Metal GPU Driver Crash on M5 Pro + macOS 26.5 — kIOGPUCommandBufferCallbackErrorOutOfMemory with <2GB working sets Summary The Metal driver AGXMetalG17X 351.2 on macOS 26.5 (25F71) for the M5 Pro chip crashes with kIOGPUCommandBufferCallbackErrorOutOfMemory (00000008) when running LLM inference workloads with working sets as small as ~1.5GB, despite 24GB of unified memory being available and Apple Diagnostics confirming the hardware is fully functional. This affects multiple tools: MLX, llama.cpp (Metal backend), and native apps using Metal for inference. System Component Value Model MacBook Pro (Mac17,9) Chip Apple M5 Pro (applegpu_g17s) GPU Cores 16 RAM 24 GB LPDDR5 macOS 26.5 (25F71) Metal Metal 4 GPU Driver AGXMetalG17X 351.2 Xcode 26.5 (17F42) Reproduction MLX (Python) pip install mlx mlx-lm python -m mlx_lm.generate \ --model mlx-community/Qwen2.5-3B-Instruct-4bit \ --max-tokens 10 \ --prompt "Hello" Expected: Normal text generation Actual: Crash with: libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Insufficient Memory (00000008:kIOGPUCommandBufferCallbackErrorOutOfMemory) llama.cpp brew install llama.cpp llama-cli --model model.gguf --prompt "Hello" --n-predict 20 --n-gpu-layers 99 Expected: Fast GPU generation Actual: Process hangs indefinitely Test Results Tool Model Peak Memory Result MLX Qwen2.5-0.5B-4bit 0.36 GB ✅ Works MLX Qwen2.5-1.5B-4bit 0.98 GB ✅ Works MLX Qwen3-1.7B-4bit 1.01 GB ✅ Works MLX Qwen2.5-3B-4bit ~1.5 GB ❌ Metal OOM crash MLX Qwen3-4B-4bit ~2.1 GB ❌ Metal OOM crash MLX Qwen3-8B-4bit ~4.5 GB ❌ Metal OOM crash llama.cpp Qwen2.5-0.5B GGUF ~0.5 GB ❌ Hangs with GPU llama.cpp Qwen2.5-0.5B GGUF ~0.5 GB ✅ Works with CPU only Key Evidence Hardware is healthy — Apple Diagnostics passed all tests Basic Metal works — matmul, array ops work fine CPU inference works — llama.cpp with -ngl 0 runs correctly The error is NOT about actual memory exhaustion — kIOGPUCommandBufferCallbackErrorOutOfMemory means the kernel rejects the Metal memory commit, not that physical memory is full. The system reports 17.76GB available for Metal working set. Crash Log Extract Thread 31 Crashed: 0 libsystem_kernel.dylib __pthread_kill + 8 1 libsystem_pthread.dylib pthread_kill + 296 2 libsystem_c.dylib abort + 148 3 Metal MTLReportFailure.cold.1 + 48 4 Metal MTLReportFailure + 576 5 Metal -[_MTLCommandBuffer addCompletedHandler:] + 104 ... Exception Type: EXC_CRASH (SIGABRT) Termination Reason: Namespace SIGNAL, Code 6, Abort trap: 6 Related Issues ml-explore/mlx#3586 — Metal compiler regression on macOS 26.5 ml-explore/mlx#3534 — M5 float32 precision issue ml-explore/mlx#3568 — M5 random divergence ml-explore/mlx#3539 — Metal residency OOM (M4 Max) Request Please investigate the AGXMetalG17X driver for M5 Pro on macOS 26.5. The driver appears to incorrectly reject Metal memory commits for LLM inference workloads, even when the working set is well within the system's reported limits (1.5GB requested vs 17.76GB available). Happy to provide full crash logs, sysdiagnose archives, or run additional tests.

Hi All, We have released our application in both iOs and Mac Os platform. But Now we have changed our plan to target Mac Os only. So, We need to disable the availability of iOs version of our app. As I have checked in 'Pricing And Availability' Page, There is one option to remove from sale. But What we need to do to disable iOs app availability alone(Not Mac app)? Please give your suggestion to do this. Thanks.

