Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
macOS 11 Big Sur Safari YouTube 4K HDR VP9 support
- Thread starter EugW
- Start date
- Sort by reaction score
Here's a sample of VTDecoderXPCService that says that the decoded frame is coming directly from the GPU driver:
Code:
Sampling process 628 for 3 seconds with 1 millisecond of run time between samples
Sampling completed, processing symbols...
Analysis of sampling VTDecoderXPCService (pid 628) every 1 millisecond
Process: VTDecoderXPCService [628]
Path: /System/Library/Frameworks/VideoToolbox.framework/Versions/A/XPCServices/VTDecoderXPCService.xpc/Contents/MacOS/VTDecoderXPCService
Load Address: 0x10d63d000
Identifier: VTDecoderXPCService
Version: ???
Code Type: X86-64
Parent Process: ??? [1]
Date/Time: 2020-08-09 08:06:11.835 +0200
Launch Time: 2020-08-09 08:05:42.596 +0200
OS Version: macOS 11.0 (20A5343i)
Report Version: 7
Analysis Tool: /usr/bin/sample
Physical footprint: 174.7M
Physical footprint (peak): 174.7M
----
Call graph:
2 Thread_8585 DispatchQueue_1: com.apple.main-thread (serial)
+ 2 start (in libdyld.dylib) + 1 [0x7fff6bc98851]
+ 2 main (in VTDecoderXPCService) + 16 [0x10d640f96]
+ 2 xpc_main (in libxpc.dylib) + 437 [0x7fff6bf254fb]
+ 2 _xpc_objc_main (in libxpc.dylib) + 825 [0x7fff6bf25a73]
+ 2 -[NSRunLoop(NSRunLoop) run] (in Foundation) + 76 [0x7fff2d3893e4]
+ 2 -[NSRunLoop(NSRunLoop) runMode:beforeDate:] (in Foundation) + 212 [0x7fff2d2fafd1]
+ 2 CFRunLoopRunSpecific (in CoreFoundation) + 563 [0x7fff2a927134]
+ 2 __CFRunLoopRun (in CoreFoundation) + 1315 [0x7fff2a927d40]
+ 2 __CFRunLoopServiceMachPort (in CoreFoundation) + 316 [0x7fff2a929650]
+ 2 mach_msg (in libsystem_kernel.dylib) + 60 [0x7fff6be07590]
+ 2 mach_msg_trap (in libsystem_kernel.dylib) + 10 [0x7fff6be0721e]
2 Thread_8595
+ 1 start_wqthread (in libsystem_pthread.dylib) + 15 [0x7fff6bed04c3]
+ ! 1 _pthread_wqthread (in libsystem_pthread.dylib) + 414 [0x7fff6bed1556]
+ ! 1 __workq_kernreturn (in libsystem_kernel.dylib) + 10 [0x7fff6be088de]
+ 1 start_wqthread (in libsystem_pthread.dylib) + 0 [0x7fff6bed04b4]
2 Thread_9123
+ 2 thread_start (in libsystem_pthread.dylib) + 15 [0x7fff6bed04d7]
+ 2 _pthread_start (in libsystem_pthread.dylib) + 224 [0x7fff6bed49b4]
+ 2 ??? (in AppleGVA) load address 0x7fff3edea000 + 0x6b7be [0x7fff3ee557be]
+ 2 ??? (in AppleGVA) load address 0x7fff3edea000 + 0x6ba03 [0x7fff3ee55a03]
+ 2 ??? (in AppleGVA) load address 0x7fff3edea000 + 0xb0978 [0x7fff3ee9a978]
+ 1 ??? (in AppleGVAVPXDecoder) load address 0x10d760000 + 0x2e1c [0x10d762e1c]
+ ! 1 _CFRelease (in CoreFoundation) + 238 [0x7fff2a9f2910]
+ ! 1 BBufFinalize (in CoreMedia) + 144 [0x7fff2bdae860]
+ ! 1 receivedMemoryAllocator_Deallocate (in CoreMedia) + 338 [0x7fff2be8b5ab]
+ ! 1 receivedMemoryAllocator_tellOriginToDecrementUseCountOfBlock (in CoreMedia) + 270 [0x7fff2be8b6e5]
+ ! 1 -[OS_xpc_object dealloc] (in libxpc.dylib) + 17 [0x7fff6bf157ec]
+ ! 1 _xpc_dictionary_dispose (in libxpc.dylib) + 34 [0x7fff6bf15bbf]
