M5 Pro external 5K 165Hz display: Window animations and scrolling UI appear to render at ~60Hz/jitter while cursor remains perfectly smooth

Hello Apple engineers,

I’m trying to determine whether what I’m seeing is expected behavior or a software issue with the new M5 Pro platform.

System

  • MacBook Pro (M5 Pro)
  • Latest macOS Beta
  • External 5K 165Hz monitor connected via DisplayPort
  • Refresh rate correctly detected as 165Hz

What I observe

The display itself is clearly running at 165Hz.

For example:

  • Mouse cursor movement is extremely smooth.
  • Dragging the desktop by holding an empty area is also perfectly smooth.

However:

  • Moving application windows feels much closer to 60Hz.
  • Scrolling in Safari, Chrome and other applications also appears to run at a much lower frame rate than the display refresh rate.
  • Mission Control animations sometimes show similar micro-stutters.

This makes the cursor and desktop movement noticeably smoother than normal window animations.

⸻

Troubleshooting already performed

  • Different DisplayPort cables
  • Different timing configurations
  • Different resolutions / HiDPI modes
  • DSC enabled and disabled
  • Refresh rate confirmed at 165Hz
  • Same behavior across multiple applications

The issue appears unrelated to the monitor itself because the cursor is clearly rendered at the full refresh rate.

⸻

Additional observation

Interestingly, I previously used another external 4K 144Hz HDR monitor and did not notice this behavior.

I also found another M5 Pro user reporting nearly the same issue:

  • external 165Hz display
  • smooth cursor
  • window dragging jitter / micro-stuttering

At the same time, I haven’t found similar reports from M4 Pro or the base M5 running the same monitor.

⸻

My question

Could this be related to the new M5 Pro display pipeline (WindowServer, Display Engine, or DCP)?

Is there any known issue regarding high-refresh-rate external displays on the M5 Pro platform?

Or is there additional diagnostic logging (WindowServer, DCP, Metal, etc.) that would help identify whether frames are actually being presented at the display refresh rate?

I’d be happy to provide:

  • sysdiagnose
  • WindowServer logs
  • Screen recordings
  • Display timing information
  • IORegistry dumps

if they would be helpful.

Thank you!

Answered by DTS Engineer in 916500022

Thanks for filing FB23616959, and for listing the cables, timings, resolutions, and DSC settings you've already ruled out. The report is open and under investigation.

Please continue to test your software in release and pre-release versions of system software made available to you through your developer.apple.com account. When you do, add details about that into your bug report to keep it up to date. Please see developer.apple.com/download.

When you update the report, include a sysdiagnose captured while the stutter is happening. The Profiles and Logs page has instructions for capturing one in macOS. I can't tell from here what's causing the stutter, and I don't know of a documented setting that logs WindowServer frame presentation rates.

The WindowServer crash you describe is an assertion in CA::WindowServer::AppleDisplay::max_src_rect_width_by_pipes(). Please file it as a separate bug report, because the crash and the stutter are separate work items. Attach the crash logs and a sysdiagnose, and post the FB number here.

If you're seeing similar stutter on a high-refresh-rate external display, please file your own report using Feedback Assistant. Include the Mac model, the display model, how it's connected, the refresh rate, and a sysdiagnose captured while the stutter is happening. Then post your FB number here.

system version is 27.0 Beta (26A5378j)

Case / Feedback Reference Feedback ID: FB23616959 (Captured after the system format)

Additional Findings from Crash Analysis

After analyzing multiple WindowServer crash reports, I noticed that the crashes are highly consistent.

Both crash logs terminate with an internal assertion (__assert_rtn) inside the same Apple display pipeline function:

CA::WindowServer::AppleDisplay::max_src_rect_width_by_pipes()

The call stack is consistently:

__assert_rtn CA::WindowServer::AppleDisplay::max_src_rect_width_by_pipes() CA::WindowServer::Server::get_display_info() _XGetDisplayInfo

This suggests that WindowServer is failing while calculating display pipeline parameters during display information processing, rather than due to a GPU hang, Metal failure, or a third-party application.

I also reviewed the CPU Resource Diagnostic generated for WindowServer. During periods of high CPU usage, most of the activity is concentrated in WindowServer’s display update and Core Animation pipeline (display updates and rendering preparation), rather than GPU recovery or driver timeout paths.

In addition, I did not find evidence of AGX GPU panics, IOAccelerator timeouts, Metal command buffer failures, or DisplayLink-related components in the collected logs.