I have an orphaned asset folder taking up 9.13GB located at: /System/Library/AssetsV2/com_apple_MobileAsset_iOSSimulatorRuntime/c0d3fd05106683ba0b3680d4d1afec65f098d700.asset It contains SimulatorRuntimeAsset version 18.5 (Build 22F77). Active Version: My current Xcode setup is using version 26.2 (Build 23C54). I checked the plist files in the directory and found what seems to be the cause of the issue: The "Never Collected" Flag: The Info.plist inside the orphaned asset folder explicitly sets the garbage collection behavior to "NeverCollected": <key>__AssetDefaultGarbageCollectionBehavior</key> <string>NeverCollected</string> The Catalog Mismatch: The master catalog file (com_apple_MobileAsset_iOSSimulatorRuntime.xml) in the parent directory only lists the new version (26.2). Because the old version (18.5) is missing from this XML, Xcode and mobileassetd seem to have lost track of it entirely. What I Have Tried (All Failed) Xcode Components: The version 18.5 does not appear in Settings -> Components, so I cannot delete it via the GUI. Simctl: xcrun simctl list runtimes does not list this version. Running xcrun simctl runtime delete 22F77 fails with: "No runtime disk images or bundles found matching '22F77'." Manual Deletion: sudo rm -rf [path] fails with "Operation not permitted", presumably because /System/Library/AssetsV2 is SIP-protected. Third-party Tools: Apps like DevCleaner do not detect this runtime (likely because they only scan ~/Library or /Library, not /System/Library). Has anyone found a way to force the system (perhaps via mobileassetd or a specific xcrun flag) to re-evaluate this folder and respect a deletion request? I am trying to avoid booting into Recovery Mode just to delete a cache file. Any insights on how AssetsV2 handles these "orphaned" files would be appreciated.

I am seeing a reproducible issue with macOS ControlCenter's per-app menu bar tracking state in a SwiftUI MenuBarExtra app. On launch the app creates its menu bar status item. ControlCenter then reads group.com.apple.controlcenter / trackedApplications, moves the app's status item host to the blocked list, sends NSStatusItemChangeVisibilityAction with visibility=0, and the app terminates because its only status item is removed. Two observations: Changing only the bundle identifier makes the menu bar item appear normally. In trackedApplications, the affected app has its own entry with isAllowed: true; a separate entry for another app has isAllowed: false, and that disallowed foreign entry's menuItemLocations contained an entry referencing the affected app. Removing only the affected app's reference from the other app's disallowed entry fixed the issue: the app launched normally, and ControlCenter no longer sent visibility=0. ControlCenter appears to associate one app's menu item identity with another app's disallowed tracked record, letting the foreign blocked record override the app's own allowed record. Is this expected? And is there a supported way to reset this per-app ControlCenter menu bar state without editing the protected group.com.apple.controlcenter plist directly? Testing Environment: macOS 26.3 (25D125) Xcode 26.3 (17C529) LSUIElement: true Sandboxed app SwiftUI MenuBarExtra Relevant log pattern: Host properties initialized; (bid:[AffectedApp]-Item-0-[pid]) State(applicationItem: true, clientRequestsVisibility: true, neverClip: false) looked up value <private> for key trackedApplications (Domain: group.com.apple.controlcenter) Moving host to blocked list; (bid:[AffectedApp]-Item-0-[pid]) Requesting blocked host to not be visible; (bid:[AffectedApp]-Item-0-[pid]) Sending action(s): <NSStatusItemChangeVisibilityAction: ...> Received action(s): NSStatusItemChangeVisibilityAction 0 agent requesting visibility=0 temporary=0 0 terminating on removal

