Skip to content

2.2.0

Latest
Compare
Choose a tag to compare
@TaylorIsBlue TaylorIsBlue released this 05 Oct 01:50

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.