Why would scaling impact M1 iMac performance but not MacBook Air?

[PatriotInvasion]

I switched from a 27" 5K iMac to a 24" M1 iMac. I see in Display preferences you can replicate the same "Looks like 2560x1440" real estate as the 27" iMac, but with the disclaimer "Using a scaled resolution may affect performance."

Now, I know this disclaimer has existed here for years dating back to the intro of Retina, but I don't understand why they don't show this for the MacBook Air or Pro - both of which ship using a scaled (non-2X) Retina resolution by default. Surely, whatever performance impact that results is affecting every MacBook Air and MacBook Pro they sell right out out of the box.

With that said, I'm guessing its fine to just use it if I prefer More Space, but curious if anyone has further insight into this discrepancy in messaging across iMac and the notebook line.

 

[rkuo]

Likely to just be a benign messaging issue or an oversight. They have to choose how and where they think it makes enough of a difference to mention a performance hit. You are correct that the performance hit should apply to all scaled resolutions ... although it's going to be a much bigger hit on a bigger monitor in general.

 

[PatriotInvasion]

Likely to just be a benign messaging issue or an oversight. They have to choose how and where they think it makes enough of a difference to mention a performance hit. You are correct that the performance hit should apply to all scaled resolutions ... although it's going to be a much bigger hit on a bigger monitor in general.

I doubt it was an oversight. Someone decided that scaling the MacBook Air and Pro resolutions by default was a good decision for users. It's likely that because of this decision, they are intentionally hiding this disclaimer on the notebook line to avoid the criticism that would surely occur.

The result of using the More Space option on the iMac and the Default option on a MacBook Air/Pro is the same. The scaling is affecting the performance to some degree.

 

[rkuo]

I doubt it was an oversight. Someone decided that scaling the MacBook Air and Pro resolutions by default was a good decision for users. It's likely that because of this decision, they are intentionally hiding this disclaimer on the notebook line to avoid the criticism that would surely occur.

The result of using the More Space option on the iMac and the Default option on a MacBook Air/Pro is the same. The scaling is affecting the performance to some degree.

I agree the messaging should be consistent across all platforms if it's the same amount of scaling involved. Maybe someone took the lazy way out and only updated it on one's where scaling was the default.

 

[Falhófnir]

I can only think that there's a lot more pixels to push on the iMac means the penalty is still significant for that computer, while it's become negligible for the MacBooks at this point? The computer does have to draw an image 4x the 'looks like' resolution, and for scaled ones it then has to interpolate it to the physical resolution, and it has to do it 60 times per second.

2560x1440 = 3.7M pixels (14.7M at retina resolution) while the MacBooks' 1440x900 = 1.3M (5.2M at retina). So that's about 3x the work for the same chip in the iMac before you even ask it to do anything else?

 

[kvic]

I like the fact that OP is paying attention to small details and share with a wider audience.

I think post #5 builds the technical background of an explanation. The crucial point is that when doing such scaling on non-integral multipliers, it hurts GPU performance. This is not new. People have been complaining since Apple introduced 'Retina' display though the noise wasn't big enough to make an impact on Apple's course.

Since I'm made aware of such a thing in recent years, personally I always hesitate hooking any 4K monitors to a Mac. I believe the performance hit is more exacerbated in M1 due to its integrated CPU+GPU design. Hence, the warning/disclosure that possibly also waives any legal liability on Apple's part.

If possible, could you check GPU power consumption at 1) default and 2) "look like 1440p" in two scenarios A) idle and B) say playback a 720p video?

 

[PatriotInvasion]

I can only think that there's a lot more pixels to push on the iMac means the penalty is still significant for that computer, while it's become negligible for the MacBooks at this point? The computer does have to draw an image 4x the 'looks like' resolution, and for scaled ones it then has to interpolate it to the physical resolution, and it has to do it 60 times per second.

2560x1440 = 3.7M pixels (14.7M at retina resolution) while the MacBooks' 1440x900 = 1.3M (5.2M at retina). So that's about 3x the work for the same chip in the iMac before you even ask it to do anything else?

I appreciate the logic, but I'm almost certain the Retina MacBooks had this disclaimer prior to 2016 when they decided to make the scaled resolutions the "Default for display" on the 13" and 15" MacBook Pro. It was then that the disclaimer was removed from the Display preferences despite the same physical resolution as the versions they replaced.

 

[joevt]

For a 5K display, "Looks like 2560x1440" does not require scaling (since it's actually 5120x2880 pixels).

The iMac has a 4.5K so only the non-HiDPI mode of 4.5K or the HiDPI mode of "Looks like 4.5/2K" require no scaling. "Looks like 1440p" does require scaling for the iMac's 4.5K screen. I don't know why the MacBook Air and MacBook Pro don't have the same disclaimer. All of their scaled resolutions are less than 1080p though. That might be the reason - while they are doing scaling, the number of input pixels (<4K) and the number of output pixels (1600p) are much lower than what the iMac needs to deal with.

 

[gilby101]

1) Makes a difference whether you have a graphics card with its own memory or just using general purpose RAM like the 24" iMac and other M1 Macs.

