We use an in-house OpenGL app to provide the out-the-window visuals for our flight simulators. The app is cross platform, but until now the Mac version was only used by desktop researchers, not in our primary sim labs. Now we are attempting to replace some Windows boxes with Apple Studios. We can easily maintain high framerate, and visual quality is excellent, but we are finding the graphics have a bit of stutter during high yaw rates (which quickly forces new assets into view). I've eliminating unnecessary processes, tried raising my priority via pthread_set_qos_class_self_np() or thread_policy_set(), and reducing textures quality, all of which helped, but it didn't eliminate the problem. For background, we are using framebuffers, we have a very large texture database (90 GB), and the render code runs in the main thread (not a secondary thread). What might I be missing?

I am in the process of developing a matrix-vector multiplication kernel. While conducting performance evaluations, I've noticed that on M1/M1 Pro/M1 Max, the kernel demonstrates an impressive memory bandwidth utilization of around 90%. However, when executed on the M1 Ultra/M2 Ultra, this figure drops to approximately 65%. My suspicion is that this discrepancy is attributed to the dual-die architecture of the M1 Ultra/M2 Ultra. It's plausible that the necessary data might be stored within the L2 cache of the alternate die. Could you kindly provide any insights or recommendations for mitigating the occurrence of on-die L2 cache misses on the Ultra chips? Additionally, I would greatly appreciate any general advice aimed at enhancing memory load speeds on these particular chips.

I've been trying to get the bash/script version of DeepFaceLab to work with Apple Silicon Macs, but this was original a Windows project that even now has non-existent support for MacOs/Apple Silicon. I am thinking of converting everything into a native macOS app using Swift, specifically optimized for Apple Silicon GPUs. Here's what I got from ChatGPT. Any help/advice on how to do this would be greatly appreciated. I don't have any Swift programming experience, but I have experience with some coding and can generally figure things out. I know that this is probably not feasible for a single individual with little programming experience, but I wanted to throw this out there to see what others think. Thank you Here's a high-level overview of the steps involved in porting DeepFaceLab to Swift with a graphical UI: Understand DeepFaceLab: Thoroughly study the DeepFaceLab project, its Python scripts, and the overall architecture to grasp its functionalities and dependencies. Choose a Swift Framework: Decide on the UI framework you want to use for the macOS app. SwiftUI is Apple's latest UI framework that works across all Apple platforms, including macOS. Alternatively, you can use AppKit for a more traditional approach. Rewrite Python to Swift: Convert the Python code from DeepFaceLab into Swift. You'll need to rewrite all the image processing, deep learning, and video manipulation code in Swift, potentially using third-party Swift libraries or native macOS frameworks. Deep Learning Integration: Replace the Python-based deep learning library used in DeepFaceLab with an appropriate Swift-compatible deep learning framework. TensorFlow and PyTorch both offer Swift APIs, but you may need to adapt the specific model implementation to Swift. Image Processing: Find equivalent Swift libraries or frameworks for image processing tasks used in DeepFaceLab. UI Development: Design and implement the graphical user interface using SwiftUI or AppKit. You'll need to create views, controls, and navigation elements to interact with the underlying Swift code. Integration: Connect the Swift code with the UI components, ensuring that actions in the GUI trigger the appropriate Swift functions and display results back to the user. Testing and Debugging: Rigorously test the Swift application and debug any issues that arise during the porting process. Optimization: Ensure that the Swift app performs efficiently and effectively on macOS devices.