A1200 DBOX Mediator Fast ata ETC

[MskoDestny]

doctorq wrote:
@Keropi

You have misunderstood something regarding the Mediator. Mediator can use the full gfx mem as well, as long as you don't use Warp3D. If you have no use for Warp3D you can use all the 16 MB of the ram without problems.

To be more specific, the Mediator only allows access to 8MB of PCI memory at a time due to the limitations of the trapdoor expansion slot. The Mediator hardware does provide banked access to the whole 4GB PCI address space (i.e. the driver has to write to a register to change the bank). The 2D graphics drivers have been written to work around this limitation, but the 3D drivers have not (and never will).

The G-Rex doesn't have this limitation as it doesn't use the trapdoor slot (it also has a lot more bandwidth as a result). Of course as others have pointed out, driver support isn't as good and some people have had trouble getting them to work with their Blizzards.

[MskoDestny]

tjaoz wrote:
@MskoDestny
More precisely speaking, it isn't the limitation of the trapdoor expansion slot, but the limitation of the A1200 turbo cards. Existing A1200 turbo cards have only 24 address lines connected to the trapdoor slot, instead of 32 address lines supported by this slot.

Well there are 8 reserved pins that could theoretically be used for the rest of the address bus, but generally speaking it's unwise to use reserved pins for anything so it's certainly arguable that it's a limitation of the trapdoor slot.

Warp3D drivers have access to the whole Voodoo graphic card resources - it means to the Voodoo registers and to 16MB (or 32MB in case of V4 and V5) of the Voodoo graphic memory. The limitation is that the Picasso96 2D graphic system can use only 8MB of the gfx card memory when Mediator works in the NO-MMU mode. The rest of the Voodoo card memory is then used by the Warp3D textures and the PCI DMA buffers.

I don't see how Warp3D can use more than 8MB of texture memory (apart from using the MMU). I suppose it could use 8MB for textures and the like and then instruct the Voodoo to render to a framebuffer in another 8MB page.

[MskoDestny]

tjaoz wrote:
These trapdoor pins (no. 1 to are not reserved for "anything".  :-D  They are reserved for A24 to A31 address lines (check the schematic of A1200 main board).

I stand correct there. I was going off of a pinout from BBoAH.

So, if the A24 to A31 address lines of the existing turbo cards are not connected to the trapdoor slot, it is certainly a limitation of the turbo cards, not the limitation of the A1200 trapdoor slot.

Well the A24 to A31 address lines aren't connected to anything on the A1200 so I would argue that it's the A1200 implementation of the A1200 trapdoor slot that is at fault even if the specification supports 32-bit addressing. If this were not the case, the turboboard would have likely allowed for 32-bit addressing.

You explained it yourself: "The Mediator hardware does provide banked access to the whole 4GB PCI address space (i.e. the driver (Warp3D driver in this case) has to write to a register to change the bank)".  :idea:

I was not aware that Hyperion had customized the Mediator Warp3D driver(s) to this degree.