These observations suggest that the issue may be located within WindowServer’s AppleDisplay/display pipeline implementation when driving this external high-resolution, high-refresh-rate display configuration.

I also had the same problem with the M5 pro running a 4K 160hz display. I did the same troubleshooting steps to no avail. It could most likely be as software bug, for now I guess we have to wait for newer updates of macOS to resolve this issue.

great

Im feeling the same here. My cursor feel slow on Macbook M5 Pro. I dont feel that on my previoous macbook. The same mouse and monitor looks smooth on Linux and Windows.

Thanks for filing FB23616959, and for listing the cables, timings, resolutions, and DSC settings you've already ruled out. The report is open and under investigation.

Please continue to test your software in release and pre-release versions of system software made available to you through your developer.apple.com account. When you do, add details about that into your bug report to keep it up to date. Please see developer.apple.com/download.

When you update the report, include a sysdiagnose captured while the stutter is happening. The Profiles and Logs page has instructions for capturing one in macOS. I can't tell from here what's causing the stutter, and I don't know of a documented setting that logs WindowServer frame presentation rates.

The WindowServer crash you describe is an assertion in CA::WindowServer::AppleDisplay::max_src_rect_width_by_pipes(). Please file it as a separate bug report, because the crash and the stutter are separate work items. Attach the crash logs and a sysdiagnose, and post the FB number here.

If you're seeing similar stutter on a high-refresh-rate external display, please file your own report using Feedback Assistant. Include the Mac model, the display model, how it's connected, the refresh rate, and a sysdiagnose captured while the stutter is happening. Then post your FB number here.

Feedback report draft

Suggested title: M5 Pro drops about 40% of frames on an external 5K 165 Hz display during window drags; WindowServer main thread is blocked in IOMFB swap_submit / DCPLink::rpc. M4 Max on the same display and macOS build delivers every frame.

Area: Displays / WindowServer (performance)

Type: Performance, UI stutter

Summary

On a MacBook Pro with M5 Pro, moving a window (or switching Spaces) on an external high-refresh display visibly judders. The display controller's own counters show that during a steady window drag only about 101 new frames per second are delivered for 165 refreshes. A Mac with M4 Max, on the same display, same cable type, same macOS build and the same mouse, delivers 164 to 166 of 165.

The CPU and GPU are mostly idle while this happens. A spindump taken during a drag shows the WindowServer main thread blocked for about 45% of wall time inside CA::IOMobileFramebuffer::swap_end / kern_SwapBegin, waiting in DCPLink::rpc, and contending on a lock with the hardware cursor thread, which spends 78% of its time in the same calls.

Configuration
AffectedComparison
MachineMacBook Pro (Mac17,8), Apple M5 Pro, 18 cores (6 super + 12 performance), 48 GBMac with Apple M4 Max, 64 GB
macOS27.2.1 (26B131)27.2.1 (26B131)
Energy mode on power adapterHigh PowerHigh Power
Built-in displayOff (lid closed)Off (lid closed)

Display: 27GX-Ultra, 5120 x 2880 at 165 Hz, "looks like 2560 x 1440" (HiDPI), SDR, fixed refresh rate, USB-C (DisplayPort). Both Macs drive it through two display pipes (DCPEXT1 + DCPEXT2 on the M5 Pro, DCPEXT2 + DCPEXT3 on the M4 Max).

Input: USB mouse at 1000 Hz (same mouse on both Macs). The built-in trackpad shows the same judder, less strongly.

The judder is also visible on a Samsung Odyssey G60SD (2560 x 1440 at 360 Hz, HDMI). I have frame counters only for the 5K display.

Steps to reproduce
  1. Connect the 5K display over USB-C, set 165 Hz, close the lid, hide all apps except Finder.
  2. Drag a Finder window in circles for about 20 seconds.
  3. Watch the motion, or count frames with the attached dragprobe4 (it reads the IOReport groups DCPEXTn / swap and DCPEXTn / scanout once per second).

A hands-free variant: run the attached movewin, which moves its own window in circles from a display link with no mouse input.

Expected

Every refresh gets a new frame while the window is moving, as on the M4 Max.

Actual

All figures are new frames delivered to the panel per second (sum of swap / subframes(n)), against refreshes per second (sum of scanout / subframes(n)).

