Valve Has Been Developing A New Mesa Vulkan Shader Compiler For Radeon

[dimko]

shame on AMD. Valve does their job as far as I can judge.

[shmerl]

Some results for TW3 here: https://www.gamingonlinux.com/forum/.../post_id=23861

[entropy]

shame on AMD. Valve does their job as far as I can judge.

LOL. Wouldn't even be possible for Nvidia hw.

[thebishop]

shame on AMD. Valve does their job as far as I can judge.

Whoever is the job, at least AMD isn't standing in the way.

Also, the fact that Valve *can* do it shows why their work on Linux is so important. If Valve built their future ambitions on a Windows base, they would be far more restricted in what aspects of the system stack and the user experience could be directly improved. Making Linux a viable base for high-end gaming paves the way for a ton of exciting innovations in how we game. Google Stadia is already an example of this. I'm also really excited about the potential for handheld systems based on x86 and SteamOS. I think it's a near-guarantee that Valve will eventually launch an end-to-end hardware platform for VR, which would also be extremely difficult to do well relying on Windows.

[AndyChow]

I really hope we're getting close to a blowout re-introduction of SteamOS. So many improvements have been made at all levels of the Steam for Linux experience, but SteamOS (the publicly available version anyway) has gone stagnant, without integrating most of the recent work.

We're in a window of maybe 6-12 months where a new push for a Steam linux set-top box could be a viable alternative to next-gen consoles. Next year's big games like Cyberpunk 2077 are essentially going to be a next-gen experience on PC. I think Google's Stadia is going to capitalize on their head start to market, and I think Valve also has an opportunity.

The way the current market is, I think that consoles have their places because they are cheap and have exclusive titles. You pay less for the hardware, but they make revenue from the "DRM" factor (in their minds at least, I doubt it it really affects actual sales). Then you have people that have powerful PCs. There doesn't seem to be much demand for a set-top gaming box which does not have exclusive, and is expensive because it's not sold at a discount (the set-top box makers won't get revenue from games sold).

As for Stadia, according to some "as 4K HDR streaming requires a minimum of 35Mbps to stream, a hour's worth of play will take up more than 15GB of data". For those that have gigabit connections, that won't be an issue. But maintaining low latency on 35Mbps connection for every frame for an hour isn't currently available for most. I bet the lag will be there. Even the Steam Link was criticized for high latency, and that was on a local network.

I just don't see a market for a set-top dedicated box which isn't also used as a pc. I don't really think either SteamOS or Stadia will have much of a future.

[Snaipersky]

What's this about an AMD path for Doom?

[dimko]

LOL. Wouldn't even be possible for Nvidia hw.

Dont get me wrong I am salary or two away from dumping 780TI. That does not excuse fact, that 3rd party does AMD's job.

[Haxk20]

Im getting consistent 2-5 fps boost in most games. This is really impressive job they did.

[stormcrow]

Dont get me wrong I am salary or two away from dumping 780TI. That does not excuse fact, that 3rd party does AMD's job.

How is Valve doing AMD's job here? AMD has company supported shader compilers available for their hardware already. Valve is doing it because they can do it and think they may be able to do a better job for their particular purpose, whatever purpose that might be, not because they have to do it.

[ms178]

Dont get me wrong I am salary or two away from dumping 780TI. That does not excuse fact, that 3rd party does AMD's job.

Think of it this way, LLVM within AMD's stack has to serve several competing goals, gaming is just one of them, e.g. their ROCm stack for compute, and there were also other fundamentals mentioned within LLVM itself. Also it is Valve's money, and they must see a business case here to put all that investment into this compiler (some NIR related changes are general enough that other vendors also profit). On Linux and with the open documentation from AMD, that is a good example where the community sees a need for something else, so they just do it. It would have been great if they had fixed these issues in LLVM instead, but they saw it as the lesser burden to do it their way. If it leads to a better open source graphics stack for AMD on Linux gaming, I am all for it.