Hi PaQ, thanks for collecting that data!


> (does the result make any sence for you Michael Oo)

Sort of... That data does seem to suggest that maybe it's better to use only half as many threads on a hyper-threading machine. Seems to be the opposite conclusion from some of Burr's previous data though.

I guess it's fairly likely that the optimal settings for different things (there are other internal variable factors too like number of buffers used per thread and vertex/index buffer size) is different on different machines depending on all kinds of factors like bus speed and video card speed, etc... and also probably even different on the same machine with different model structures as well.

- Michael