RADV Driver Sees Dramatic Improvement To Reduce CPU Overhead For Draw Calls

[kozman]

Such draw. Much win! And he's only had this profiling tool up and running for a short amount of time. Imagine the other stuff he's going to find once he continues in his endeavors and improves his tool. AMDGPU-PRO driver be like...I'll hold your beer.

[ms178]

I hope the Intel GPU driver developers around here took notice of Mike's tool, Igor's Lab did some great analysis on ARC's drawoverhead on Windows. And I am sure that it's also bad on Linux right now. I am also sure that Mike will want to profile your driver on Zink sooner or later, too.

[rmfx]

He's unstoppable.

Can't wait to see the next Radeon vulkan and vulkan+zink benchmarks.

[Mahboi]

Aaaah, modernisation. Finally meme images to illustrate articles. Zoomeria approves.

[Spacefish]

What should we do with all those free CPU cycles now?
Do online PGO for shaders? (Profile while they are running and recompile on CPU + replace)
Does RDNA2/3 chips have hardware performance counter registers which could be used for something like that?

[M@GOid]

Every morning, I wake up and walk up to my nearest mirror and repeat after my reflection three times, "The flavour (of winning), the atmosphere (of being a winner), the experience (of being #1)"

Probably the best laugh I got today.

[gnarlin]

So, when will we get new gaming benchmarks with these patches on Phoronix?

[Dukenukemx]

So, is running OpenGL through Vulkan now faster than the direct OpenGL drivers?

I don't think that'll be a thing. It'll run as good and maybe in some tests, but on average it'll still be better on OpenGL to run OpenGL.

[grigi]

That was some fantastic writing Mike Blumenkrantz!
Are you sure you're not a successful writer under a pen name?

[ET3D]

Every morning, I wake up and walk up to my nearest mirror and repeat after my reflection three times, "The flavour (of winning), the atmosphere (of being a winner), the experience (of being #1)"

Hat off for turning this into proper English.