AMD Catalyst Linux Driver Performs Differently Based on Program's Name

Gaming benchmarks by preforming better in an older game? Yeah, no :)

A lot of the big games are often optimized on the driver level by both Nvidia and AMD which is why new graphics drivers are released very often compared to other hardware drivers. It's very common to see driver releases the same week as new triple A title that improve performance 40-50% and fix various broken things such as multi-gpu scaling.

In this case it's two games with the same engine (source 1) where the same optimizations happens to work but AMD likely hasn't gotten around to actually verify the optimizations and as such haven't enabled them for CS:GO (Remember steam hardware survey says only ~1% game on Linux so it's hardly worth their time to spend time optimizing Linux gaming)

No, this is just how modern game drivers work nowadays [1]. With older APIs (pre-Mantle/Vulkan/DX12) it's really opaque how to produce correct and performant code. And even past that, AAA game devs have enough clout to basically force hardware companies to fix their bugs, eg one major game never made BeginFrame/EndFrame calls which is a horrendous violation of API rules. You don't want to have "broken drivers" on launch day while it works on your competitor's hardware.

This results in a game of pattycake where the game developers and the driver developers are both guessing at what the other will do, and trying to optimize for it. So game releases basically require having an NVIDIA/AMD developer sit down, profile your code, and figure out which of the tens-to-hundreds of thousands of behavior triggers and optimizations should be turned on for your game.

What's probably going on here is that CS:GO is having a different set of optimizations applied from its parent game (or hasn't had any at all applied yet on Linux). The big question here is whether CS:GO actually works 100% reliably on those settings, on all AMD GPUs, on all possible combinations of graphics settings. Because it's very possible that it doesn't, or that no one has verified that it does.

[1] http://www.gamedev.net/topic/666419-what-are-your-opinions-o...

Only thing I recall similar to this was Intel's compiler not using SSE instructions intentionally on AMD processors. Or were you thinking of something else GPU related?

If by "a few years back" you mean 14 years ago with the quake/quack thing.

But nobody is saying this is the cheating-style "optimizing" which has been judged as invalid. Both Nvidia and ATI do heavy amounts of game-specific optimizations in drivers and so long as there is no visual difference this is perceived to be fair and valid optimization.