Valve's ACO Helps The Radeon RX 5600 XT Compete With NVIDIA's RTX 2060

[Guest]

How is the current stability for AMD drivers? I'm seeing lots of posts on social media about driver issues and BSODs on windows.

I had a RX 560 and 580 that both had an issue trying to handle 4K@60Hz over HDMI on a display I had; would randomly become unstable and either garble or go to black the entire screen at times. OS didn't matter (happened on Windows, macOS, and Linux with the same GPU and display), and it wasn't the cables (tried numerous 4K@60Hz HDR certified cables). Some further searching around shows some threads pointing at it being a hardware-related issue specific to AMD GPUs. I could fix this by creating a custom reduced-blank resolution, but this was messy (not ideal with Wayland, and required a paid program on macOS).

Got a laptop with a GTX 1060, and the display with the same cables works perfectly at 4K@60Hz; no instability at all.

[LoneVVolf]

How is the current stability for AMD drivers? I'm seeing lots of posts on social media about driver issues and BSODs on windows.

Judging by posts on archlinux forums things for the RX 5000 series cards have stabilised a lot.
Especially for those willing to run mesa / llvm git versions .

The APU stability ... varies.

An important factor seems to be motherboard manufacturer and the quality/frequency of firmware releases.
One or 2 manufacturers appear to have a worse track record (more problems) then others .

Those with motherboards/chipsets designed for the new models seem to get the best results.

[Teggs]

These are impressive numbers. Compliments to all the developers.

I understand ACO will next be incorporated into RadeonSI, but doesn't this have wider implications? AMD would want this on Windows, Playstation, and what remains of OpenGL on Mac, at least for gaming? And what do nVidia, Intel, and Apple think about it?

I assume that Linux Navi users are beta testers for the PS5 driver code and Windows Navi users are beta testing for the...whatever the fuck Microsoft decided to name their next Xbox...driver code.

And all beta-testing the Navi achitecture in general for RDNA 2 console hardware. From a certain point of view. At least the desktop segment still matters to them in that scenario, as overachieving rodents if nothing else.

[RussianNeuroMancer]

I had a RX 560 and 580 that both had an issue trying to handle 4K@60Hz over HDMI on a display I had; would randomly become unstable and either garble or go to black the entire screen at times. OS didn't matter (happened on Windows, macOS, and Linux with the same GPU and display), and it wasn't the cables (tried numerous 4K@60Hz HDR certified cables). Some further searching around shows some threads pointing at it being a hardware-related issue specific to AMD GPUs. I could fix this by creating a custom reduced-blank resolution, but this was messy (not ideal with Wayland, and required a paid program on macOS).

Got a laptop with a GTX 1060, and the display with the same cables works perfectly at 4K@60Hz; no instability at all.

Did you tried to contact AMD support regarding this issue? Is this issue HDMI-specific? (I asking if you can check it with DisplayPort outputs and DisplayPort display.)

[geearf]

These are impressive numbers. Compliments to all the developers.

I understand ACO will next be incorporated into RadeonSI, but doesn't this have wider implications? AMD would want this on Windows, Playstation, and what remains of OpenGL on Mac, at least for gaming? And what do nVidia, Intel, and Apple think about it?



And all beta-testing the Navi achitecture in general for RDNA 2 console hardware. From a certain point of view. At least the desktop segment still matters to them in that scenario, as overachieving rodents if nothing else.

It's AMD only, why would Intel or nVidia think anything about this?
I doubt the current PlayStation needs it, as for Windows well ACO so far only supports part of Vulkan shaders, but AMD probably does not want to keep supporting their other backend compiler for OpenCL and co.

[Zgembo]

Wow, I see 14% performance up in SotTR (both high and highest bench profiles) on my 5700 (fanless, limited to 110W)!

[theriddick]

I wonder how many people having AMD driver issues are also using wayland? Wayland is still pretty alpha version territory!

AMD need to start up their open-source driver branch for windows, that be great and help bring features to Linux down the line.

[Teggs]

It's AMD only, why would Intel or nVidia think anything about this?

Because of the reason ACO was created. Doesn't nVidia use LLVM as well for shaders? Wouldn't it behoove them to do the same thing?

[geearf]

Because of the reason ACO was created. Doesn't nVidia use LLVM as well for shaders? Wouldn't it behoove them to do the same thing?

It does not make sense, ACO was created specifically for AMD cards (ACO => Amd COmpiler).
You want to support a vastly different set of GPUs, you'd most likely be looking at restarting close to from scratch.

I think nVidia uses LLVM but not fully like AMD does, and maybe their backend is fast enough anyway.
It's not because ACO is faster that it proves that using LLVM was wrong, it might simply be a less efficient implementation, etc.

[khnazile]

5600XT looks nice, but the performance of older Vega chips... that's just disappointing Also, there's no Vega20 in this benchmark, was it too bad?