Xavier Verguin Gonzalez

Optimize your app's GPU performance with Xcode 26.4. This update introduces new Neural Accelerator and Occupancy profiling counters for M5 and A19 GPUs, plus per-line shader live register counts. Discover these new tools in: developer.apple.com/videos/play/te… and developer.apple.com/videos/play/te…

The second tech talk: Boost your graphics performance with the M5 and A19 GPUs developer.apple.com/videos/play/te… youtube.com/watch?v=_5yEcJ…

We released two tech talks today going over how to take advantage of the new architecture, features and associated developer tools. 1) Accelerate your machine learning workloads with the M5 and A19 GPUs developer.apple.com/videos/play/te… youtube.com/watch?v=wgJX1H…

We brought significant architectural advancements and feature set improvements to A19/M5 GPUs. Scalable GPU Neural Accelerators 2nd Gen Dynamic Caching Shader Architecture 3rd Gen Ray Tracing The best GPU Driven Pipeline New graphics features, rate and perf increases.

My "No Graphics API" blog post is live! Please repost :) sebastianaaltonen.com/blog/no-graphi…

@ErikWDev The Heat Map view shows you the execution history for the selected SIMD, which could vary greatly and not be indicative of overall performance. I recommend using the cost graph in the Shaders tab to get a better understanding of cost breakdown: developer.apple.com/documentation/…

It is a single indirect draw call of quads, and to my limited understanding it appears in #Xcode as if most of the time is spent inside a single dot product? Or being limited to do so. How can I get to the bottom of this? #gamedev #indiedev

Can someone help me get to the bottom of this render pass that leaves me a bit baffled? We have one draw call for our particles that has a lot of overdraw, but it is taking an insane 25ms on my M4. I am happy to provide more info #Metal #Graphics #GPU #performance #indiedev

@bztree Hi Sasha. I am sorry you're hitting these issues profiling your kernel. Can you file a bug report via feedbackassistant.apple.com with a repro case / more information? Happy to help via DM too. Thank you.

I guess the profiler is somehow replaying my kernel on the GPU, and if it does it wrong it hangs the window server, which then causes a kernel panic because the window server doesn't check in?! I guess that's my theory for now.. What a Rube Goldberg machine

Oh dear I’ve done it again. Obviously Apple didn’t expect their products to be used by a REAL DEVELOPER like me. The fact that is crashing so much must mean I’m pushing limits here (actually I’m 2x slower than MLX right now :( I just want this profiler to work)

Learn to unlock the full potential of game development for Apple Platforms with an all-day online event tomorrow, November 6. You'll learn how to leverage Metal, optimize for Apple silicon, design great handheld experiences, and navigate the App Store. youtube.com/watch?v=H4Dm97…

Documentation for the new Metal Performance HUD in macOS/iOS/iPadOS/tvOS 26 is now available – covering in great detail HUD customization, available metrics, performance insights, and performance reports: developer.apple.com/documentation/… (5 articles in total)

@warrenm @Luckyballa Hi Warren. Could you file a report for the capture issues via feedbackassistant.apple.com and share the id? We will have a look asap. Thanks!

@Luckyballa I’ve experienced crashes or unusable results on 80% of my captures so far with Metal 4, but that’s mostly because I’ve been abusing the daylights out of the API. Once I get acclimated I’m sure it’ll be more stable

Metal Debugger is the sole reason why Metal is the best framework for learning GPU programming

In Xcode 26, the Dependency Viewer in Metal Debugger has been updated to visualize fine-grain per-stage dependencies.

The Metal Performance HUD has been updated with new metrics and a customizable user interface. It also now has the ability to identify performance insights and generate performance reports. Learn more at developer.apple.com/videos/play/ww…

Metal Debugger in Xcode 26 comes with brand new debugging support for Machine Learning in Metal 4. Use the new Machine Learning network debugger to inspect and debug your ML network, including inspecting intermediate tensor data. Learn more at developer.apple.com/videos/play/ww…

Metal 4 comes with a streamlined API for optimal performance. Compilation is explicit and faster. And comes with first-class support for ML. Execute full models in your command buffers or inline inference operations alongside your shader code. developer.apple.com/metal/whats-ne…

Are you interested in joining the team that builds Apple's GPU Developer Tools? We're looking for a dedicated engineer to work on our core GPU tools technologies stack. Reach out, or apply here: jobs.apple.com/en-us/details/…

📣Instruments in Xcode 16.3 Beta 2 contains a brand new tool for analyzing CPU workloads: Processor Trace. It uses hardware execution tracing to accurately reconstruct program execution and is available on M4 and all iPhone 16 devices. Learn more on developer.apple.com/documentation/…

@Alecazam123 @s1ddok What's the issue that you're seeing with MoltenVK and Metal Capture?

@s1ddok One good reason not to use MoltenVK. Having Metal capture is critical.

Metal + Swift + Metal GPU Debugger is such a good productivity combo that I now routinely implement stuff first in it and only once debugged, port to C++/Vulkan world. It is way faster for me to implement algorithm twice then debug it on Vulkan.

Measuring the performance of the Unity 6 URP deferred renderer with frame buffer fetch.

@timd_ca If you don't pass -fmetal-enable-logging to the Metal compiler then you won't pay any runtime cost, even if you keep the log calls in your shader. This is assuming that you don't define the runtime EVs that enable logging (MTL_LOG_LEVEL/MTL_LOG_BUFFER_SIZE).

@Xavyiy This has been a very welcome feature. Is there any runtime cost when logging is disabled? (does everything nop away with dead code stripping, etc?)

Did you know that in macOS 15 and iOS 18 you can log debug messages directly from your Metal shaders? Learn more about it at developer.apple.com/documentation/…