AMD Radeon RX 7900 XTX + RX 7900 XT Linux Support & Performance

[L_A_G]

I was hoping to see my 3090 to get a bit of a hiding and outside of heavily ray tracing focused workloads it seems to have gotten one. Good on AMD.

While that may have been a bit of an odd thing to say, I always like to see real competition in any market and the significant tit-for-tat improvements that follow when its working. Which it seems to be doing considering the good performance and superior value offered by these new cards. However the fly in the ointment is that AMD's being AMD again in how long it takes for full (mostly bug/issue free) support to materialize in the open source drivers. Then again Nvidia's way, way worse in that regard so you can excuse it if you're being realistic.

[finalzone]

Hmm, about the 6900 series, I may test Unigine benchmark and Shadow of Tomb Raider with my system: AMD Ryzen 9 5900X 12-Core testing with a ASUS PRIME X570-PRO (4403 BIOS) and Sapphire AMD Radeon RX 6950 XT 16GB on Fedora Linux 37.

[NeoMorpheus]

We'll see if NVIDIA now decides to send out the RTX 40 series hardware...

1- If you didnt blindly worship them somewhere, then dont hold your breath. Seems that they have doubled down on that, since even Steve for HU has becomes a nvidia shill with their bullshit RT propaganda push.

2- If they do send something, hopefully they also send you a fire extinguisher.

Thanks for the review!

[dimko]

• According to Steam HW Survey fewer than 1% (minus Steam Deck) of people game under Linux - maybe NVIDIA finally started paying attention to the figures.
• NVIDIA is universally hated and mocked here on Phoronix - no cards, less hatred.
• Why do you care about NVIDIA results anyways? I've been under the impression that Linux user and NVIDIA do not mesh.

Historically they did a lot.
And for simple reason. Nvidia had and still has superior binary driver. Before AMD was even on the scene for Linux. I remeber gaming on Linux 10+ years ago on Nvidia. AMD was not even something I would remotely think about.

[avis]

Historically they did a lot.
And for simple reason. Nvidia had and still has superior binary driver. Before AMD was even on the scene for Linux. I remeber gaming on Linux 10+ years ago on Nvidia. AMD was not even something I would remotely think about.

No one disputes that NVIDIA has high quality closed drivers but that's not what Linux users believe they want.

And there's a huge stinking issue of UEFI Secure Boot. You either have it or you have NVIDIA drivers (there's an option of your own MOK certificate but it's not for 99.9% of users out there). Neither RedHat nor Ubuntu want to digitally sign NVIDIA drivers even they perfectly can. What am I saying, they are even refusing to sign VirtualBox GPL 2.0 kernel drivers.

[Berniyh]

Beware of the multimonitor idle power consumption: I hope it's an issue which can be remedied in software.

Would be interested in that as well. Especially the power consumption with multi-screen setups in idle and video play (e.g. Youtube) seem to be high.
This was an issue with AMD cards for a long time, but they got it under control with RDNA2. However, RDNA3 seems to be a huge step backwards there, most likely due to the interconnects and – in case of playing a video and multi-screen setups – having to fire up more of them to get higher memory bandwidth.
Would be nice if Michael could check for this. Would be interesting if the problem on Linux is the same as on Windows.

[AdrianBc]

Phoronix: AMD Radeon RX 7900 XTX + RX 7900 XT Linux Support & Performance

Today's the day that the embargo expires on being able to provide reviews on the AMD Radeon RX 7900 series graphics cards... After testing both the Radeon RX 7900 XT and RX 7900 XTX graphics cards the past two weeks, today I have the initial performance numbers to share on these graphics cards and the current state of the open-source Linux graphics driver for these first RDNA3 graphics cards. Here is the first look at AMD's new flagship desktop Radeon graphics cards running under Linux with fully upstream and open-source graphics drivers.

https://www.phoronix.com/review/rx79...x7900xtx-linux

Thanks for these results, but what I really want to see is a simple benchmark showing the computational speed of RX 7000 for both FP32 and FP64.

This would solve the mystery of the fate of FP64 in RX 7000, i.e. whether FP64 has been completely removed or the FP64:FP32 throughput ratio has been reduced to 1:32 or the FP64:FP32 throughput ratio has been reduced to 1:64.

Because AMD has not said a single word about FP64, it might have been completely removed, but prior to the official launch there were rumors that the FP64:FP32 throughput ratio has been reduced to values similar to that of the NVIDIA GPUs.

[frytaped]

I hope someone sends Michael something that's hot, thick and massive.

[jayaura]

Beware of the multimonitor idle power consumption: I hope it's an issue which can be remedied in software.

Fix for this was merged upstream, but has not landed in stable release yet. Next release should have the patch.

[Grinness]

Would be interested in that as well. Especially the power consumption with multi-screen setups in idle and video play (e.g. Youtube) seem to be high.
This was an issue with AMD cards for a long time, but they got it under control with RDNA2. However, RDNA3 seems to be a huge step backwards there, most likely due to the interconnects and – in case of playing a video and multi-screen setups – having to fire up more of them to get higher memory bandwidth.
Would be nice if Michael could check for this. Would be interesting if the problem on Linux is the same as on Windows.

Those are numbers under windows.
In linux with open source drivers the story is different -- at least measured with sensors (via amdgpu kernel driver):

GPU	Single Monitor	Dual Monitor
rx 480	7W	10W
rx 6800	7-11 W	8-14 W

The article mentions 41 W in multi-monitor for rx 6800, not even close