I have the issue with simply trying to ssh to any other host on the LAN since Tahoe 26.5. last entries from ssh -v answer are debug1: Host 'answer' is known and matches the ED25519 host key. debug1: Found key in /Users/dgerman/.ssh/known_hosts:5 debug1: rekey out after 134217728 blocks debug1: SSH2_MSG_NEWKEYS sent debug1: expecting SSH2_MSG_NEWKEYS debug1: SSH2_MSG_NEWKEYS received debug1: rekey in after 134217728 blocks debug1: SSH2_MSG_EXT_INFO received debug1: kex_ext_info_client_parse: server-sig-algs= debug1: kex_ext_info_check_ver: publickey-hostbound (AT) openssh.com=<0> debug1: SSH2_MSG_SERVICE_ACCEPT received debug1: Authentications that can continue: publickey,password debug1: Next authentication method: publickey No problem from older MacBooks This worked before the update.

We have been having unexplained failures with the codesign tool recently on macosx aarch64 and x64 hosts. Every once in a while when signing an app locally using the following command: /usr/bin/codesign -s - -vvvv --force /home/me/FooBarCalculator.app results in the following error: /home/me/FooBarCalculator.app: timestamps differ by 185 seconds - check your system clock The number of seconds reported in the error message keeps varying (but usually in that range). We have checked the system clock but there isn't anything wrong (from what we can see) with the host. In fact, we have been seeing this error on several hosts now, so it isn't specific to one host. While looking into this issue, we even printed the details of an already signed binary using the following command: codesign -dvvv HelloWorld.app and that prints among other things, similar warning message: ... Timestamp=12 May 2026 at 5:36:0 AM HelloWorld.app: timestamp mismatch: internal time 12 May 2026 at 5:32:59 AM (184 seconds apart) I'm looking for inputs on how we go about debugging this issue and where/how the codesign tool sources these timestamps from (any specific API?) and what value is it comparing against to notice a difference. These affected hosts have different operating system versions some 15.x and some 26.x.

I have a chat app with a floating composer pill (NSGlassEffectView) pinned to the bottom of my content area. Messages scroll behind it. The pill looks fine but content behind it is completely clear. Fully legible text showing through, no frosting at all. Mail, Messages, Notes all have this frosted look where content behind a floating bar is blurred enough that you can't read it. That's what I'm going for. My scroll view has wantsLayer = true. I tried hasVerticalScroller = true + scrollerStyle = .overlay based on a tip I saw in another thread. No change. Is NSGlassEffectView supposed to blur what's behind it, or is the frosting in Apple's apps coming from the scroll edge effect? If it's the scroll edge effect, is there any way to trigger it for a custom floating view at the bottom of a scroll view? I can't find any AppKit API for this. I also spent time trying this in SwiftUI using .glassEffect() on macOS. Same result, the material renders nearly clear with no way to control the frosting level via any public API. If I'm missing something obvious I'd love to know. I've been stuck on this one for a while. Thanks for any advice! I've attached photos of my input bar (clearly transparent) and Apple Messages, which has a style that I see all over iOS and Mac OS 26.

Device Details: MBP M2 Pro AND MBP M3 Pro macOS 26.3 (25D125) Xcode Version 26.3, 26.2, 26.1 (I reinstalled all 3 of these after the macOS update) BUG: Xcode hangs indefinitely when performing Source Control operations (Pull or Push) on a Git repository that uses SSH authentication. The same repository works correctly when performing the equivalent Git operations from the Terminal using the Git CLI. The issue appears to be specific to Xcode’s internal Source Control integration. When the operation is triggered from Xcode, the UI shows a spinning progress indicator and never completes. Terminal Git commands (git fetch, git pull, git push) complete normally using the same SSH key and repository. A hang sample taken during the issue shows the Xcode main thread blocked in Source Control authentication and fingerprint handling code paths, including: IDESourceControlUIHandler IDESourceControlFingerprintManager handleAuthenticationFailure showFingerprintAlertOnWindow This suggests Xcode may be waiting on a Source Control authentication or host fingerprint UI flow that never resolves. SSH connectivity itself is functioning correctly: ssh -T git@bitbucket.org and ssh -T git@github.com both authenticate successfully. git ls-remote, git fetch, git pull, and git push all work correctly from Terminal. No Source Control accounts are configured in Xcode Preferences. Authentication relies entirely on SSH keys. Steps to Reproduce Configure an SSH key for Git access (e.g., Bitbucket or GitHub) and confirm it works via Terminal. Clone or open an existing Git repository that uses SSH (git@host:repo.git). Open the project/workspace in Xcode. In Xcode, attempt a Source Control operation such as: Source Control → Pull Source Control → Push Observe that Xcode displays a spinning progress indicator and does not complete the operation. Logs available on this Feedback Assist ID: FB22146913