2) When you use 2x scaling (looks like 2230x1260) macOS and apps draw on a 4480x2520 (twice the size) virtual display (in RAM) which is the same size as the physical display. This consumes part of the RAM (8 or 16 GB).

3) When its says "looks like 2560x1440", macOS and all apps will draw on a virtual display which is 5120x2880 (twice the looks like value). The screen driver will then downsize to the size of the physical screen 4480x2520. So that will use extra RAM for the virtual display as well as the RAM for the physical display.

So non-default scaling will certainly effect performance compared with default 2x scaling on an 8GB iMac due to the extra virtual display, which is much bigger than any virtual display on the small screen laptops.

>> Get 16GB RAM if you want non default resolution.

 

[PatriotInvasion]

1) Makes a difference whether you have a graphics card with its own memory or just using general purpose RAM like the 24" iMac and other M1 Macs.

2) When you use 2x scaling (looks like 2230x1260) macOS and apps draw on a 4480x2520 (twice the size) virtual display (in RAM) which is the same size as the physical display. This consumes part of the RAM (8 or 16 GB).

3) When its says "looks like 2560x1440", macOS and all apps will draw on a virtual display which is 5120x2880 (twice the looks like value). The screen driver will then downsize to the size of the physical screen 4480x2520. So that will use extra RAM for the virtual display as well as the RAM for the physical display.

So non-default scaling will certainly effect performance compared with default 2x scaling on an 8GB iMac due to the extra virtual display, which is much bigger than any virtual display on the small screen laptops.

>> Get 16GB RAM if you want non default resolution.

Makes sense I suppose, but the More Space scaled resolution on the 16" MacBook Pro ("Looks like 2048x1280") is virtually rendering a canvas of 4096x2560 and scaling down to the 3072x1920 physical resolution.

So you're saying this doesn't warrant the performance disclaimer because of the discrete GPU in the MBP along with the virtual resolution still being significantly less than 5120x2880? If that's the case, its the answer to my original question.

 

[joevt]

Makes sense I suppose, but the More Space scaled resolution on the 16" MacBook Pro ("Looks like 2048x1280") is virtually rendering a canvas of 4096x2560 and scaling down to the 3072x1920 physical resolution.

So you're saying this doesn't warrant the performance disclaimer because of the discrete GPU in the MBP along with the virtual resolution still being significantly less than 5120x2880? If that's the case, its the answer to my original question.

I have an eGPU for my Mac mini 2018. It gets the disclaimer for any scaled resolution for a 4K (3840x2160) display including smaller ones such as "looks like 1504 x 846". There's no disclaimer for "looks like 1920x1080".

I have another 4K display that shows a "Looks like 3840x2160" option but it's not a HiDPI mode (7680x4320) - instead, it's a low resolution mode 3840x2160 (but macOs doesn't call it low resolution because it's the native resolution of the display). Do this: take a screenshot then view the info to see how many pixels were saved.

 

[AAPLGeek]

I have an eGPU for my Mac mini 2018. It gets the disclaimer for any scaled resolution for a 4K (3840x2160) display including smaller ones such as "looks like 1504 x 846". There's no disclaimer for "looks like 1920x1080".

I have another 4K display that shows a "Looks like 3840x2160" option but it's not a HiDPI mode (7680x4320) - instead, it's a low resolution mode 3840x2160 (but macOs doesn't call it low resolution because it's the native resolution of the display). Do this: take a screenshot then view the info to see how many pixels were saved.

How is your experience running the scaled resolution closest to looks like 1920x1080 without the eGPU? Like 2304x1296 via something like EasyRes.

 

[gilby101]

So you're saying this doesn't warrant the performance disclaimer because of the discrete GPU in the MBP along with the virtual resolution still being significantly less than 5120x2880? If that's the case, its the answer to my original question.

Yep. MBP's GPU has its own RAM, smaller virtual display is secondary.

 

[joevt]

How is your experience running the scaled resolution closest to looks like 1920x1080 without the eGPU? Like 2304x1296 via something like EasyRes.

In the Displays preferences panel, the next step up from "Looks like 1080p" for Intel GPU on Mac mini 2018 with 4K DisplayPort display is "Looks like 1440p". Holding the Option key down and clicking Scaled has many modes between 1152x648 up to 3360x1890 with 3840x2160 at the top (there's a couple modes between 1080p and 1440p). With SwitchResX, I can get access to the non-HiDPI modes up to 6720x3780 (which is what "Looks like 3360x1890" uses). Of course, with SwitchResX, I can add any scaled mode I want up to 8Kx8K for Intel GPU and 16Kx16K for AMD GPU.

I can't really say I notice a difference in performance with any of the modes. I've never run the benchmarks to find out for sure if there's a difference between the modes. Generally, I choose a mode that fits what I'm doing - not because of how it affects performance.

 

[kvic]

I can't really say I notice a difference in performance with any of the modes. I've never run the benchmarks to find out for sure if there's a difference between the modes. Generally, I choose a mode that fits what I'm doing - not because of how it affects performance.

First of all, thanks for the last two posts. I like the details of anecdotes.

