| Bug ID | 85580 |
|---|---|
| Summary | [RadeonSI] Bad performance in TF2. |
| Product | Mesa |
| Version | git |
| Hardware | Other |
| OS | All |
| Status | NEW |
| Severity | normal |
| Priority | medium |
| Component | Drivers/Gallium/radeonsi |
| Assignee | dri-devel@lists.freedesktop.org |
| Reporter | aaronbottegal@gmail.com |
I built Mesa just a few days ago with LLVM 3.5, git mesa, 3.18-rc2 kernel. Performance is just horrible with tons of dips when playing TF2, and probably other Valve games. Polygon count just makes it too heavy to what it used to be, dipping down to 20FPS or so when moving objects and such. I don't have time to bisect, don't even know how much longer I will be using this R9 270X, so any tests I can do without a recompile I can try. Benchmarks like glmark2 are fine, but in-game performance is just bad.
(In reply to Aaron B from comment #0) > Polygon count just makes it too heavy to what it= used to be, dipping down to > 20FPS or so when moving objects and such. Have you double-checked that direct rendering and hardware acceleration are working for 32-bit processes? Did you build 32-bit Mesa as well? > I don't have time to bisect, [...] So it's a regression? In what kind of time-frame?
It has gotten better since I've re-compiled llvm and mesa, and moved back to just 3.5. But it still drops down to the red in the source engine FPS counter a lot still. It should be a regression, afaict, it never did it as often as it does with llvm 3.5, current mesa, radeonsi.
| What | Removed | Added |
|---|---|---|
| Status | NEW | RESOLVED |
| Resolution | --- | WORKSFORME |
Probably fixed eventually, never tested but going to just close assuming it's fixed since it's the biggest linux game really.