54 months ago
I've been farting around with an old HP xw6400 workstation - exclusively uses a special kind of RAM called FB-DIMMs. They run very hot, have mandatory heat spreaders, and feature a little asic right the center to help them (I assume with some kind of error correction, since FB-DIMMs are classed as a kind of ECC).
To get to the point, I've upgraded the machine now to an 8 core (2 CPU) configuration, but upgrading the RAM is proving very difficult. I bought a 2x4gb set off ebay that met all the listed specifications required by HP for this workstation: 4gb PC2-5300F, CAS 555-11, 1.8V. The system supports up to 16gb RAM across 4 slots, so 4gb must be supported. This RAM is not supported. Turns out, it's "4RX8" RAM, while this xw6400 ONLY accepts " 1RX4" or MAYBE "2RX4" RAM.
What are these bizarre numbers, you ask? Well, I was baffled, so I did some research. I've found that at least the first half of this number refers to something called "rank," which apparently refers to the method of adding RAM caches to the stick - single rank RAM is faster because of its construction, dual rank RAM is slower, and quadruple rank RAM is slowest of all - and my xw6400 doesn't support quad rank RAM at all in the BIOS - just doesn't recognize the memory (if I have 1RX4 - single channel - RAM in slot 1, and 4RX8 RAM in any other slot, it boots with an error, but I can get into BIOS, where it shows me that the total system memory detected is only the sum of the 1RX4 RAM, even though the 4RX8 RAM is detected and listed in the slot that it's in!).
So, obviously there are real differences here, but I'm intrigued by the suggestion that lower rank construction methods yield RAM that is somehow faster. Is this still true for everyday consumer RAM, or even modern ECC RAM, or is this just something that used to be important but no longer really is?