Just posted this feedback regarding macOS 26 "Tahoe" (FB19853155) - please support with additional submissions if you share my view. I will miss the beautiful and individual designed icons of the past! "macOS 26 is enforcing squicles for app icons, falling back to a grey background for 3rd party apps without a compliant AppIcon asset. As a result many original app icons are reduced in size and hard to distinguish because they share the same background color. Although I respect Apple's strive for an iOS-like UI on Macs, a smooth transition path would be more user- and developer-friendly ... e.g. with some info.plist property to opt-out icon migration, potentially ignored by a future macOS version. The current solution causes a bad usability, and makes the system look inconsistent as many - especially free - software will not be updated with new icon designs. Please reconsider this bad design decision!"

Hello! I'm trying to implement an editor app (macOS) that allows the user to write code, which will be compiled and executed, showing the result in the editor window. Imagine it like SwiftUI previews, but the graphic output is created with Metal, not SwiftUI. I found that IOSurface can be used to share that kind of data over XPC, so I would not have to rely on the private NSRemoteView. However, I'm confused if it is, at all, possible for my editor app to connect to an XPC Service, that was NOT bundled with it (but compiled by it at runtime). I succeeded to launch an XPC service defined as: <?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd"> <plist version="1.0"> <dict> <key>Label</key> <string>com.myteam.myproject.service</string> <key>MachServices</key> <dict> <key>com.myteam.myproject.service</key> <true/> </dict> <key>Program</key> <string>/Path/to/service/run_my_service.sh</string> </dict> </plist> But the call to let connection = NSXPCConnection(machServiceName: "com.myteam.myproject.service") let proxy = connection.remoteObjectProxyWithErrorHandler { error in continuation.resume(throwing: error) } as? MyServiceProtocol fails with "The connection to service named com.myteam.myproject.service was invalidated: Connection init failed at lookup with error 3 - No such process." I have added <key>com.apple.security.temporary-exception.mach-lookup.global-name</key> <array> <string>com.myteam.myproject.service</string> </array> to my entitlements. Since the tutorials I followed are quite old, I'm wondering if support for something like this was dropped at some point. Thanks for any advice!

In SwiftUI 6 when developing for MacOS in Xcode 26, there does not seem to be a way to provide an automatic increment of the current build number, i.e. "Current Project Version" in settings. Does anyone have a solution that works in the latest version of Xcode. i.e. 26? All the solutions that I have tried are based on older versions that involve running scripts to update the build file, (.xcconfig) or the .info.plist file which simply will not work in the current configuration.

I was recently charging my macbook, just like any normal time i do. i expected to see the battery at 80%, but the macbook was hiding a surprise for me. the magsafe glowed amber, so i knew it cant be on 80%, so when i opened the lid, the surprise was a shocking 94%. I didn't know what to expect, but my macbook is running macOS tahoe 26.5, so everything seemed normal to me until: someone please help in fixing this bug, as it isn't the first time this happens to me. please link this to my magsafe 4 idea here, as the magsafe could activate pulsing amber (1 second on 1 second off, for both amber and yellow pulsing lights) in case of failure to stop on set limit, making it visually much easier to understand if there is an issue.