+ ! 1 _xpc_dictionary_node_free (in libxpc.dylib) + 68 [0x7fff6bf15cdf]
+ ! 1 -[OS_xpc_object dealloc] (in libxpc.dylib) + 47 [0x7fff6bf1580a]
+ ! 1 _objc_rootDealloc (in libobjc.A.dylib) + 62 [0x7fff6aa15f42]
+ ! 1 objc_destructInstance (in libobjc.A.dylib) + 146 [0x7fff6aa16011]
+ ! 1 objc_object::sidetable_clearDeallocating() (in libobjc.A.dylib) + 83 [0x7fff6aa16315]
+ ! 1 _os_unfair_lock_lock_slow (in libsystem_platform.dylib) + 162 [0x7fff6bec5145]
+ ! 1 __ulock_wait (in libsystem_kernel.dylib) + 10 [0x7fff6be0896e]
+ 1 ??? (in AppleGVAVPXDecoder) load address 0x10d760000 + 0x2ee4 [0x10d762ee4]
+ 1 VTDecoderSessionEmitDecodedFrame (in VideoToolbox) + 1247 [0x7fff375093c8]
+ 1 ??? (in VideoToolbox) load address 0x7fff37500000 + 0xf8cdd [0x7fff375f8cdd]
+ 1 ??? (in VideoToolbox) load address 0x7fff37500000 + 0xe4074 [0x7fff375e4074]
+ 1 xpc_connection_send_message_with_reply_sync (in libxpc.dylib) + 238 [0x7fff6bf1b852]
+ 1 dispatch_mach_send_with_result_and_wait_for_reply (in libdispatch.dylib) + 50 [0x7fff6bc57abe]
+ 1 _dispatch_mach_send_and_wait_for_reply (in libdispatch.dylib) + 518 [0x7fff6bc576c6]
+ 1 mach_msg (in libsystem_kernel.dylib) + 60 [0x7fff6be07590]
+ 1 mach_msg_trap (in libsystem_kernel.dylib) + 10 [0x7fff6be0721e]
1 Thread_8591
+ 1 start_wqthread (in libsystem_pthread.dylib) + 15 [0x7fff6bed04c3]
+ 1 _pthread_wqthread (in libsystem_pthread.dylib) + 414 [0x7fff6bed1556]
+ 1 __workq_kernreturn (in libsystem_kernel.dylib) + 10 [0x7fff6be088de]
1 Thread_8594
+ 1 thread_start (in libsystem_pthread.dylib) + 15 [0x7fff6bed04d7]
+ 1 _pthread_start (in libsystem_pthread.dylib) + 224 [0x7fff6bed49b4]
+ 1 _dispatch_worker_thread (in libdispatch.dylib) + 284 [0x7fff6bc505e3]
+ 1 _dispatch_semaphore_wait_slow (in libdispatch.dylib) + 58 [0x7fff6bc42066]
+ 1 _dispatch_sema4_timedwait (in libdispatch.dylib) + 76 [0x7fff6bc41c3a]
+ 1 semaphore_timedwait_trap (in libsystem_kernel.dylib) + 10 [0x7fff6be07272]
1 Thread_8594 DispatchQueue_93: vtdecoder-connection-event-queue-0x7f9331419a00 (serial)
+ 1 thread_start (in libsystem_pthread.dylib) + 15 [0x7fff6bed04d7]
+ 1 _pthread_start (in libsystem_pthread.dylib) + 224 [0x7fff6bed49b4]
+ 1 _dispatch_worker_thread (in libdispatch.dylib) + 218 [0x7fff6bc505a1]
+ 1 _dispatch_root_queue_drain (in libdispatch.dylib) + 326 [0x7fff6bc507a5]
+ 1 _dispatch_lane_invoke (in libdispatch.dylib) + 426 [0x7fff6bc4805e]
+ 1 _dispatch_lane_serial_drain (in libdispatch.dylib) + 263 [0x7fff6bc47407]
+ 1 _dispatch_mach_invoke (in libdispatch.dylib) + 498 [0x7fff6bc59508]
+ 1 _dispatch_lane_serial_drain (in libdispatch.dylib) + 263 [0x7fff6bc47407]
+ 1 _dispatch_mach_msg_invoke (in libdispatch.dylib) + 441 [0x7fff6bc58997]
+ 1 _dispatch_client_callout4 (in libdispatch.dylib) + 9 [0x7fff6bc417c7]
+ 1 _xpc_connection_mach_event (in libxpc.dylib) + 935 [0x7fff6bf1c140]
+ 1 _xpc_connection_call_event_handler (in libxpc.dylib) + 56 [0x7fff6bf1d2d4]
+ 1 ??? (in VideoToolbox) load address 0x7fff37500000 + 0xe4f33 [0x7fff375e4f33]
