[vishnu]

...proper desolder braid cleaning.

When the situation comes to having to get the desolder braid out, you know you're going to have a bad day. Good luck with it though!

[Tech_and_Music]

...proper desolder braid cleaning.

When the situation comes to having to get the desolder braid out, you know you're going to have a bad day. Good luck with it though!

Eh, for BGA work, it's pretty much required to clear off the pads on the PCB, prior to soldering on the new chip, or to remove the remnants of the old solder balls off the IC prior to reballing it.
I just don't get why the solder mask on these PMC R7Ks is so weak, as the QED RM5271s don't have this issue.

[Tech_and_Music]

A friend of mine checked out the RM7000A datasheet, and pointed out to me that the datasheet actually describes the bitstream data, as it's the CPU reading it, rather than the SGI itself, unlike what I misunderstood.

   

This is very valuable information, as it frees me from having to stick to (close to) 600MHz rated chips.
Sure, having the fastest of the fastest is nice, but any R7K is going to be a nice upgrade for R5K based O2s.

I do note that it does not tell the CPU what the cache size is.
Knowing it has a couple of index lines and control signals going to the cache, with the actual data traveling on the SysAD bus, this does once again make me wonder about the possibility of putting larger SRAM chips in place, to see if it can enumerate more cache, opening another avenue to SGI O2 upgrades.

Another thing I realize, is that due to the PROM being connected to the CPU, a drop-in PGA module for the older R5K modules could potentially provide a fully solder-less install for an end user, by putting the R7K BGA CPU, voltage regulator and PROM on an interposer module that goes into the R5K's PGA socket.
If it indeed then just checks the cache size by polling the SRAM ICs themselves, it would just use the 512K already on the board with no problem.

Again, all of that is theoretical for the moment, but if I can get myself CAD-savvy enough, open source interposer modules would be my end goal.
That would potentially make R7K upgrades much more accessible to O2 owners.

Enough to think about for a while, I think :-)

[chulofiasco]

awesome!

[RageX]

I'd be very interested in a drop in upgrade.

[Tech_and_Music]

Absolutely amazing work, also consider using 304 pin BGA sockets - they are available.

Then once you get a working CPU module you could use that known good working CPU module to test other socketed CPU - it would simplify making the transposer board also, which would be impressive!

Do you have a link for such a socket?
I do see cheap ones available on eBay, but they're a two piece deal, where you still have to solder the BGA chip to an adapter, that then plugs into the socket that would be soldered to the PCB.

That would still require multiple reflows per chip, so not necessarily easier than just soldering the chip down and trying it that way.
There may be sockets with mechanisms to lock down the chip, requiring no soldering to the chip for testing purposes, but I know those are generally extremely expensive.