Apple Silicon GPU support by Ruaneri-Portela · Pull Request #1470 · aristocratos/btop

Ruaneri-Portela · 2026-01-13T22:01:42Z

With help from the folks at Issues #701 , I organized it using only IOKit, which provides the data without root.

It has only been tested on my M4 Pro platform, so anyone willing to test it would be very welcome.

Apparently, we can use the IOReport framework to obtain energy data for the CPU, such as usage in watts and clock speed.

We can also use the same framework to add the ANE (Apple Neural Engine).

deckstose · 2026-01-13T22:32:54Z

So this adds a powemetrics api, and then removes it?

deckstose

Most of the code is not properly formatted.

There is a huge amount of AI assisted code in there, this is what I was getting from the issue comments (?).

Most of the code needs COMMENTS. Especially the function pointer magic.

Ruaneri-Portela · 2026-01-13T22:42:12Z

I had kept the power metrics to read the CPU data. But I think it's outside the scope to just add the GPU. I removed it for now.

With a little work, you can get the data via IOKit.

I took Emanuele Zattin's code, which was allegedly written by him using AI, and reworked some things that I thought were relevant.

I can redo it by scheduling with AI.

Ruaneri-Portela · 2026-01-14T00:03:01Z

I added some comments, but I think it might not be enough, and I haven't formatted it properly yet.

deckstose · 2026-01-14T11:36:22Z

I don't care about the boolean operators ! == not etc, it's more about indentation, line length and naming (snake_case) functions.

Also IMO comments should not include the question mark (although it's done in other parts of the code). It's fine either way, but redundant in a way. And I'm not sure if editors with their commenting feature support it.

vandabbin · 2026-01-14T15:03:36Z

Also IMO comments should not include the question mark (although it's done in other parts of the code). It's fine either way, but redundant in a way. And I'm not sure if editors with their commenting feature support it.

whoops that how I was commenting code here because almost all the comments I already saw in the code had it. I will keep that in mind in the future

deckstose · 2026-01-14T17:10:40Z

No worries, I have merged code with this style

deckstose · 2026-01-14T17:22:09Z

Please mark this ready for review when this is ready

Ruaneri-Portela · 2026-01-15T18:40:07Z

I made the changes, but I'm not sure if it's any good.

If you could take a look

deckstose · 2026-01-16T14:44:44Z

I will take a look but it's not going to be anytime soon

Copilot

Pull request overview

This PR adds GPU monitoring support for Apple Silicon Macs by leveraging IOKit framework to collect GPU metrics without requiring root privileges. This addresses issue #701 by providing a sudoless alternative to powermetrics for GPU monitoring on Apple Silicon.

Changes:

Added new IOKit utility framework for CoreFoundation object handling and IOService iteration
Implemented Apple Silicon GPU monitoring via IOAccelerator and IOReport frameworks
Refactored existing code to use new safe helper functions for CF object conversions
Integrated GPU support into build systems (Makefile and CMake)

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 17 comments.

Show a summary per file

File	Description
src/osx/smc.hpp	Removed unused `masterPort` variable
src/osx/sensors.cpp	Refactored to use new safe CF conversion helpers and moved HID declarations
src/osx/iokit.hpp	New header defining IOKit utility functions and IOReport framework bindings
src/osx/iokit.cpp	Implementation of safe CF converters and IOService iterators
src/osx/gpu.hpp	New header defining GPU monitoring classes (GPU, GPUActivities, IOGPU)
src/osx/gpu.cpp	Implementation of GPU metrics collection via IOAccelerator and IOReport
src/osx/btop_collect.cpp	Integrated GPU collection into main collection loop
src/btop_shared.hpp	Added Apple-specific shutdown namespace for GPU support
src/btop_config.cpp	Added "apple" to default shown_gpus config on macOS
src/btop.cpp	Added IOAccelerator shutdown call for macOS
Makefile	Enabled GPU_SUPPORT for macOS ARM64
CMakeLists.txt	Added iokit.cpp and gpu.cpp sources, enabled GPU_SUPPORT definition

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

aristocratos · 2026-01-19T20:17:29Z

Seems to work as intended on a Mac Mini M1 running Tahoe.

But would be great to have some testing done on other Mac hardware.

What of the code is AI generated? Because that code would need to be checked so it hasn't been "lifted" from some other project (and might have missing attribution or incompatible licensing).

Ruaneri-Portela · 2026-01-19T21:01:48Z

Seems to work as intended on a Mac Mini M1 running Tahoe.