+ 1 ??? (in AppleGVAVPXDecoder) load address 0x10d760000 + 0x2c8d [0x10d762c8d]
+ 1 ??? (in AppleGVA) load address 0x7fff3edea000 + 0xb0609 [0x7fff3ee9a609]
+ 1 ??? (in AppleGVA) load address 0x7fff3edea000 + 0xafab6 [0x7fff3ee99ab6]
+ 1 ??? (in AppleGVA) load address 0x7fff3edea000 + 0x6b6b8 [0x7fff3ee556b8]
+ 1 ??? (in AppleGVA) load address 0x7fff3edea000 + 0x6b77d [0x7fff3ee5577d]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x2474b [0x11436874b]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x277456 [0x1145bb456]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x277312 [0x1145bb312]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x20c222 [0x114550222]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x210005 [0x114554005]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x6b65d [0x1143af65d]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x65cae [0x1143a9cae]
+ 1 ??? (in AppleIntelKBLGraphicsVADriver) load address 0x114344000 + 0x253b7e [0x114597b7e]
+ 1 IOAccelResourceFinishEvent (in IOAccelerator) + 148 [0x7fff529e6354]
+ 1 IOConnectCallMethod (in IOKit) + 186 [0x7fff2da5491f]
+ 1 io_connect_method (in IOKit) + 383 [0x7fff2da54b08]
+ 1 mach_msg (in libsystem_kernel.dylib) + 60 [0x7fff6be07590]
+ 1 mach_msg_trap (in libsystem_kernel.dylib) + 10 [0x7fff6be0721e]
1 Thread_8597
1 start_wqthread (in libsystem_pthread.dylib) + 0 [0x7fff6bed04b4]
Total number in stack (recursive counted multiple, when >=5):
Sort by top of stack, same collapsed (when >= 5):I have no idea how you can say that's not hardware accelerated.
Interesting. Not the case on my MacBook and my Radeon VII based Hackintosh. Thanks for finally sharing a valid proof. 
‘Software encoding’ need only be mentioned when a browser app isn’t using Intel Quick Sync or discreet GPU or if that browser needs to be updated for sites like YouTube. There are examples from members above of some browsers failing to use Quick Sync even when the operating system (Big Sur and Windows) allows it.
Good job Apple but bad job Firefox and Brave for not using Quick Sync or improving YouTube support. Partial VP9 support was introduced with Skylake and full support came with Kabylake, Coffeelake etc
There is no VP9 support in quick sync in ivy bridge and no quick sync at all in Xeon x-5690 what I was testing with. VP8 support starts with Broadwell and partial VP9 support starts with Skylake. Kaby Lake, Coffee Lake, Whiskey Lake, Comet Lake add full fixed-function VP9 8-bit and 10-bit decoding acceleration and 8-bit encoding acceleration. So everything would be either software or hardware (through the video card) decoding.
Last edited:
It doesn‘t work like that on macOS. Software doesn‘t have direct access to the decoders and encoders. They need to use Apple‘s VideoToolbox (AppleGVA) framework to implement it. It‘s Apple who did the poor job not supporting VP9 for so long. Now as they do, other browsers beside Safari will for sure offer accelerated VP9 playback as well.