I expected your dGPU to burn more electricity sometimes even at idle if Radeon driver happens to be buggy for a given macOS. Otherwise, if you're not pushing the GPU to its limit, I believe in most scenarios won't offset performance other than running at somewhat higher wattage.

I expect this is more troublesome for iGPU inside Intel (same could apply to M1) as power budget is shared by both CPU+GPU. If GPU is burning more, not only limiting its max performance but also cutting into CPU's available power budget, limiting max core boost.

(See if @MrMacintoshBlog with his M1 iMac is interested in the tests I suggested in my previous post :-D)

 

[driven01]

I found this thread as I've recently run into this issue on my 4K display.

I find it interesting that on the internal display there isn't a resolution that shows issues related to scaling. Just the external display.

Do we either need to get a native 2560x1440 display or a 5k display to avoid these scaling issues?

 

[gilby101]

Do we either need to get a native 2560x1440 display or a 5k display to avoid these scaling issues?

That will depend on the specifics of the display and your sensitivity to any hopefully minor issues. But with macOS, for the best size for text and graphics along with clearest resolution you need a display with about 220 ppi. That is 5K for 27" display or 4K for 23.5" display. For other sizes and ppi, you need to test with each display. As you are probably aware, there are many people who like 4K 32" displays - each according to their own.

 

[joevt]

Do we either need to get a native 2560x1440 display or a 5k display to avoid these scaling issues?

If they are really issues. I haven't seen actual benchmarks or measurements to indicate the severity of the issues.

 

[kvic]

Re-visiting the thread near a year after I still find the small details posted here interesting.

I didn't and still don't think extra memory usage (for the off-screen frame buffer) will cause performance penalty because it's too minor compared to multiple GiB main memory.

Scaled resolutions causing less crispy display is real. The perception by people is, however, quite subjective. If 'performance' is meant by Apple regarding this then the warning shall show for all non-integrally scaled resolutions. But they don't. Hence, we probably can rule this out.

The rendering of displayed objects to off-screen framebuffer requires more power for a larger virtual screen. If this is part of Apple's "performance warning", then again non-integrally scaled resolutions should carry the message. Hence, we can rule this out too.

Now seems the remaining down-scaling portion is the only possible culprit. Down-scaling requires work either by CPU, GPU or some ASICs. CPU is obviously less power efficient than the other two options.

So why do Intel MBA & MBP not carry the warning message? But M1 & AMD dGPU do? One possibility: where Quick Sync is it's used for down-scaling and it's very power efficient. Hence, nearly no performance penalty.

When Quick Sync (or ASICs) is not available, general purpose pixel shaders on GPU are used for this job. That uses up power, and potentially momentarily stall GPU's rendering pipeline at regular intervals and hence hit on FPS (if playing games).

--

The above is just my theory. Interested folks can help to verify by performing experiments:

1. I was told in the other thread the GPU in M1 reports idle power of only "15mW". Set your iMac to a scaled resolution, the GPU is expected to report way higher power consumption.

2. The same experiment as in #1 but perform on a Radeon eGPU.

3. Put your iMac back to native resolution. Run Unigene Valley benchmark (@1080p for comparison purpose) and jot down the average FPS score. Now set your iMac to a scaled resolution. Run the same benchmark again. Check if any non-trivial drop in FPS score.

4. The same experiment as in #3 but perform on a Radeon eGPU.

 

[Tagbert]

I found this thread as I've recently run into this issue on my 4K display.

I find it interesting that on the internal display there isn't a resolution that shows issues related to scaling. Just the external display.

Do we either need to get a native 2560x1440 display or a 5k display to avoid these scaling issues?

I’ve notice the difference messaging on an M1 Air between the internal display (no message) and an external 4K display (message about performance). There may be thresholds beyond which they choose to display that message.

In practice, the performance impact seems quite theoretical. I haven’t noticed any performance hit, whatsoever on a scaled resolution. Perhaps on older systems with slower display controller that might have been an issue but now, it seems like a relic.

 

[kvic]

The above is just my theory. Interested folks can help to verify by performing experiments

Answering myself and providing one data point. I tried the experiments on a 13 inch Intel MBP with only iGPU and no external monitor attached.

Observations:

1. Idle GPU power consumption is same between Default and any Scaled Resolutions.

2. Idle system power consumption is roughly same between Default or any scaled resolutions.

3. Valley benchmark average FPS is about 3-5% loss when comparing Scaled Resolutions against Default

Conclusions:

Not much could be drawn with one data point. However, it appears indicative down-scaling happening in Quick Sync indeed (?)

 

[kagharaht]

Answering myself and providing one data point. I tried the experiments on a 13 inch Intel MBP with only iGPU and no external monitor attached.

Observations:

1. Idle GPU power consumption is same between Default and any Scaled Resolutions.

2. Idle system power consumption is roughly same between Default or any scaled resolutions.

3. Valley benchmark average FPS is about 3-5% loss when comparing Scaled Resolutions against Default

Conclusions:

Not much could be drawn with one data point. However, it appears indicative down-scaling happening in Quick Sync indeed (?)

Thanks for the Benchmarks. I've been looking for this information! Thank you.