just wondering, Bards Tale

[FlatOut]

I didn't get more than a few extra fps in Baldur's Gate. Does this hack work best with Nvidia?

[Dreadmoth]

Hum, you could also try to set Blending accuracy set to the first level (basic). It could be faster on some games.

Basic Blending Unit Accuracy

Unfortunately Blending Unit Accuracy set to anything higher than None results in the game (or at least the textures) looking like it was rendered using a lower colour depth - I think this occurs on every Snowblind engine game that I have (and doesn't seem to be affected by "Allow 8-bit textures" setting).

Native resolution frame dumps:

None | Basic
Software mode

-----

I didn't get more than a few extra fps in Baldur's Gate. Does this hack work best with Nvidia?

Maybe it has a bigger impact in certain areas? The tavern is probably the most intensive area I've seen so far in BG: DA - particularly during the intro cutscene ("Elfsooooonnggg Taverrrrnnnnn" is what I used to hear when the game dropped to 25fps - now it runs full speed in every cutscene and area I've tried so far).

-----

simple enough and very good to know. ill have to do that then. thank you all very much. this has all been very helpful. and i assume this all applies to dark alliance and champions of norath as well?

It has worked well for me in all of the Dark Alliance and Champions games (in the areas I've tested, at least). Champions: Return to Arms in particular gets a nice boost to performance in the Plane of Tranquility (which used to run okay when standing still, but slowed down drastically when moving around or rotating the camera - now it is quite smooth).

[dogen]

Actually I would give it a try.
It appeared to have the famous half-screen bug and needs the sw mode as well as the 4GB patch. Very similar problems to the ones snowblind engine games showed. The half-screen bug in snowblind games was recently resolved.

Depending on your hw you could always play in sw mode but maybe it even works in hw.

That's because it uses the same engine.

[FlatOut]

Maybe it has a bigger impact in certain areas? The tavern is probably the most intensive area I've seen so far in BG: DA - particularly during the intro cutscene ("Elfsooooonnggg Taverrrrnnnnn" is what I used to hear when the game dropped to 25fps - now it runs full speed in every cutscene and area I've tried so far).

I tried an intensive section, that's where I got a few extra fps. In a non-intensive section I got 15 fps or so.

[Topken]

Yeah openGL works better on Nvidia hardware. Also whats your CPU FlatOut? Also check to see if anything is running in the background on windows that might cause you slowdowns.

[FlatOut]

Yeah openGL works better on Nvidia hardware. Also whats your CPU FlatOut? Also check to see if anything is running in the background on windows that might cause you slowdowns.

The question isn't if OpenGL works better on Nvidia, it's whether it should be holding back this hack so much because of it. This hack is supposed to give up to 50% performance increase, and I'm getting no more than 10%. CPU and background tasks aren't an issue.

[gregory]

I'm surprised of your result. I would have expected a bigger boost on AMD actually.
The hack has a double action
1/ it reduces texture cache management overhead. Less work on the GSdx thread. However it increases pcie transfers (instead to send some part of the texture, the full texture is uploaded). From the top of my head, it was a 20% increases in bandwidth.
2/ it limit the read of the rt to the useful part. Before you read the full 2k buffer now only small 64x64 block.

Hum could you check the impact with/without 8 bits texture.

[gregory]

Hum the blocking behavior is related to the color format of the game. 5 bits by channel. It would require some dithering to reducing the banding effect. BA none will emulate a 8 bits by channel (less accurate but nicer)

[FlatOut]

I'm surprised of your result. I would have expected a bigger boost on AMD actually.
The hack has a double action
1/ it reduces texture cache management overhead. Less work on the GSdx thread. However it increases pcie transfers (instead to send some part of the texture, the full texture is uploaded). From the top of my head, it was a 20% increases in bandwidth.
2/ it limit the read of the rt to the useful part. Before you read the full 2k buffer now only small 64x64 block.

Hum could you check the impact with/without 8 bits texture.

Without 8-bit I get poor performance, with 8-bits it runs fairly smooth, especially on D3D11.
I didn't get much of any boost without 8-bit, it just runs more stably at about the same framerate(1-2% more). With 8-bit on I get about 10% better performance with the hack.

[gregory]

Hum. Hum. I don't understand. Is your GPU PCI2 or PCI3?

I reduce the work on the GSdx thread. The speed boost on Nvidia imply that driver thread was idling (otherwise perf impact would have been null). I will try to bench it on a single thread.

Or maybe, texture upload is done in async fashion (even on AMD). Maybe AMD stall the draw to wait the texture whereas Nvidia stall the driver thread. I need to check also how many threads are enabled in MT mode.