GT218, Nvidia 340 failure is back on Quadro NVS 300

[illwieckz]

We hardcoded some string detection to fix the same issue with the Geforce 210 (See Unvanquished/Unvanquished#919), but the driver string detection does not match the Quadro NVS 300 and then the bug is reproduceable on current 0.52.0/sync branch.

We currently test for the Geforce string to apply the workaround and then NVS does not match.

I noticed both GPU are using the 340 version of the driver, it may be better to look for it instead.

[illwieckz]

While waiting for a fix, players facing the issue can workaround it by running the game that way:

./daemon -set r_arb_texture_gather 0

[illwieckz]

freem owns two GPUs running with the nvidia 340 driver:

• GeForce 8400 GS rev.2 (NV98, G98) from year 2007,
• GeForce 8400 GS rev.3 (NVA8, GT218) from year 2010;

Only the GT218 based one reproduces the bug, while the G98 does not reproduce the bug with the same driver (same computer, same system, cards were swapped out).

Because we can assume all GT218 use the 340 driver, we may detect the 340 driver to apply the patch, but it will apply the patch for nothing on some hardware (preventing the use of some optimization). It would still be better to detect all Geforce and Quadro cards…

[illwieckz]

The G98-based GPU is also driven by the Nvidia 340 drivers but does not support ARB_texture_gather, so disabling this extension for all GPUs using this driver will not degrade experience for this card (this extension would be disabled because it's missing anyway).

It's very likely the condition to reproduce the bug is not the GT218 GPU itself, but the Nvidia 340 driver advertising the ARB_texture_gather extension.

We have to make sure the bug does not affect newer driver that supports ARB_texture_gather, I'll test a GPU using the Nvidia 390 driver soon. But well, I have a feeling we would have known about it before…

[illwieckz]

Fixed by #370.