As part of the OpenJDK testing we run several regression tests, including for Java SE networking APIs. These APIs ultimately end up calling BSD socket functions. On macos, starting macos 26, including on recent 26.2 version, we have started seeing some unexplained but consistent exception from one of these BSD socket APIs. We receive a "ENOBUFS" errno (No buffer space available) when trying to construct a socket(). These exact same tests continue to pass on many other older versions of macos (including 15.7.x). After looking into this more, we have been able to narrow this down to a very trivial C code which is as follows (also attached): #include <stdio.h> #include <sys/socket.h> #include <string.h> #include <unistd.h> #include <sys/errno.h> static int create_socket(const int attempt_number) { const int fd = socket(AF_INET6, SOCK_STREAM, 0); if (fd < 0) { fprintf(stderr, "socket creation failed on attempt %d," " due to: %s\n", attempt_number, strerror(errno)); return fd; } return fd; } int main() { const unsigned int num_times = 250000; for (unsigned int i = 1; i <= num_times; i++) { const int fd = create_socket(i); if (fd < 0) { return -1; } close(fd); } fprintf(stderr, "successfully created and closed %d sockets\n", num_times); } The code very trivially creates a socket() and close()s it. It does this repeatedly in a loop for a certain number of iterations. Compiling this as: clang sockbufspaceerr.c -o sockbufspaceerr.o and running it as: ./sockbufspaceerr.o consistently generates an error as follows on macos 26.x: socket creation failed on attempt 160995, due to: No buffer space available The iteration number on which the socket() creation fails varies, but the issue does reproduce. Running the same on older versions of macos doesn't reproduce the issue and the program terminates normally after those many iterations. Looking at the xnu source that is made available for each macos release here https://opensource.apple.com/releases/, I see that for macos 26.x there have been changes in this kernel code and there appears to be some kind of memory accountability code introduced in this code path. However, looking at the reproducer/application code in question, I believe it uses the right set of functions to both create as well as release the resources, so I can't see why this should cause the above error in macos 26.x. Does this look like some issue that needs attention in the macos kernel and should I report it through feedback assitant tool?

I am coding in Perl and I want to create Environment Variables (External to the app) specifically for this app only, however my custom ENV variables are deliberately blocked by the OS for security purposes (as described in documentation).

Hi, I’m pretty new to AppKit and I’m trying to make an NSTextField inside an NSTableView both: Editable Multi-line / wrapping Right now, wrapping works fine until I set: tf.isEditable = true Then the text becomes a single line. How do I make it editable while still wrapping correctly? import AppKit final class ViewController: NSViewController, NSTableViewDataSource, NSTableViewDelegate { let tableView = NSTableView() let text = String(repeating: "A", count: 500) override func viewDidLoad() { super.viewDidLoad() view = tableView tableView.addTableColumn(NSTableColumn()) tableView.usesAutomaticRowHeights = true tableView.dataSource = self tableView.delegate = self } func numberOfRows(in tableView: NSTableView) -> Int { 1 } func tableView(_ tableView: NSTableView, viewFor tableColumn: NSTableColumn?, row: Int) -> NSView? { let cell = NSTableCellView() let tf = NSTextField(wrappingLabelWithString: text) tf.lineBreakMode = .byCharWrapping if let tableColumn { tf.preferredMaxLayoutWidth = tableColumn.width } tf.isEditable = true // comment this out and wrapping works cell.addSubview(tf) tf.translatesAutoresizingMaskIntoConstraints = false NSLayoutConstraint.activate([ tf.leadingAnchor.constraint(equalTo: cell.leadingAnchor), tf.trailingAnchor.constraint(equalTo: cell.trailingAnchor), tf.topAnchor.constraint(equalTo: cell.topAnchor), tf.bottomAnchor.constraint(equalTo: cell.bottomAnchor), ]) return cell } }

Hello, I'm evaluating possibility to use virtualization to setup on-permise parallel testing system for a product I work on. My compatibility range is wide, i.e. it would cover macOS 12 - 26, so any testcase needs to be executed on at least 5 different OS versions (and introducing any parallization to the test execution would mean I deal with higher number of VMs). As far as I understand, there is a constraint in Apple Software License, that limits number of OS VM installations per one physical system to 2 (section 2.B.(iii)) in case of OS downloaded from the Internet or through App Store. Clearly these days a single high-end Mac hardware could sustain more than 2 VMs running in parallel. The license also mentions it is also possible to be in a volume or maintenance license program and then the terms of this program apply instead. So I wonder how do people normally deal with the above limitation? What is the path I need to follow if I want to be able to run more than "2 additional copies or instances of" macOS on my VMs?

