There's Finally A Decent Vulkan Ray-Tracing Benchmark

[bug77]

The OpenBenchmarking.org screenshots I took, but the scenes are static so basically appear the same...

Ok, so I learned one thing about the benchmarks. Only the lights move. Neat.
Not much to screenshot there, I guess.

[blacknova]

Ok, so I learned one thing about the benchmarks. Only the lights move. Neat.
Not much to screenshot there, I guess.

So, they simplified it as much as possible -> no BVH updates and I'd guess that is most expensive operation.

[bridgman]

Just had a nagging thought - the Vulkan ray tracing extension may have only been provisional at Navi21 launch. It appears that we might have enabled RT support after the extension was ratified, and then re-spun the 20.45 driver package for Windows but not for Linux. Checking now.

OK, looks like the 20.45 release does not include the RT extensions since it came out a week before the spec was ratified, but the upcoming 20.50 release should include the RT extensions. Sorry for the noise.

[qarium]

I checked with our Vulkan team and they confirmed that we do support ray tracing in the closed source Vulkan driver for both Linux and Windows.
Just had a nagging thought - the Vulkan ray tracing extension may have only been provisional at Navi21 launch. It appears that we might have enabled RT support after the extension was ratified, and then re-spun the 20.45 driver package for Windows but not for Linux. Checking now.
EDIT - I'm now getting conflicting answers about RT support in the Linux driver - hold on until we get consensus.

amd should stop to totally confuse their customers with multible drivers.
AMDVLK vs RADV+ACO
in my point of view this situation of multible drivers just hurt us.
so please just stop AMDVLK on linux. also stop the closed source OpenGL driver and use the opensource RadeonSI openGL driver for the workstation tasks to.
i am sick of all the people confused by amd and we have to explain the driver situation what driver they should use on every forum thread. this Confusion is really irritating.
for years it maybe was a idea to use multible diverend drivers for different tasks but right now AMD with RADV+ACO wins every gaming benchmark on linux against Nvidia. so it is really time to stop this Confusion.

don't you get the point that the PR of AMD is hurt just because of the confusion of the multible drivers?

whats the point of closed source AMDGPU-PRO OpenGL driver if RadeonSI opensource OpenGL driver wins EVERY benchmark against Nvidia?...

i really do not get the point.

also put SR-IOV GPU virtualization on every GPU even gaming GPUs... the time is over you have to admit today everybody need some SR-IOV GPU virtualization to play games with the ability to protect against trojan horses in closed source and or cracked games.

[JoshuaAshton]

OK, looks like the 20.45 release does not include the RT extensions since it came out a week before the spec was ratified, but the upcoming 20.50 release should include the RT extensions. Sorry for the noise.

dude just release it lol

[bridgman]

dude just release it lol

Yep, now that we know it's not already out there that's what we are going to do. The 20.50 release is already close to completion and the Vulkan team tells me that their 20.50 code already includes RT extensions.

And yes I have asked why the Linux team was not contacted for the initial release after spec ratification.

[qarium]

Sorry for the noise.

Noise?... its Confusion... amd should take simple steps to stop confuse people.

tell your openGL PRO customers to test RadeonSI openGL and report bugs they need to be fixed.
to finally end this closed source opengl driver.

add SR-IOV GPU virtualization to any future GPUs anyone need this in modern times
your PRO customers will buy PRO cards with ECC and other true PRO featurs.
if you have to increase price for this?.. well look at the market the people pay crazy high price for GPUs right now
in germany a 6900XT is at 1389€ right now and a nvidia 3090 is at 2000€
do you really think the people dont buy a 6900XT with SR-IOV it it costs 100€ more? means 1489€ ?

it is just a way to confuse customers if they buy a 1400€ card from amd and can not play a game in Virtual Machine at full speed.

stop confuse linux customers with multible drivers if you read in the forum people report massiv bugs with the AMDVLK as soon as they install RADV+ACO the bugs are gone.

it is insanity to confuse your customers like this.

[smitty3268]

Yep, now that we know it's not already out there that's what we are going to do. The 20.50 release is already close to completion and the Vulkan team tells me that their 20.50 code already includes RT extensions.

And yes I have asked why the Linux team was not contacted for the initial release after spec ratification.

Can you share why it's not being released in amdvlk? Hopefully there aren't legal or trade secret issues with open sourcing it.

[Michael]

OK, looks like the 20.45 release does not include the RT extensions since it came out a week before the spec was ratified, but the upcoming 20.50 release should include the RT extensions. Sorry for the noise.

Ahhh great, had been banging my head against the wall today trying to figure out why I couldn't get the current driver exposing it via vulkaninfo or the benchmark... (And also pondering why it wasn't mentioned in the Linux driver change-log if was the case.)

[qarium]

Ahhh great, had been banging my head against the wall today trying to figure out why I couldn't get the current driver exposing it via vulkaninfo or the benchmark... (And also pondering why it wasn't mentioned in the Linux driver change-log if was the case.)

not your fault... maybe just an accident but the feeling that AMD thinks linux people are second class citizens because of the 1% market share never go away...

google stadia just shows the world is insane native linux version for home customers is not possible but native linux version for linux servers for the use of google statdia is possible.

i do not know the true reason why this is so but i can imagine something like this: DRM/Copyprotection don't work for linux this means there will never be linux end-user native linux version of the game. but Cloud gaming/streeming of the game from google stadia is the perfect copy protection so native linux version for stadia servers is perfectly fine.

what a sick mindset. feels like they do not only think linux users are second class citizens but in addition of this they trust a crack whore with HIV on the street more than linux users.