besweeet said:
@boumbo: I thought that maybe your experiments would help us out with the HD 2000, but nothing.
As far as I can see there are three different cases of Sandy Bridge graphics:
HD3000 4 threads, 12 EUs
HD2000 4 threads, 6 EUs
HD - 2 threads, 6 EUs, GPU may work or not
http://en.wikipedia.org/wiki/Sandy_Brid ... processors
Not sure about the i7 models with 8 threads
As seen in the Intel Linux drivers the driver must distinguish between HD3000 and HD2000 for memory allocation and probably memory offsets. These depend on the number of threads per EU, and number of EUs.
The Mac driver does seem to know only about HD3000 and classifies also HD2000 and HD processors as HD3000. My best guess is that as the driver does not account for reduced EUs or threads we see strange artifacts.
The Linux drivers also disable completely graphics acceleration for generation/stepping/revision 6 Sandy Bridge chips. The reason is that the BLT ringbuffer is buggy on these chips and locks up. My humble guess is that all Sandy Bridge Pentium and Celeron are stepping 6 chips with buggy GPU, hence only "HD". At least this is the case with a Celeron B810 which I have sitting here.
Patching the Mac SNB driver for HD2000 should be possible IMVHO because the changes are relatively small - given enough time and a proper decompiler.
HD only Sandy Bridge chips (Pentium/Celeron) will probably never work satisfactory with the Sandy Bridge driver.
Cheers
Geejay