Linux Support Expectations For The AMD Radeon RX 6000 Series

[Ftelf]

Not quite. There are a few raytracing items for simulation of thermal dynamics that are done from the sources. It roughly 99.9 percent of ray tracing implementations are from the camera and 0.1 are only for on supercomputers simulating real world physics and are done from sources. Yes the other way requires monster amount of processing power in the class of start at 100000 GPU and up with processing time even with all this power in weeks to months so extremely inefficient is a understatement but when you are modelling something like a rocket engine where you must be exact being inefficient is not the problem exactness is.

So the other way you are not going to see in games or desktop applications because if you tried to run them on a desktop computer level of processing power you would not render a single frame in a complete human life time so the computer would likely die before it gave any results.

Ftelf it does not pay to use the every word without at least some restruction. Every real implementation of raytracing on deskitop computers casts rays from cameras itself is true. Every real implementation of ray-tracing on any class system its not true there is just a small corner case on supercomputers. Some people get their understanding of how raytracing is done by reading supercomputer white papers that is basically wrong in all general cases of raytracing.

Nice to have some additional information about somebody who knows what he's talking about and do it without unnecessary emotions and offending others
I was kinda focused on say software and implementation what common-user could interact with. And speaking of wider group of raytracing algorithms as you include some where there is not even a camera (I did not even though about them), we could probably add Global Illumination (GI) algorithms which are tracing from the light source to surfaces and are just used as a lightmaps for secondary classic raytrace. I think that that was demonstrated on PS5 in real time, pretty impressive. When I was younger I did played with 3DSMax a lot and this was extremely time-consuming

[mdedetrich]

....

Apologies for quoting again, but I found the original reference issue (actually Level1 techs did a video on it), you can see it here https://www.youtube.com/watch?v=k-tnFzPaVBY .

There is a kernel patch to fix the issue, but according to them a proper FLR implementation would be ideal ¯\_(ツ)_/¯

[oiaohm]

Apologies for quoting again, but I found the original reference issue (actually Level1 techs did a video on it), you can see it here https://www.youtube.com/watch?v=k-tnFzPaVBY .

There is a kernel patch to fix the issue, but according to them a proper FLR implementation would be ideal ¯\_(ツ)_/¯

That a 2019 video just before BACO reset appears. This is before AMD has pulled the problem complete apart. Please note level1 has missed that the cards are also disappearing random-ally when your reboot the complete system because on some of the cards PCIe hot reset is not working 100% of the time either. The horrible here the issues that make a PCIe hot reset not dependable also make a FLR for of reset not dependable you see this in Nvidia cards where neither works dependably. When you have a card with a miss behaving hot reset or FLR the solution is you have to take control of the cards power management chip that is what baco does for AMD cards. Guess what you know that signed power management firmware you need so you can clock up and down nvidia cards that ends up containing hardware quirk fixes. There is one solution to these problems be it AMD or Nvidia and its not implement FLR.

AMD developers have been implementing BACO reset on all cards that can support it.

There is a lot of miss information around this bug

https://wiki.gentoo.org/wiki/GPU_pas...Navi_reset_bug

AMD Navi 10 series GPUs require a vendor specific reset procedure. According to AMD a PSP mode 2 reset should be enough however at this time the details of how to perform this are not available.

You see this is that statement correct no because how to is in the Linux kernel driver and was there the complete time.

https://www.kernel.org/doc/html/latest/gpu/amdgpu.html

GPU reset method (-1 = auto (default), 0 = legacy, 1 = mode0, 2 = mode1, 3 = mode2, 4 = baco)

Yep method 3 is PSP mode 2 has been there since the AMDGPU driver was added to the Linux kernel. The windows driver also has a register setting to change reset mode. Only one problem Navi 10 cards some models mode2 works other cards it does not. Some mode2 works random-ally. Navi cards have issues.

Proper final fix for Navi is baco reset.

[mibo]

Concerning the AMD SAM / BAR support:
https://www.igorslab.de/en/b550-or-x...-to-reinstall/

It seems, one needs to set the BIOS from CSM to UEFI mode to really, really activate the BAR support.
I guess, most Linux users have CSM enabled - as I do. And with this, it seems the BAR=256MB is a fixed value.

[Zan Lynx]

Concerning the AMD SAM / BAR support:
https://www.igorslab.de/en/b550-or-x...-to-reinstall/

It seems, one needs to set the BIOS from CSM to UEFI mode to really, really activate the BAR support.
I guess, most Linux users have CSM enabled - as I do. And with this, it seems the BAR=256MB is a fixed value.

Why in the world would you be using CSM? For what?

I think I disagree with your "most Linux users" because there are several of us where I work and we all use UEFI and most of us use Secure Boot, except for the poor guys who have an Nvidia card.

If you have a problem with Secure Boot just turn it off. There's no reason to use CSM.

If you think you need CSM then you probably installed your OS the wrong way.

[mibo]

Why in the world would you be using CSM? For what?

I think I disagree with your "most Linux users" because there are several of us where I work and we all use UEFI and most of us use Secure Boot, except for the poor guys who have an Nvidia card.

If you have a problem with Secure Boot just turn it off. There's no reason to use CSM.

If you think you need CSM then you probably installed your OS the wrong way.

Why in the world would you be using UEFI? For what?

up to today I saw not a single good reason for using it.
But maybe I try it out later.

[Zan Lynx]

Why in the world would you be using UEFI? For what?

up to today I saw not a single good reason for using it.
But maybe I try it out later.

Because it boots faster with Fast Boot. It's the default setting and you have to go out of your way to change it.

[agd5f]

Why in the world would you be using UEFI? For what?

up to today I saw not a single good reason for using it.
But maybe I try it out later.

CSM is the legacy backward compatible interface. A lot of features including extended MMIO over 4GB, require UEFI.

[mibo]

CSM is the legacy backward compatible interface. A lot of features including extended MMIO over 4GB, require UEFI.

Because, UEFI was problematic for Linux in the beginning (signed bootloader, etc.) I up to now never used it. But for extended BAR/SAM I will try it out.
Can you say, if the AMD graphics drivers on Linux will use/benefit from extended BAR/SAM functionality?
(if you only can answer after the 18th of November - it's fine for me :-)

[fenixex]

Because, UEFI was problematic for Linux in the beginning (signed bootloader, etc.) I up to now never used it. But for extended BAR/SAM I will try it out.
Can you say, if the AMD graphics drivers on Linux will use/benefit from extended BAR/SAM functionality?
(if you only can answer after the 18th of November - it's fine for me :-)

The performance penalty is game dependent. For me the fps in Witcher 3 doubled (~50 to ~110) after disabling the option while for BL3 it was only an increase of around 10-15fps in total. See bug 1039