RADV Gets A Big Performance Boost Thanks To DCC

[phoronix]

Phoronix: RADV Gets A Big Performance Boost Thanks To DCC

Not only is Vega now Vulkan 1.0 conformant with RADV, but it's also much faster if pulling down the very latest Mesa bits...

http://www.phoronix.com/scan.php?page=news_item&px=RADV-DCC-Performance-Boost

[FireBurn]

very good... i think the Open-Source AMD Vulkan AMDVLD driver will lose any performance benefit over the RADV in less than 3 months..

I think it has already in some benchmarks

[boxie]

sweet as! RADV is looking good - I wonder how much AMDVLK driver can get from it to push it's FPS higher :P

[oooverclocker]

It would be amazing to see new benchmarks with color compression. As long as it's lossless there can't be any complaints. Probably RADV will do better than expected when there is still headroom for such improvements.

[humbug]

very good... i think the Open-Source AMD Vulkan AMDVLD driver will lose any performance benefit over the RADV in less than 3 months..

Hard to say.

according to Bridgman up-to now the performance optimization on their official Vulkan codebase has been focused around windows. So once they spend some time and focus on Linux performance a bit expect amdvlk also to get a lot quicker. Consider how blazing fast their Windows driver is (comfortably beats competing Nvidia cards).

[Michael]

It would be amazing to see new benchmarks with color compression. As long as it's lossless there can't be any complaints. Probably RADV will do better than expected when there is still headroom for such improvements.

As said in the article, benchmarks are coming.

[Kemosabe]

AMD wasted a lot money and effort by developing something in house which is never going to be used :-D

[duby229]

I think it has already in some benchmarks

Yes it has. Radv definitely has lower CPU utilization, so it actually is doing less work than amdvlk. Smaller is basically always better. In every single scenario a lighter program that can do the same thing as a heavier program will perform better than the heavier program.

[duby229]

Hard to say.

according to Bridgman up-to now the performance optimization on their official Vulkan codebase has been focused around windows. So once they spend some time and focus on Linux performance a bit expect amdvlk also to get a lot quicker. Consider how blazing fast their Windows driver is (comfortably beats competing Nvidia cards).

Yes, that's true, but I think what you're seeing there is hardware differences. Nvidia tweaks their architectures to get the best possible fixed function hardware performance, AMD tweaks their architectures to get the best possible 3d engine and compute performance. It makes perfect sense why. Nvidia has a highly developed DX9-11 driver, which existed long before DX12-Vulkan, and so they designed their architectures for the drivers they were developing. AMD on the other hand designed Mantle, which was a new idea and so they designed their architectures for -that- driver, which also suits DX12 and Vulkan.

It's not hard to say, it's plainly obvious.

[numacross]

AMD wasted a lot money and effort by developing something in house which is never going to be used :-D

Isn't the AMD's driver shared with Windows? In either case opening drivers is always a good thing, even if there's already another one. Having a choice is always preferable to being locked-in