Thread 106232752

7900 XTX is pretty kino on Linux. Why do AMD GPUs get so much hate?

>>106232752 (OP)
It is as old as the internet, newfriend.

>>106232752 (OP)
Nvidia has a virtual monopoly and her fanboys want to keep it that way.

>>106232752 (OP)
windows jeets because AMD can't make stable drivers. And before anyone says they do now they've been saying that for the last 15 years and its never been true

>>106232752 (OP)
nvidia mindshare is real
>106232815
>her-baiting with a company
plz just shush T_T

>>106232752 (OP)
no it doesnt , its just shills

>>106232752 (OP)
How is that HDMI2.1 working for you?

>>106233133
>HDMI2.1
>2025
No serious person would prefer HDMI over DP.
Also it's not AMD fault but HDMI forum, Nvidia won't also have an open-source driver.

>>106233188
Most modern TVs only have HDMI 2.1, also DP 1.4 requires DSC for high res/refresh rate and DP 1.4 cables all suck (good like finding a monitor that doesn't use DP 1.4.) Ive had included monitor DP cables artifact at high refresh rates while all my 2.1 HDMI cables are fine. DP 2.1 seems great but good like finding monitors that have it

>>106233206
Using a TV as a display is a compromise, you are getting an inferior display. If you reach the needs where you MUST use 2.1, you should use a proper monitor.
I wish it was available, but, honestly, when using Linux we normally have to buy things after researching. AMD tried, they even had the code, the problem is bureaucracy and licensing.
On the cabling part, that is hit and miss unfortunately because we can't trust manufacturers. I use DP2.1 with two monitors and never had an issue.

I was hoping for an RX 9050, but instead they release an RX 7400. I just want something with <100W TDP from factory.

>>106233351
What exactly is the issue with undervolting and powerlimiting

>>106233133
i just use a dp to hdmi adapter. full 4k 120hz hdr vrr on linux.

>>106232752 (OP)
Explain how a GPU can be "kino" on a certain operating system, nigger

>>106232752 (OP)
their drivers are trash

>t. 6750XT

>>106232752 (OP)
i got one like 6 months agop as an upgrade to my 6950xt good card sucks theyre dropping out of high end cards now though this might be the last good card for a long time

>>106233373
Just that it's not the default when booting, but I've been looking at that too. The RX 9060 are quite a good value proposition and Linux support is pretty good.

mfw this is the 3rd time Radeon had a 7000 series

>>106233606
They're not dropping out
The 9070xt was just a low effort filler card because they don't want to make rdna any more
There's an argument to be made that with the reunification of architectures you'll see AMD going for flagship status because they're no longer wasting wafers on a gamer only market

>>106233477
>HDR
>on Linux
Yeah, sure

>>106232935
the drivers have been fine for years now, cope retard, most of the driver problems were actually hardware based and they never admitted it.

>>106232752 (OP)
its great, i own one, its mostly indian shills, fanboy shills, and buyers remorse idiots who are coping hard. If you paypigged for a 5090 like a retard, you bet your arse you're going to shill it and mock the 'poors' with more sensible pcs, because thats what you overpaid for.

>>106233628
I'm a bit sad they changed the naming scheme to pander to nvidiots with the new 9000 series. I would've loved to have a new Radeon 9600 Pro, which I had back in the day. I've had a 7970 though, so a new 7000 series wouldn't be too shabby either.

>>106233484
By the virtue of not being dogshit like linux nvidia and not having to manually update the drivers most likely

>>106233844
It works on KDE
xfce and mate chuds won't ever get it with their one commit per year but still

>>106233844
Both Gnome, KDE and Hyprland have full HDR support. The merger started 5 years ago.

>>106233484
AMD and Intel have fully official open source drivers and are pretty much available as soon as they are out. Due to that, the support is not for recent boards but also very old ones.

>>106232752 (OP)
Because it doesn't run their preconfigured ai cp generators.
It's that shrimple.

>>106233628
>>106233934
Playing devil's advocate here, it makes sense, the numbers themselves are bullshit but AMD is barely a competitor so they have to make things as easy to understand as possible.
Now motherfuckign ROCm is the real problem.

