Comment by jorvi

Comment by jorvi a day ago

Really cool stuff! Especially nice to see the groundwork being laid for what could become very efficient handhelds, considering how much performance Apple's M-series and Qualcomm's Elite series with relatively few watts. Much better than AMD, Intel or Nvidia.

One nit: it's too bad Valve / Igalia choose to completely ignore the lessons from Bazzite.

Bazzite already runs a scheduler like LAVD, called BORE[0]. It would have saved them a lot of work to extend and improve that rather than invent the wheel again. I'm not sure if Valve and Igalia are unaware of Bazzite and BORE or if this is a case of NIH.

[0]https://github.com/firelzrd/bore-scheduler

Maxious 15 hours ago

Igalia benchmarked against EEVDF https://youtu.be/wQbiqKUIsMI?si=rT-zMXJkVR6RYG_D&t=2353

Also bazzite uses LAVD by default for steam deck hardware https://universal-blue.discourse.group/t/bazzite-buzz-18/379...

Reply View 1 reply

jorvi 12 hours ago

EEVDF is not BORE. Well, technically BORE is something wrapped around either CFS or EEVDF. Interestingly enough, specifically for high sustained load situations (which handhelds are likely to hit given their limited power budget), the CFS version of BORE actually performs better than the EEVDF version.
And Bazzite used to use BORE, they just like tracking upstream as much as possible.
Anywho who is to say, maybe LAVD is indeed better. We don't know.

Reply View | 0 replies

bigyabai 3 hours ago

> Much better than AMD, Intel or Nvidia.

My impression is that Nvidia still leads in power-efficiency when compared node-for-node. The Switch 2 is a miracle beyond anything Apple or AMD ever did with 8nm silicon.

Everyone else is shipping less power-efficient raster hardware unless you're super pedantic about idle draw.

Reply View 0 replies