But would be great to have some testing done on other Mac hardware.

What of the code is AI generated? Because that code would need to be checked so it hasn't been "lifted" from some other project (and might have missing attribution or incompatible licensing).

My hardware is an M4 Pro, and that's where I tested it.

Regarding the code being marked by AI, and because of Emanuele Zattin's contribution, parts of what the code estimates to be CPU usage were inspired by AI. I say inspired because I didn't request a code per se, but rather the procedure.

Our reference code is https://github.com/dehydratedpotato/socpowerbud
Whose code is MIT.

deckstose

I've pointed out a couple of patterns, they should apply everywhere.

I can't say much about the actual implementation, but it looks good over all.

@aristocratos

I wonder if it is necessary to have a struct indicating no GPU support, the undefined GPU_SUPPORT macro should be enough, right?

btop/src/btop_shared.hpp

Lines 192 to 196 in 0c11d73

    
           #else 
        
           	struct gpu_info { 
        
           		bool supported = false; 
        
           	}; 
        
           #endif

EDIT: Resolved in #1495.

deckstose · 2026-01-19T21:35:08Z

+
+
+namespace Gpu {
+#ifdef GPU_SUPPORT


Namespace should be inside ifdef

This depends on #1495

deckstose · 2026-01-19T21:41:39Z

 namespace rng = std::ranges;
 using namespace Tools;

+


Needless newline.

deckstose · 2026-01-19T21:41:47Z

 #include "sensors.hpp"
 #endif
 #include "smc.hpp"



Needless newline.

deckstose · 2026-01-19T21:42:57Z

 		Mem::old_uptime = system_uptime();
 		Mem::collect();
+
+		#ifdef GPU_SUPPORT


Please put macros at the start of the line.

deckstose · 2026-01-19T22:03:56Z

+    };
+
+    if (auto it = map.find(key); it != map.end()) {
+        usage.*(it->second) = value;


This needs to be rewritten in a more readable syntax.

deckstose · 2026-01-19T22:06:46Z

+        proc = std::stoul(s.substr(pidPos + 4, commaPos - (pidPos + 4)));
+        name = s.substr(commaPos + 2);
+    } catch (...) {
+        return;


Missing log?

deckstose · 2026-01-19T22:11:27Z

+if(APPLE AND BTOP_GPU)
+target_compile_definitions(libbtop PUBLIC GPU_SUPPORT)
+endif()


Combine this branch with the Linux branch with GPU_SUPPORT defined

vandabbin · 2026-01-19T23:52:06Z

Seems to work as intended on a Mac Mini M1 running Tahoe.

But would be great to have some testing done on other Mac hardware.

I tested it just now on my M1 Macbook Pro running 15.7.3 and it also seems to be working. I don't have any newer apple hardware to test with though.

However maybe I'm wrong and its just apple silicon proving how efficient it is but I have trouble believing that I was getting 1 watt power usage playing 4k video. It fluctuated between .73 and 1.03

also I'm not so sure that the power meter is displaying very nicely? Those numbers just don't seem to match well. when the usage drops the meter spikes. and then half the meter is .05 watts but the full meter is 1 watt? If it is supposed to scale like that It would be nice if there was some sort of indicator. (ex. the way net has numbers to indicate the scale)

video example

Screen.Recording.2026-01-19.at.3.47.45.PM.mov

Ruaneri-Portela · 2026-01-20T00:56:28Z

Seems to work as intended on a Mac Mini M1 running Tahoe.
But would be great to have some testing done on other Mac hardware.

I tested it just now on my M1 Macbook Pro running 15.7.3 and it also seems to be working. I don't have any newer apple hardware to test with though.

However maybe I'm wrong and its just apple silicon proving how efficient it is but I have trouble believing that I was getting 1 watt power usage playing 4k video. It fluctuated between .73 and 1.03
also I'm not so sure that the power meter is displaying very nicely? Those numbers just don't seem to match well. when the usage drops the meter spikes. and then half the meter is .05 watts but the full meter is 1 watt? If it is supposed to scale like that It would be nice if there was some sort of indicator. (ex. the way net has numbers to indicate the scale)
Screen.Recording.2026-01-19.at.3.47.45.PM.mov

This metric in W is only for the GPU module, excluding use for CPU and ANE, which also need to be computed.

Considering that your video is actually running on Apple's decode ASICs and not on the GPU shaders, I see it as a correct value.

vandabbin · 2026-01-20T00:58:29Z

This metric in W is only for the GPU module, excluding use for CPU and ANE, which also need to be computed.