Scenario on the 5K displayM5 ProM4 Max
Mouse drag, 165 Hz101 of 165 (96 to 107)164 to 166 of 165
Mouse drag, 120 Hz94 of 120not measured
Mouse drag, 120 Hz, recorder in passive mode86 of 122 (79 to 93)not measured
Window moved by script, no input, 165 Hz84 to 90 of 167[to add]
Window moved by script while the mouse is wiggled85 of 167[to add]
Mouse wiggled, nothing else moving1 to 2[to add]

Load during the mouse drag at 165 Hz:

M5 ProM4 Max
GPU active residency20%, all in the lowest state (338 MHz), about 0.2 W23%, all in the lowest state (338 MHz)
Fastest CPU cluster46% active, about 3690 MHz average75 to 95% active, about 4450 MHz
Second CPU cluster40% active, about 1500 MHz70 to 88% active, about 2500 MHz
leftMouseDragged events delivered per secondabout 440about 760
WindowServer CPUabout 36% of one corenot measured

Forcing the clocks up on the M5 Pro recovers only part of the gap:

Condition during a mouse drag, 165 HzNew frames per second
Nothing102
One spinning user-interactive thread (super cluster at 4608 MHz)126
Background Metal compute load (GPU at about 1590 MHz)123
Both128
Trace findings

Spindump, 8 s, 801 samples at 10 ms, taken during a mouse drag at 165 Hz (116 to 124 frames per second were delivered while it ran). Sample counts out of 801:

WindowServer ws_main_thread (2.41 s CPU, 3.62 GHz average):

  • 464 in CGXUpdateDisplay
  • 361 in CA::WindowServer::IOMFBServer::finish_skylight_update
    • 245 in CA::IOMobileFramebuffer::swap_end -> kern_SwapEnd -> IOMobileFramebufferShim::swap_submit
      • 173 asleep in IOMobileFramebufferAP::swap_submit_dcp -> DCPLink::rpc -> rpc_caller_gated -> IOWorkLoop::sleepGate
      • 53 in IORecursiveLockLock, blocked by a turnstile waiting for thread 0xe4f (the cursor thread)
    • 116 in kern_SwapBegin -> IOMobileFramebufferUserClient::s_swap_start
      • 74 asleep in IOMobileFramebufferAP::swap_start_dcp -> DCPLink::rpc
      • 42 in IORecursiveLockLock, blocked by the same cursor thread
  • about 61 in compositing and Metal command buffer submission
  • 291 idle in mach_msg

WindowServer thread com.apple.coreanimation.cursor.external-0 (0.11 s CPU):

  • 625 in CA::WindowServer::IOMFBServer::cursor_callback
    • 334 in CA::IOMobileFramebuffer::swap_end -> swap_submit_dcp -> DCPLink::rpc (asleep)
    • 291 in CA::WindowServer::IOMFBDisplay::set_cursor_surface -> kern_SwapBegin

Kernel threads active in the window all ran at about 1.2 to 1.36 GHz average (cycles divided by CPU time): IOGPU Completion (0.25 s), AGXHardwareWorkLoop (0.19 s), four AppleDCPLinkServiceSoC threads, IOSurfaceRoot, AppleT8142USBXHCI. sched_update_powered_cores -> ApplePMGR::enableCPUCluster also appears.

powermetrics over the same 8 s: P1-Cluster 63% active with 61% of the time at 1344 MHz (its lowest state); P0-Cluster down 89% of the time; S-Cluster 83% active at about 3680 MHz (this includes the profilers themselves).

Ruled out
  • Variable refresh rate, HDR, HDMI versus USB-C, built-in display on versus off: the shortfall is present with VRR off, SDR, USB-C and the lid closed.
  • GPU capacity: 20% active at the lowest performance state. Geekbench Metal and OpenCL scores are normal for this chip.
  • Memory pressure and thermal throttling: 90% memory free, no swap, no thermal warnings.
  • Other apps: all hidden during the measurements.
  • Recorder overhead: a passive mode that only reads the counters once per second gives the same result.
  • Refresh rate: at 120 Hz the Mac still delivers only 86 to 94 frames per second.
  • The cursor alone: a scripted window move with no mouse input is still at about half rate. Moving the cursor at the same time costs about 5 more frames per second.
Notes

The reading of the swap and scanout counters is my own: scanout tracks the refresh rate exactly, and swap falls to 0 to 2 per second when nothing on screen changes and matches the refresh rate on the M4 Max during a drag.

Thanks for the detailed draft, especially the M4 Max comparison on the same display and macOS build. Please file it using Feedback Assistant. Attach a sysdiagnose and the spindump, both captured during a window drag. Then post the FB number here so we can make sure it gets routed to the right team.