Linux Support Expectations For The AMD Radeon RX 6000 Series

[skeevy420]

Anyone else waiting on, I'm gonna call it, a mid-range 6-8GB 6400XT in the $175-$250 range? I like that they're taking on Nvidia's best, but all I really need is something equivalent to my 580 with better power usage and thermals and the 5300XT just isn't it.

[smitty3268]

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> )

No one has AV1 hardware encode, and everybody now has all the rest.

Honestly if you are mainly concerned with video I'd say just stick with NVidia. AMD has the same stuff on paper now, but in practice I'd argue it's not really the strong point of their hardware and NVidia has focused more on the software side of it so far.

[oiaohm]

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"....

Missing DLSS for Linux users may not be a deal breaker. https://github.com/Plagman/gamescope Remember AMD linux users have working XWayland that equals having gamescope. Now of course this might at long last get NVidia off ass and fix their Xwayland support so using Xwayland with a Nvidia card ceases to be return to software rendering on cpu.

[leipero]

I'm still on a GeForce GTX750 and I always spec my CPU to 65W TDP or less because I have no air conditioning. This is the main detail I care about aside from whether the drivers will match nVidia's for stability when a desktop session is left open for months on end.

This. I will never buy a CPU over 65W TDP, and for GPUs, I'm still debating it (with myself), there's no debate for above 75W TDP, if GPU requires extra power aside from PCIe I will not buy it, but I'm thinking 50W TDP would be reasonable upper limit. I do have AC at the place I am now, but, it is kinda wasteful to go over it, plus it corelates with my budget as well (lower TDP parts = lower/low end hardware = low price).
The only issue is, there's no decent 50-75W GPU on the market that is worth buying (I feel for 2020, 40GPixel/s 80+GTexel/s with about 3-4+ GFLOPs at around 50W TDP should be a standard for low end GPUs).
For about 80$ (or at least under 100$), such GPUs would be seeling like crazy IMO, for all sorts of purposes and low-end "gaming". I'm not sure if it's feasable from thecnological/financial perspective, but if it isn't in 2020, that would suggest that technology didn't really advance that much in the last decade or so, because those are similar specs to the high end GPUs from almost a decade ago.

[leipero]

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.

Honestly, AMD Image Sharpening (tm) is quite decent, I'm sure it's nowhere near DLSS (in function and practise), but doing some sort of upscaling that is decent enough shouldn't be that dificult IMO.

But yeah, AMD is spreading like COVID in the US . (Before anyone gets offended, US numbers are far away from what media makes it look like, the matter of fact is, it's bad everywhere, and US for sure isn't even close to be "the worst" as media wants to present it).

[polarathene]

"nvidiot"

Seems you're just making a fool of yourself atm tbh.

I've even expressed interest in using AMD, just that it's not feasible commercially until the software vendors for the industry support it. Here's a popular open-source project with AMD feature request:

https://github.com/alicevision/meshroom/issues/595

subj are gaming videocards.

You're comparing to NVIDIA 3080/3090, the GTX/RTX line isn't only for gaming, it's useful for industries like photogrammetry too.

You're thinking of workstation Quadro's I guess? The extra VRAM is nice, but other benefits aren't as important, mostly just need solid CUDA performance with enough VRAM for the given dataset to process.

cuda is vendor lock-in, i.e. you have to be not very smart to depend on it for starters.

That's a non-issue when you need to make money and the competition cannot practically compete in said industry. If NVIDIA gets the job done and AMD GPUs aren't supported by the proprietary vendors (lock-in already), you're not going to get very far with AMD. ROCm has been available for a while, but not even open-source alternatives are adopting it yet, and those even with CUDA support are considerably behind in what the proprietary software offers (unlike other software you might be familiar with).

amd has converter from cuda to their chips

Yeah, that doesn't work with proprietary software. You need to have the CUDA source code. For the open-source alternative projects, it can help port the code to support AMD, but it's not a magic automated process, you can still run into road blocks with unsupported API

https://github.com/ROCm-Developer-Tools/HIP/blob/master/docs/markdown/hip_faq.md#is-hip-a-drop-in-replacement-for-cuda

https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-porting-guide.html#scanning-existing-cuda-code-to-scope-the-porting-effort

https://github.com/ROCm-Developer-Tools/HIP/blob/master/docs/markdown/CUDA_Runtime_API_functions_supported_by_HIP.md

https://github.com/ROCm-Developer-Tools/HIP/blob/master/docs/markdown/CUDNN_API_supported_by_HIP.md

Here the Meshroom project had trouble, but has since improved trying to gain HIP support, however the manual part is what's holding it back for over 9 months apparently:

https://github.com/alicevision/AliceVision/issues/439#issuecomment-417422887

If it's so easy though, why not contribute to the project to promote AMD some more and help others not be stuck with CUDA at least in the open-source photogrammetry world

Or are you just going to deflect with some BS and name calling again?

[Alexmitter]

But yeah, AMD is spreading like COVID in the US . (Before anyone gets offended, US numbers are far away from what media makes it look like, the matter of fact is, it's bad everywhere, and US for sure isn't even close to be "the worst" as media wants to present it).

Ah yes, this bullshit again.

[mibo]

From: https://www.anandtech.com/show/16202...ovember-18th/2

"As things currently stand super Resolution is still under development, so it won’t be available to game developers (or gamers) at the time of the RX 6000 series launch. But once it is, like AMD’s other FidelityFX graphics libraries, it will be released as an open source project on GPUOpen, and AMD is explicitly noting that it’s being designed as a cross-platform solution."

DLSS -> Super Resolution: open source and cross-platform.
That sounds good to me.

[leipero]

Ah yes, this bullshit again.

What BS? It was a joke.

[Mez']

Anything foreseen around 150-200€ (or $)? You know, for the vast majority of normal users (excluding those satisfied with an iGPU)...