I know that and that is what I’m saying. They should be updated for Big Sur by now after 2 months and 4 betas.
VP9 hardware decoding has been added in beta 4, so less than a week ago. Firefox and Chrome surely will add support for it soon.
oh yes you’re right. The hot weather and this messy thread made me confused 😝😂
Anyway I don’t use the other browsers. I just want Safari and Big Sur to make better use of GPUs in every way possible.
a webm downloaded from youtube? what process using to try? could attempt it here.
I already posted a link from Intel indicating the Haswell and Broadwell support for partial VP9 decode acceleration, but here it is again:
Intel® Iris™ and HD Graphics Driver update posted for Haswell and Broadwell version 15.36.14.4080
This update addresses several previously found issues. Please see the Release Notes and excerpt below for details. As always, feedback is appreciated. The 15.36.14.4080 has been posted to Intel Download Center at the following direct links to the drivers: 32-bit...
community.intel.com
Intel® Iris™ and HD Graphics Driver update posted for Haswell and Broadwell version 15.36.14.4080
New features with this release:
- Improved video playback through partial hardware acceleration support for the VP9 video format.
- Additional Open CL extension support:
- cl_intel_subgroups – this extension enables work items to share data without use of shared local memory
- cl_intel_advanced_motion_estimation – provides access to Intel's motion estimation hardware acceleration block that can be used by media processing applications including custom transcoders and image stabilization
- Support for OpenGL extension:
- GL_EXT_polygon_offset_clamp
- GPU accelerated decode of HEVC video file format including both 8-bit and 10-bit support. This will provide improved video playback capabilities on the platforms listed in the following table.
It will never be supported on the iMac Pro nor the mac pro with the current graphic cards on offer as they do not support Vp9.
hardware decoding.
Only Navi based AMD card can do VP9.
At some stage on Windows, AMD had written an OpenCL accelerated driver, performance were awful and power usage very high (over 250W for the graphic card alone)
They have disabled that driver since.
If you look at the webkit source code, you'll see that they only report vp9 as available (software) if using a desktop, or using a laptop when plugged on power.
For the brave, you can test firefox with VP9 hardware decoder support added (Big Sur only)
It also use CoreAnimation so will perform similarly to Safari when it comes to power usage.
This will play 4K @ 60 fps on my macbook pro 2018 using less than 15% CPU usage; and the app isn't even showing in the list of application using power.
It also use CoreAnimation so will perform similarly to Safari when it comes to power usage.
This will play 4K @ 60 fps on my macbook pro 2018 using less than 15% CPU usage; and the app isn't even showing in the list of application using power.
For now Safari only supports webm via Media Source Extension (which is what youtube is using) you can't directly play a webm file. WebM in MSE was only enabled 2 weeks ago https://bugs.webkit.org/show_bug.cgi?id=214529
It sucks in Windows because Apple EFI disable the intel GPU when booting into Windows ; you only get the AMD graphic card which can't do VP9 (only the navi based one like the 5700 can).
So no thanks to Apple for crippling the performance when booted under Windows.
This partial acceleration is useless. It's so bad that even Intel removed it from their drivers within 1 year of releasing it.
As far as features goes, you can consider both Haswell and Broadwell to *not* have VP9 at all ; they will do VP8 however.
Big Sur only supports VP9 hardware acceleration for Intel macs > Skylake.
This is the info I was given. TBH, I don't know if it's > or >=
The new MBP 2020 with a navi AMD could do it; but at this stage it seems that the VP9 HW decoder is always using the Intel CPU. I don't have any other machines to try with but this macbook pro.
I have an iMac Pro which will never do it (Xeon CPU don't have the required HW and the AMD Vega definitely can't do VP9)
I've noticed that the performance of VP9 playback when connected to an external screen sucked big time. Like you see the CPU usage shoot to 200% and the video completely freeze. It seems that the VideoToolbox reverts temporarily to software decoding whenever the playback window is covered by another or when switching from full screen to windowed.
In the Firefox image I linked to earlier; the VideoToolbox VP9 decoder is only used if it fully supports hardware decoding; otherwise it will use the ffvp9 software decoder.