>>106233373
>what is the issue with powerlimiting
idk, ask amd, on rdna 3/4 you can increase powerlimit by 12% or decrease it by 6% (yes, i didn't swap those around, you can increase it more than you can decrease it, extremely retarded)
you can underclock which achieves a similar result but it's not optimal as some workloads wil use much more power than others for similar clocks, so when you target a low power limit such as 75w you either lose a fuckton of performance in an attempt to stay under 75 all the time or you go over it a lot in some tasks.
I'm surprised there hasn't been more backlash about this, no one seems to give a fuck.

>>106232752 (OP)
I wish AMD GPUs had better path tracing performance. The 4080 Super has nearly twice the rendering performance in Blender as the 7900 XTX despite the two GPUs having the same MSRP.

I got a 7900 xtx on release and I haven’t had any issues, performance is great and I have no real need to switch. From what I heard and isn’t going to do flagships anymore so that’s a little depressing.

>>106235462
How much is the AMD's fault and how much is not enough work in Blender?

it was shit at launch because it was "world first chiplet gpu". OS didn't know what to do with it so 100w at idle and lower performance than 4080.

today the drivers\firmware are way better and now the gpu is on par or better than a 5080 in rasterization.

>>106234415
>>106235368
what about cinnamon?

>>106235462
How much of that is because of CUDA

>>106235548
>cinnamon
For what I saw Cinnamon use Muffin as a WM, so, no, not really supported. I don't know if you can use another WM with it though.

>>106235623
Thanks for responding anon

>>106232752 (OP)
I don't hate Radeon GPUs, it's just that local AI stuff is already hard enough on CUDA, why would I want to make it even harder by using ROCM?
Also, no chink hacked 2x memory Radeon GPUs, right?

>>106235711
Don't mention it. I do recommend giving KDE a try, we got 6.4 recently and Debian is shipping 6.3. It's a great experience and you run it live just to check the results with HDR in your setup.

>>106235845
>Also, no chink hacked 2x memory Radeon GPUs, right?
There has been mods on the RX5600XT to put 16GB on it.

AMD gave hints of improving the ROCm compatibility with retro versions but we all know how AMD is.

>>106235493
>>106235553
>How much is the AMD's fault and how much is not enough work in Blender?

I think it's probably a mixture of both hardware and software aspects. It's hardly disputable that RDNA has been lagging behind Nvidia at ray tracing on the hardware level, but there's also vendor APIs at work.

>How much of that is because of CUDA

You can use the pure CUDA option for rendering in Blender, but most Nvidia GPU users in Blender use Nvidia's OptiX ray tracing API which is faster as it directly utilizes the dedicated RTX cores to accelerate path tracing.

https://developer.nvidia.com/rtx/ray-tracing/optix

For a while, AMD didn't have an equivalent in Blender, using ROCm HIP compute for rendering instead, but they recently integrated HIP-RT in Blender which competes with OptiX, however it isn't as mature as OptiX.

https://gpuopen.com/hiprt/

Intel also has their own ray tracing API which Arc GPUs use called Embree, GPU utilization was integrated into Blender at the same time as HIP-RT.

https://www.embree.org/

>>106232752 (OP)
>crashes kernel

>>106233206
>Most modern TVs only have HDMI 2.1
don't use a TV then, tard monkey.

>>106235958
>For a while, AMD didn't have an equivalent in Blender, using ROCm HIP compute for rendering instead, but they recently integrated HIP-RT in Blender which competes with OptiX, however it isn't as mature as OptiX.
>
>https://gpuopen.com/hiprt/
Oh that's interesting, I didn't know about that.
Hilarious to see that the SDK documentation doesn't mention Navi3 though but the repository does. AMD never learns.

>>106232752 (OP)
I would like to see an FSR4 compatible GPU with at least 24GB VRAM.

>>106236007
Find me a 50" 4k monitor

>>106233934
>>106235399

Meanwhile Intel Arc does its own thing and the naming system is brilliant because using letters means you can do 26 GPU generations without rolling over

>>106232752 (OP)
There is only one white mans choice.

>>106232752 (OP)
most people don't even try aldi shit
little do they know some of it is pretty good