Linux Support Expectations For The AMD Radeon RX 6000 Series

[usta]

I am really pretty far away from amd ( for gpu side ) could some one tell me some good reasons to give a try?
For example is it have AV1 hardware encode and decode ? is it have h265/254 hardware encode / decode ? is it have v9 hardware encode / decode ?
I am asking because more than gaming i am interested in streaming and i need hardware acc for encode/decode
(And i know it is a bit hard in here not to be trolling but pretty please just 1 time <troll off> )

[pal666]

The diffrence between them is 8 compute units (72 vs 80) clocked the same way, but also both cards have exactly same TDP

novideo has cheating tdp, amd has total board power

So in reality you will have slighty faster 72 compute units vs slighty sloer 80 compute units.

in reality not all pieces of silicon are the same

Honestly from price point it doesn't compute.

it's called binning. i wonder how nvidiots compute novideo top prices

[fenixex]

Don't mind me. I'm just waiting for piotrj3 to say that 3090 is a Titan to prove him wrong

[polarathene]

novideo card has some magical qualities which make price irrelevant

Hey you moronic imbecile! 😂

CUDA and proprietary software for professional work is a perfectly valid magical quality that NVIDIA has over AMD. It's considerably more niche I guess, but perfectly valid reason to not buy an AMD GPU.

I would love the new AMD cards tbh, but if they're incompatible with industry software (Photogrammetry comes to mind), not much that can be done about that. The open-source alternatives don't come close, and one of the leading open-source ones that comes to mind also only has CUDA and lacks ROCm implementation, so that must be saying something? (apart from ROCm support on Navi GPUs with Linux being in a bad state last I heard)

[zoomblab]

The open source driver delivery to the end users is fundamentally broken. By end users I always mean consumers, common people that want nothing to do with terminals, source codes and third party software. It is fundamentally broken because it relies on provisioning and syncing hardware releases with the development cycle of the linux kernel and also the distributions. There are some examples where this model is applied successfully, Intel comes to mind. However this is the exception and one can not expect every other company in the world to be able to follow the afformentioned kernel and distribution schedules, because of countless reasons. I am amazed how educated people can not grasp / understand / accept this common sense.

To the extent that issues like this are perpetuated, but also others like never having to deal with terminals and everything not having a GUI, linux will not have a chance in the mainstream. Sadly.

[ernstp]

agd5f Which one is Sienna and Navy now exactly?

[pal666]

CUDA and proprietary software for professional work is a perfectly valid magical quality that NVIDIA has over AMD.

subj are gaming videocards. for compute amd has crdna. i.e. it's a bit off-topic. cuda is vendor lock-in, i.e. you have to be not very smart to depend on it for starters. amd has converter from cuda to their chips, i don't watch it, so i can't tell you how good it is

It's considerably more niche I guess, but perfectly valid reason to not buy an AMD GPU.

well, i'll file cuda reasoning under "nvidiot"

I would love the new AMD cards tbh, but if they're incompatible with industry software

they are good enough for supercomputers, so again i'll file this complaint under "nvidiot"

[oiaohm]

The open source driver delivery to the end users is fundamentally broken. By end users I always mean consumers, common people that want nothing to do with terminals, source codes and third party software. It is fundamentally broken because it relies on provisioning and syncing hardware releases with the development cycle of the linux kernel and also the distributions. There are some examples where this model is applied successfully, Intel comes to mind. However this is the exception and one can not expect every other company in the world to be able to follow the afformentioned kernel and distribution schedules, because of countless reasons. I am amazed how educated people can not grasp / understand / accept this common sense.

https://www.osr.com/blog/2020/10/15/microsoft-driver-updates-allowed-win7-win8/

Sorry but driver delivery open source or closed source is not that simple. Microsoft will in future be mandating all drivers for Windows be submitted to them for approval as well.

The reality as you attempt to make sure the drivers the OS is using are quality the process for those making drivers get more of pain in ass. At least the Linux kernel mainlining process you are dealing with humans who can in fact think not a automated suite that says your driver is bad so you are out absolutely that is the current Microsoft model.

https://github.com/dell/dkms

DKMS has existed for a very long time for tell for allowing drivers for Linux to be installed on many kernels. Spectre bugs have also made it clear what was theory that binary drivers were a security nightmare turned out to be true.

The problem is a lot more complex than one likes.

1) You need to remove or redesign structures in kernel space at times because they are security flawed. This will break binary drivers this did happen with Windows with the meltdown and spectre class bugs and has happened with different Windows 10 updates since. So Linux kernel developers being worried that drivers not matched with the kernel were going to cause problems when security fixes had to be done has been proven true.

2) Drivers need to be peer reviewed. Reason why Microsoft is being forced into Mandating validation. Linux kernel validation process is submit you driver for mainline.

Something people miss is if you get your driver into mainline and you have a product that need that driver you can get the driver mainline backported into the LTS branches that distributions do auto pickup.

zoomblab its really simple to say make distribution of drivers work well. Its turns out to be a really hard problem for everyone. Every model has it problems. Every model will required companies to be putting their drivers in early for approval before the hardware is on the market in future. So this will mean working with like Google for Andriod, Microsoft for Windows, and Linux kernel mainline for Linux. Yes working with those upstreams will mean being aligned with their cycles.

[piotrj3]

novideo has cheating tdp, amd has total board powerin reality not all pieces of silicon are the same
it's called binning. i wonder how nvidiots compute novideo top prices

Normally I wouldn't offend people but screw it if you can I should be able to do it as well.
Listen absolute piece of shit moron. I am not talking in those posts(with compute units) about RTX 3090 or Nvidia TDP or whatever you illiterate person. I am comparing 6800XT and 6900XT. If you want to talk about RTX 3090 go somewhere else. Perhaps to primary school where they teach reading with understanding, but I bet with your behaviour no one teached you ever you would not pass any classes. Also factually TDP is not information for you how much GPU consumes power but it is around value of heat produced for sake of orientation of buying coolers. Also just so you know AMD "cheats" TDP way more. Comparing to power draw like 2070 super FE (official 215W TDP) averages 214W and peak 226W. AMD 5700XT(225TDP) in same furmark does 223W but peaks 263W. I don't see who is cheating more, only one who really cheats TDP is intel but Intel strictly claims diffrent definiton of TDP.

Also that is really intesting binning that seems to be first binning in the world on mass scale to increase slighty better binned product by more then 50% of price. Because all binnings I am aware of among Intel, AMD, Nvidia resolve usually around 2 products from same chip where one is only slighty better binned and slighty more performance. See Ryzen 3000 and 3000X (and now XT serie) see Intel 10850k and 10900k, see Nvidia normal and super lineups. Binning that costs more then 50% more for probably 10% performance (practicly less) without more VRAM or wider memory bus or something is horrible trade and somehow this is first time I see it (except maybe limited 8086k that was for hardcore overclockers).

[smitty3268]

Summary:
AMD has cards with approximately the same performance as the 3090, 3080, and 3070 TI available for a bit cheaper, a bit less power usage, and a bit more memory. Ray tracing is slower and a DLSS like option (Super Resolution?) is "coming in a driver update"....

Seems competitive, at least. Hopefully they can keep the cards in stock unlike NVidia so far with their launch, and hopefully it actually works on linux this time right from the start.