Is there still a performance warning when using scaled resolutions?

[MBHockey]

Hey all,

Not a developer, but i'm curious if there is still the message that is present in Yosemite about performance decreases when using scaled resolutions.

See the attached picture. This is from a 2015 15" rMBP on Yosemite. It's definitely true -- the iris pro struggles with normal system animations at scaled resolutions. The dGPU does not, but it'd be great to know if these were working better with iGPUs in Capitan.

 

[swestu]

Yes, it's still there. I'm running El Capitan on a Mid 2014 13" rMBP with (only) the Intel Iris.

 

[neutrak]

This really isn't the miracle update that people will have you believe. Yes it's a lot better but highest scaled resolution will still affect performance.

 

[MBHockey]

This really isn't the miracle update that people will have you believe. Yes it's a lot better but highest scaled resolution will still affect performance.

Are you running the beta? Can you comment on performance at scaled resolutions?

 

[Traverse]

There will always be a performance hit because the system has to do much more calculating.

2880/1440 = 2
2880/1680 = 1.714285714
2880/1920 = 1.5

There is a lot more calculating and estimation overhead that will affect performance. Software can make this more efficient, but ultimately more powerful hardware will make it so that the overhead difference is negligible. My 2013 15" rMBP has always ran better at 1920 than 1680 presumably because it is a cleaner calculation ratio.

 

[kwokaaron]

There will always be a performance hit because the system has to do much more calculating.

2880/1440 = 2
2880/1680 = 1.714285714
2880/1920 = 1.5

There is a lot more calculating and estimation overhead that will affect performance. Software can make this more efficient, but ultimately more powerful hardware will make it so that the overhead difference is negligible. My 2013 15" rMBP has always ran better at 1920 than 1680 presumably because it is a cleaner calculation ratio.

Huh... never really thought about why 1920 would perform better than 1680. Still though, like everyone has said, just because it may run more optimally/better doesn't change the fact that procedure-wise it still isn't as efficient as the "Default" resolution. In addition to a performance hit your battery life might also be impacted, which is why I kept it on the default resolution.

 

[bobbie424242]

I'm also curious in scaled performance with the 2014 15" rMBP with integrated graphics. In Yosemite, the Iris Pro really struggle in scaled modes (the ones where performance warning is displayed in Settings), while it is perfectly fine unscaled.

 

[MBHockey]

Huh... never really thought about why 1920 would perform better than 1680. Still though, like everyone has said, just because it may run more optimally/better doesn't change the fact that procedure-wise it still isn't as efficient as the "Default" resolution. In addition to a performance hit your battery life might also be impacted, which is why I kept it on the default resolution.

Interesting. 1680 seems to run better on my 15" rMBP (2015, AMD gpu) than 1920. maybe i'm just seeing things.

 

[Fthree]

Interesting. 1680 seems to run better on my 15" rMBP (2015, AMD gpu) than 1920. maybe i'm just seeing things.

that is a possibility because I've been running 1920 seamlessly

 

[kwokaaron]

From swestu's screenshot I see there's a new "Ambient light compensation" option. Does anyone know what that does?

 

[bobbie424242]

Can someone report if performance is increased using the Iris Pro of a rMBP, using scaled resolution on an external 4K monitor, at resolutions not an integer multiple of the monitor native resolution (eg 1296p or 1440p @ HiDpi) ? On Yosemite, such resolutions are noticeably more jerky (scrolling, animations, ...) than either 1080p HiDpi or 2160p (on a Dell P2415Q monitor whose native resolution is 3840x2160), the latter one being of course unusuable (unreadable text)

 

[xmichaelp]

There will always be a performance hit because the system has to do much more calculating.

2880/1440 = 2
2880/1680 = 1.714285714
2880/1920 = 1.5

There is a lot more calculating and estimation overhead that will affect performance. Software can make this more efficient, but ultimately more powerful hardware will make it so that the overhead difference is negligible. My 2013 15" rMBP has always ran better at 1920 than 1680 presumably because it is a cleaner calculation ratio.

That's not how it works... The high you go in scaled resolution the more pixels you're pushing.

At 1920x1200 you're pushing 3840x2400, at 1440x900 you're only pushing 2880x1440

 

[kwokaaron]

That's not how it works... The high you go in scaled resolution the more pixels you're pushing.

At 1920x1200 you're pushing 3840x2400, at 1440x900 you're only pushing 2880x1440

You're right actually. Totally forgot about this. However, if apps aren't updated (which most are), I think they're simply "pixel-doubled", which should use these scaling factors.

 

[Traverse]

That's not how it works... The high you go in scaled resolution the more pixels you're pushing.

At 1920x1200 you're pushing 3840x2400, at 1440x900 you're only pushing 2880x1440

It may just be a miscommunication of words, but the graphics chip only has to "push" 2880x1800 pixels. The system renders them at 3840x2400 and then down samples to 1920x1200 which isn't a clean downsample so there is some loss of quality and more stress on the GPU.

 

[swestu]

From swestu's screenshot I see there's a new "Ambient light compensation" option. Does anyone know what that does?

I've read it's supposed to do the same as f.lux, but I haven't seen it doing stuff yet. ^^