Considering that your video is actually running on Apple's decode ASICs and not on the GPU shaders, I see it as a correct value.

Yeah you are probably right about that.

I do still think the meter shouldn't scale as the value gets smaller but maybe that's just me. It looks super weird to be jumping from half to full for a value change of 0.05w to 0.09w. I would expect both those values to only produce 1 bar on the meter if even that. I don't usually have the GPU box open though on my other systems (relying mainly on the small GPU meter in the CPU box) so maybe that is how they all do it?

edit: just checked on my framework laptop and the gpu power meter does not scale like that on amd gpu. (picture is with super tux kart running in the background) And it idles around 5w or so

regarding scaling of the power meter. I think it would make sense for it to be scaled based on the total power usage ever expected from a given gpu model

deckstose · 2026-01-24T08:19:42Z

CI is failing

Ruaneri-Portela · 2026-01-24T10:21:59Z

CI is failing

Fixed

deckstose requested changes Jan 13, 2026

View reviewed changes

Comment thread btop.desktop

Ruaneri-Portela force-pushed the main branch 6 times, most recently from 2e6951e to 215cbe0 Compare January 13, 2026 23:58

Ruaneri-Portela force-pushed the main branch from b5fd2e7 to 39bdb08 Compare January 14, 2026 02:52

Ruaneri-Portela force-pushed the main branch from c9fd2f5 to acca72c Compare January 14, 2026 16:58

deckstose marked this pull request as draft January 14, 2026 17:16

Ruaneri-Portela force-pushed the main branch 2 times, most recently from 0331902 to 0321cd7 Compare January 14, 2026 17:44

aristocratos added the AI generated Majority of included code is AI generated label Jan 15, 2026

Ruaneri-Portela requested a review from deckstose January 15, 2026 23:18

aristocratos requested a review from Copilot January 16, 2026 22:34

Copilot started reviewing on behalf of aristocratos January 16, 2026 22:35 View session

Copilot AI reviewed Jan 16, 2026

View reviewed changes

Ruaneri-Portela force-pushed the main branch from b5d3c2d to 90cdf86 Compare January 17, 2026 01:36

aristocratos removed the AI generated Majority of included code is AI generated label Jan 19, 2026

deckstose requested changes Jan 19, 2026

View reviewed changes

This comment was marked as outdated.

Sign in to view

Ruaneri-Portela force-pushed the main branch 5 times, most recently from a19adf5 to 4348f11 Compare January 23, 2026 03:59

Ruaneri-Portela requested a review from deckstose January 24, 2026 02:28

Ruaneri-Portela force-pushed the main branch from 4348f11 to 143e27f Compare January 24, 2026 10:21

deckstose added the gpu Issues or pull requests related to GPU functionality label Jan 25, 2026

Ruaneri-Portela requested a review from aristocratos February 10, 2026 20:38

aristocratos mentioned this pull request Feb 14, 2026

feat: Add Apple Silicon GPU support via IOReport API #1541

Merged

Ruaneri-Portela force-pushed the main branch from 29e72ce to c84ca8b Compare February 16, 2026 01:40

Reseted

de3450f

Ruaneri-Portela force-pushed the main branch from c84ca8b to de3450f Compare February 16, 2026 01:43

Uh oh!

Conversation

Ruaneri-Portela commented Jan 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

deckstose commented Jan 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

deckstose left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Ruaneri-Portela commented Jan 13, 2026

Uh oh!

Ruaneri-Portela commented Jan 14, 2026

Uh oh!

deckstose commented Jan 14, 2026

Uh oh!

vandabbin commented Jan 14, 2026

Uh oh!

deckstose commented Jan 14, 2026

Uh oh!

deckstose commented Jan 14, 2026

Uh oh!

Ruaneri-Portela commented Jan 15, 2026

Uh oh!

deckstose commented Jan 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aristocratos commented Jan 19, 2026

Uh oh!

Ruaneri-Portela commented Jan 19, 2026

Uh oh!

deckstose left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vandabbin commented Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Ruaneri-Portela commented Jan 20, 2026

Uh oh!

vandabbin commented Jan 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as outdated.

deckstose commented Jan 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Ruaneri-Portela commented Jan 13, 2026 •

edited

Loading

deckstose commented Jan 13, 2026 •

edited

Loading

deckstose left a comment •

edited

Loading

vandabbin commented Jan 19, 2026 •

edited

Loading

vandabbin commented Jan 20, 2026 •

edited

Loading

deckstose commented Jan 24, 2026 •

edited

Loading