new | show | ask | jobs Github

yencabulator 5 days ago

Yes clearly that's why the from-scratch RADV driver was often faster.

The dirty open secret in the tech industry is that the special sauce almost always just isn't all that special.

▲

fidotron 5 days ago | parent [-]

> Yes clearly that's why the from-scratch RADV driver was often faster.

Because AMD didn't actually care about the Linux driver since it didn't make them moeny.

> The dirty open secret in the tech industry is that the special sauce almost always just isn't all that special.

Only among people where that's true. In the computer industry just look at the M series of chips where it's very clear that their direct competitors can't establish why it does what it does.

▲

yencabulator 5 days ago | parent [-]

> can't establish why it does what it does

This is weird Apple fanboy head-in-the-sand thinking. The Mx chips have been dug into plenty and are just good engineering, not magic. AMD's horribly-named "Ryzen AI Max+ 395" chip is definitely moving in the same direction.

▲

fidotron 5 days ago | parent [-]

Right, so we have other ARM64 implementation that are this good?

Bonus points for ones without ex-Apple employees involved in their design, because maybe those people might know something about it.

▲

michaelmrose 5 days ago | parent [-]

Qualcomm Snapdragon X Elite. Qualcomm has been making ARM chips prior to Apple. Admittedly its 4nm vs 3nm for the M4.

	▲	fidotron 5 days ago \| parent [-]
		Those did involve ex-Apple people, and there isn't proof that they are quite as good either, but they are the closest that anyone has publicly come. Qualcomm have never actually caught up with Apple performance wise since the introduction of Arm64. They had a very nice 32 bit implementation and were completely caught off guard. Prior to their NuVia acquisition their 64 bit efforts were barely improvements on what you can just license from Arm directly, to the point for a while that is all they were.