Wassup with Xeon chipsets and RAM?

[PCBONEZ]

Looking at some boards with Intel E7525, E7520, and E7320 Chipsets and I'm seeing stuff like:

Up to 16GB DDR 333 (or) Up to 32GB DDR 266.

Seems odd to me that the max changes with speed.
Anyone know the reason?

Thx!

.

[zandrax]

Re: Wassup with Xeon chipsets and RAM?

Not 100% sure it's the same issue, but I remember Athlon Via chipset could address up to 3 ram slot at full speed, but using the 4th forced the chipset to reduce ram frequency to avoid data corruption: perhaps modules with bigger density chip share this problem.

Zandrax

[PCBONEZ]

Re: Wassup with Xeon chipsets and RAM?

That makes sense,,,

... and you reminded me of something I'd forgotten about.
I have some old Intel N440BX server boards I'm not using anymore.
Those would take either registered or non-registered PC100.
non-registered maxed at 1GB and registered maxed at 2GB.
- That must have been for the same reason.
.

[gonzo0815]

Re: Wassup with Xeon chipsets and RAM?

I know similar issue from old BX chipset, which could only handle the full memory density with registered modules.
Same for most SKT754 systems, which are going down to ddr333 when more then two sticks of ram are used.

From my understanding this is due to increased capacitive and inductive load to the output drivers of ram controller.
Either buffering the dims or reducing frequency alleviates the problem because of a lot less load at the ram drivers.
A better layout or better output drivers would certainly fix this problem too, but both is probably more expensive or not possible at all.

[gdement]

Re: Wassup with Xeon chipsets and RAM?

The BX limitation was officially 512MB unbuffered, or 1GB registered. But that chipset was so conservatively spec'd that 1GB unbuffered is really not a problem. The only time I've noticed minor issues is at PC133 (it's only supposed to run at PC100).

I think something else that happens in chipset specs is that, at the time they're written, it's only possible to make the biggest modules by using 4-bit RAM chips. That means activating 16-18 chips at once, and that's too much load without making it a registered module. Some el-cheapo modules use 4-bit chips unbuffered, but they're in violation of specs.

Up to 16GB DDR 333 (or) Up to 32GB DDR 266

I think there was a rule like that for the i815, if you had too many modules (or maybe too many rows) then it was supposed to drop to PC100. There's definitely a DDR333/266 capacity rule for the SiS645, I remember that from setting up my sister's computer. I just wish the consumer DDR chipsets would support registered memory, like the BX did.

[gg1978]

Re: Wassup with Xeon chipsets and RAM?

It's pretty commonplace, really. As others have said BX only officially supported 512 unbuffered, or 1G registered, though it never seemed to have issues with 1G unbuffered at 100FSB.. The later Via KT series (KT333, KT400, KT400A, KT600) all dropped the max RAM speed with all slots full. Intel 815 had a 512M hard limit, it would not address any more than that.. The AMD 760 supported only two slots unbuffered, if you wanted to fill all 4 slots, they HAD to be registered.. Serverworks chipsets, at least on the old P3 ones, require registered ECC DIMMs.. Older AMD 64's would drop the DIMM speed with all 4 slots loaded..


These are all older SDR and DDR systems.. DDR2 did make signal integrity a bit better, i'm guessing because of the BGA packaging that DDR2 comes in.. Though even with DDR2, you still see that typically filling all DIMM slots will reduce the memory speed..