It only shows that the OS sees the installed dimm modules but it doesn't show if it is able to use it. The same thing happened to the early core 2 duo systems. The chipset had a memory allocation bug so Apple limited it to 3GB instead of 4. You could put 4GB in the machine and it showed up as 4GB in system profiler and about this Mac however only 3 GB could be used. The last reply from OWC is a link to a list with the testing results in various configurations. But those are just what they are: numbers. That's about all you can find at OWC about it. Mac Performance Guide also
questions it. The only source saying you can use 32GB memory is OWC and they not really have proven it to work.
They officially support up to 16 GB for those models so having 4 slots would make sense (4*4GB).
I realize what you're getting at here, but there's a serious flaw in the logic by DigiLloyd. Yes, the 24GB spec was determined from 6x DIMM slots, and Yes it was also via 4GB UDIMM's. Thus 24GB.
But that was done before 8GB DIMM's of any kind were out for DDR3 (Samsung's UDIMM's or any vendor making RDIMM's).
Now since the memory controller is actually the same between the SP and DP parts, RDIMM's actually do work in the SP versions of the Xeon line (35xx, 36xx, 55xx, and 56xx parts - all LGA1366 sockets and use the same memory controller - the difference is the 2nd QPI channel in the DP versions, which require a different chipset). The chipset does not matter in terms of memory capacity, as the memory controller is now on the CPU's die, not the chipset (memory controllers were located in the Northbridge in previous families, such as the parts used in the 2006 - 2008 systems, but changed in 2009 with the Nehalem architecture = memory controller on the CPU die). Systems engineering at it's finest (same memory controller on the LGA1366 parts, as ECC functionality is just disabled for the i7 variants).
This can be confirmed by looking to non Apple sources if it works or not (SP servers/workstations), and the information demonstrates that it does in terms of the memory controller within the CPU's (8GB UDIMM or RDIMM works on Nehalem architecture).
What I don't get, is why Intel hasn't updated that spec since RDIMM's have released to market.
In the case of the i7 variants on the LGA1366 parts, it's no longer correct either, but only if the Samsung parts are used for 8GB DIMM's (RDIMM is not supported; UDIMM can work, but the ECC functionality won't, so it operates as standard non-ECC DDR3).
@nanofrog: thought you meant the dimms used in the 2006 models but you're correct about registered memory. I took that for granted since unregistered ecc is nearly impossible to find. I never doubted the 8GB dimms to work but that in absolutely does not mean the cpu is able to support a higher max than stated by Intel. It isn't proven that 4*8GB works properly in the single cpu Mac Pro. OWC is the only one I could find who claims it does and does very little to actually prove it. I think we need more user experience in this area.
See above.
The information is based on what the memory controller can actually do, and there's no evidence that Apple's firmware implementation is borked in this case (one user here is running the Samsung 8GB UDIMM's IIRC without issue, as does the 8GB RDIMM's by a few users as well).
Now it's possible that the memory capacity is reported at 32GB via UDIMM, but only work as 16 in a MP, but there won't be a gain in performance between 4GB UDIMM and 8GB UDIMM. No such report yet, though as 8GB UDIMM's are rare, it's not likely we'll see this one bore out one way or the other in a MP.
The Register chip on RDIMM changes things a bit (gets around the impedance problem of so many memory chips - Samsung did it by increasing the memory density in their memory chips thus keeping it within spec impedance wise). What reports we do have here on MR, seem to indicate that there is a significant performance increase (i.e. applications can use the additional capacity).