Hello, with macOS' and iOS' recent updates (26.4.x), Handoff in my app appears to have become dysfunctional, as I receive the following message in the logs: getContinuationStreamsWithCompletionHandler(), inputStream=(null) outputStream=(null) error=Error Domain=NSPOSIXErrorDomain Code=94 "Bad message" UserInfo={NSLocalizedDescription=Could not set up internal streams} It ONLY works: From a Mac with 15.x to other Macs or iOS devices (on 26.4.x). All other combinations (see below) do NOT work: iOS (26.4.x) > iOS (26.4.x) iOS (26.4.x) > Mac (26.4.x) iOS (26.4.x) > Mac (15.x) Mac (26.4.x) > iOS (26.4.x) The Handoff prompt is shown in the Dock on Mac, in the app switcher on iOS, and in the Dock on iPadOS, but once I click/press it, I get the continuationStreams error. Now, is it my app, or the OS? Thank you kindly, – Matthias

In our macOS solution, we have a few processes and a few plugin modules which communicate with each other over XPC. We have recently started enforcing library validation flag along with hardened runtime for all processes and plugins. To enforce that, we are trying to get signing information from the XPC audit token using SecCodeCopySigningInformation with kSecCSDynamicInformation flag. As per documentation, this flag requires a live SecCode not SecStaticCode to be passed to SecCodeCopySigningInformation. However, SecCodeCopySigningInformation explicitly requires SecStaticCode in its parameters. So I am unsure how to pass live SecCode to SecCodeCopySigningInformation without copying SecStaticCode from it using SecCodeCopyStaticCode. Force cast from SecCode to SecStaticCode fails. Is unsafeBitCast a valid option in this case? Note that we support macOS version 12 and later.

In macOS Tahoe 26.4.1 we noticed that when we use our custom authorisation plugin to perform unlock using SF Authorisation Plugin View, the SF window isnt focused. User has to manually click on it to type in the password. We also noticed that this wasnt the case in macOS Tahoe 26.2 . Can you kindly suggest us if any flags have to be enabled for the same? Any help on this issue is highly appreciated

I'm writing a read-only filesystem extension. I see that the documentation for loadResource(resource:options:replyHandler:) claims that the --rdonly option is supported, which suggests that this should be possible. However, I have never seen this option provided to my filesystem extension, even if I return usableButLimited as a probe result (where it doesn't mount at all - FB19241327) or pass the -r or -o rdonly options to the mount(8) command. Instead I see those options on the volume's activate call. But other than saving that "readonly" state (which, in my case, is always the case) and then throwing on all write-related calls I'm not sure how to actually mark the filesystem as "read-only." Without such an indicator, the user is still offered the option to do things like trash items in Finder (although of course those operations do not succeed since I throw an EROFS error in the relevant calls). It also seems like the FSKit extensions that come with the system handle read-only strangely as well. For example, for a FAT32 filesystem, if I mount it like mount -r -F -t msdos /dev/disk15s1 /tmp/mnt Then it acts... weirdly. For example, Finder doesn't know that the volume is read-only, and lets me do some operations like making new folders, although they never actually get written to disk. Writing may or may not lead to errors and/or the change just disappearing immediately (or later), which is pretty much what I'm seeing in my own filesystem extension. If I remove the -F option (thus using the kernel extension version of msdos), this doesn't happen. Are read-only filesystems currently supported by FSKit? The fact that extensions like Apple's own msdos also seem to act weirdly makes me think this is just a current FSKit limitation, although maybe I'm missing something. It's not necessarily a hard blocker given that I can prevent writes from happening in my FSKit module code (or, in my case, just not implement such features at all), but it does make for a strange experience. (I reported this as FB21068845, although I'm mostly asking here because I'm not 100% sure this is not just me missing something.)