xf86-video-modesetting Driver Optimization Helps Conserve Intel Power Consumption

[phoronix]

Phoronix: xf86-video-modesetting Driver Optimization Helps Conserve Intel Power Consumption

A change merged to the X.Org Server Git for the generic xf86-video-modesetting DDX driver is helping conserve some power consumption at least for Intel graphics by determining the optimal hardware cursor size...

https://www.phoronix.com/news/Optimal-HW-Cursor-Size

[varikonniemi]

Why even support larger cursors in the first place? They look better. So intel "fixes" their driver by forcing a worse looking cursor.

[binarybanana]

Why even support larger cursors in the first place? They look better. So intel "fixes" their driver by forcing a worse looking cursor.

That's not what the patch does. It just picks the smallest size overlay that fits the cursor, either 64x64, 128x128, 256x256, etc. Usually cursors fit into 64x64 or maybe 128x128. No need for a 256x256 overlay in most cases, but if you pick a huge cursor it should still work.

[heliosh]

Uhm.. does this only save power, while the cursor is moved?

[FireBurn]

You either conserve power, or lower power consumption

[varikonniemi]

That's not what the patch does. It just picks the smallest size overlay that fits the cursor, either 64x64, 128x128, 256x256, etc. Usually cursors fit into 64x64 or maybe 128x128. No need for a 256x256 overlay in most cases, but if you pick a huge cursor it should still work.

So intel was using a 256x256 buffer but only filling it up with 64x64 pixels?

[binarybanana]

So intel was using a 256x256 buffer but only filling it up with 64x64 pixels?

Hard to say if it's only Intel driver defaulting to 256x256 (could be the same on AMD/Nvidia, etc.) or if that even is the maximum cursor size on Intel or other hardware (could be higher), but yes, as far as I understand you get a 256x256 overlay for the cursor by default and with a smaller cursor image most of that is simply transparent. What's surprising is the difference in power consumption. Cursor size should not have that much of an effect, not when the difference in size is so small. The patch suggests it has to do with how Intel driver handles buffers for cursor images.

[kerneltoast]

Uhm.. does this only save power, while the cursor is moved?

While the cursor is visible on the screen, actually. On my Alder Lake laptop, this change reduced power consumption by 2 W while the cursor was visible on the screen.

[Vorpal]

What on earth would make a larger cursor buffer use so much more power? That seems absurd, and worth trying to optimise: Presumably a 0x0 cursor would use even less power. Since reducing the cursor to 0x0 is obviously absurd, it seems worth investing time in making it more efficient instead of making users have to pick small cursors for power efficiency.

Another reason that seems more useful: Someone with vision problems or on a high DPI screen might want a larger cursor and they shouldn't be penalised for that.

[loganj]

While the cursor is visible on the screen, actually. On my Alder Lake laptop, this change reduced power consumption by 2 W while the cursor was visible on the screen.

how do you measure that?