Prepacked CUDA and CUDNN for user friendliness, it is now download and play.
+ Refactor namespaces, enhance initialization, and UI
+ Improved logging and error handling across Aimmy for easy error pinpointing.
+ Refactor RequirementsManager, now checking for CUDA and TensorRT requirements accordingly, and more. 
+ Organize UI elements
+ Refactored namespaces and using directives for better organization and less confusion.
+ Updated various classes for better readability and maintainability.
+ General Bug Fixes, such as proper disposal on resolution change.
+ General Performance Enhancement
+ Updated ONNXRuntime, requirements for CUDA have changed, 12.6 CUDA and 9.3 CUDNN
+ Added new UI elements and event handlers
 -> Drag and drop Models and Configs for simple loading.
 -> Add a model progression, just a spinning square to let you know its loading (Useful for TensorRT)

- You cannot spam load models anymore, as it doesnt allow you to select a model as one is loading (This is because TensorRT takes a long time to load/cache)
- Removal of ddxoft mouse movement
- Removal of bloat
- Configs are having issues, will be fixed soon.

# Screen Capture
+ Improved Direct3D initialization and error handling.
+ Rewrote Direct3D11 Screen Capturing
+ Added GDI fallback for screen capture.
+ Seamless screen capture swapping (besides the 1s delay for direct3d11 disposing)


# AI Loop
+ Enhanced model initialization with CUDA and TensorRT support.
+ Updated prediction methods for efficiency.

This is one of the biggest updates in Aimmy CUDA, at least on the backend. You will notice significant performance changes and a lot less bugs. It must suck to be an AMD user.

This update brings TensorRT & CUDA combined. If you're having issues with DirectX, you can simply switch back to GDI (but we have a fix for all you laptop users so GDI is unnecessary, but there).

Not only is this update larger, now anyone ranging from tech nerd to a 7 year old, you can now install Aimmy CUDA like as if it was regular aimmy. Just download the zip and run the exe.

AI Confidence was moved to Aim Config, btw.

What is TensorRT?

@mastere4 says, "Pretty much an add-on for CUDA. While it does make your gameplay smoother and faster, it's a double edge sword by making your models loading time drastically slower for 1st time instances."

Comparison between TensorRT and CUDA

+ TensorRT has a 5-10ms difference for me.
+ TensorRT optimizes a model and uses precision methods for accuracy.
+ TensorRT has a GPU Memory Limit
+ TensorRT has INT8 Precision and FP16 Precision.
- TensorRT Caching takes a lot of space if you are optimizing a lot of models.
- TensorRT takes way longer to load a model (for the 1st load), 384 - 60 seconds, but after caching, it'll take about 40-9 seconds.
- ONNXRuntime's caching system is difficult, being way slower than the DLL i have created, but no way to use my DLL in conjuction with ONNXRuntime.
- About 560MB, and extra installation (though quite simple)

+ CUDA model loadtimes are almost instantaneous
+ CUDA also has a GPU Memory Limit
+ CUDA can use TF32 as a Math Mode.
+ CUDA is 560MB less than the TensorRT installation
- CUDA does not have the optimizations that TensorRT provides
- CUDA is approximately 10-5ms slower than TensorRT

Which RAR File do I download?!

Pick DLLS-INCLUDED.2.2.0-AC.rar if you don't know what you're doing and don't have 12.x cuda installed or 9.x cudnn.

Pick 2.2.0.Aimmy-CUDA.rar if you do know what you're doing and do have 12.x cuda and 9.x cudnn installed.

IF you don't have the 12.x and 9.x installed but you want to, then read the README file on how to install those two.

No, you cannot use this on your AMD or Intel GPU.

Description

Enhance screen capture with Direct3D and refactor code... This is a preview of whats to come to regular Aimmy so... yay

beware. I made these updates at 4 in the morning, if there are bugs just let me know and be patient...

Bug Fixes (8/16/24)

• stops detecting out of no where using closest to mouse (fixed)
• collect data auto-collecting when off
• collect data not collecting when its not detecting anything
• open directory doesnt open the right folder (maybe fixed)
• rewrote screen capture method for performance and bug updates
• added a debug mode for iterations and error logs.
• added more razer checks
• fixed resolution changes (for ex alt tabbing to fullscreen/windowed breaks aimmy)
• add a clamping thing to stop aimmy from crashing when your cursor goes out of bounds (to a different monitor)

Github Copilot log:

Transitioned screen capture from GDI+ to Direct3D for improved efficiency

• Integrated Direct3D-based screen capture (InitializeDirect3D, D3D11Screen, ScreenGrab).
• Added InitializeCaptureMethod to select capture method based on user settings.

Improved performance and debugging:

• Updated AiLoop to include FPS calculation and display.

Refactored code for better readability and maintainability:

• Split UpdateOverlay into smaller methods (UpdateConfidence, UpdateTracers, UpdateFocusBox).
• Optimized HandlePredictions and GetClosestPrediction.
• Enhanced PrepareKDTreeData with pre-allocated lists.

Enhanced resource and thread management:

• Introduced DisposeResources for proper resource disposal.
• Ensured proper thread termination and cleanup.

Minor cleanups and UI updates:

• Removed deprecated GDI+ method and commented out unused CaptureMethod.
• Added new settings and UI elements for screen capture method, debug mode, and FPS display.
• Commented out and removed references to ddxoft Virtual Input Driver.
• Adjusted dropdown items and added new sliders and toggles in MainWindow.xaml.cs.
• Added FPS label in DetectedPlayerWindow.xaml.

Updated project dependencies:

• Added Vortice.Direct3D11 package reference in Aimmy2.csproj.

Full Changelog: 2.1.6...2.17

CUDA!

Use your NVIDIA gpu to its fullest with a simple line change, this has a significant performance improvement compared to DirectML, in our cases anyway. For many users, DirectML is simpler and easier to setup; plug and play basically, but CUDA has significantly improved performance... 50/50.

Features (Basically beta...)

• Percentage System
• EMA Smoothing (thanks @whoswhip)
• Fixed RZCTL and Razer Process detections (maybe)
• n more cleanup/optimizations lol

README

Check README file in the github to see how to set this up properly... hopefully it works for you, goodluck