Today I probed a little deeper into this issue.
Some mainboard manufacturers have obviously choosen the strategy of "crippling a feature" to save on costs here by providing one single socket for TAG-RAM on HX boards. To be precisely what they really spared was the costs of a second 28-pin DIL socket.
As long as there´s just one TAG-RAM socket present, the cachable area of the L2 cache is limited to 64MB of RAM. The single socket could even be populated witch a 32x8 chip (instead of the needed 16Kx8) without fixing this, because this SRAM design has 8 I/O lines only resulting in an 8-bit TAG-RAM bus-width.
For the Intel HX northbidge to enter the enlarged cache mode that allows all 512MB of RAM to be cached by L2 cache there need to be 3 additional I/O lines for a total bus-width of 11 bit.
The attachment HX_cache_io.JPG is no longer available
This is done by adding a second TAG-RAM socket that is populated with a second 16x8 cache chip. Three of it´s I/O ports are used, 5 are left unused, ALL remaining pins like Vcc, Vss, address-lines, chip select, write enable and output enable are wired parallel to the existing (first) SRAM socket.
The 3 used I/Os need to be connected to the pins TIO8, TIO9 and TIO10 of the HX northbridge.
The attachment HX_pinout.JPG is no longer available
On first sight this looks like an easy job: Make a tiny daughterboard with one 28-pin plug at the downside and two 28-pin sockets on top, wire accordingly, populate with 2 x 16x8 15nS SRAMS and connect 3 wires for the additional I/O lines to the mainboard.
BIOS should take care of enabling a cachable area of 512MB at start with the correct register.
Maybe a pull-down resistor needs to be added at pin "TIO10" the turn the extended cacheability mode on.
Now here´s the situation on one of my HX boards:
The attachment HX_single_tag_1.JPG is no longer available
The problem is, if the pcb´s layout does not provide a second TAG-RAM, the needed pins of the HX-northbridge (that sadly comes in the BGA324 package) might not be connected at all. In my case I detected a trace for TIO10 (probably just in case TIO10 needs to be pulled up) but could not find any connections to TIO9 and TIO8 so far.
It´s possible that there are just isolated pads for the soldering balls.
There are countless small traces and vias on the green pcb of the chip where there´s no plastic coating but I have no idea where they lead...
Maybe someone has an idea how to get a grip on the 3 pins / pads?
Meanwhile I managed to find a connection with pin TIO9. No luck with TIO8 so far. I think I´ll give ot a try anyway and prepare a small daughterbord for two cachechips while searching for TIO8.