Higher density DIMM's = less overclocking headroom

[magellan]

Is this fact? Or 'net rumor?

 

[Automata]

Higher density in what way? I assume you mean within the same family (DDR2, DDR3) of RAM.

If not, I have some old sets in my Pentium Pro server that don't overclock well, and they aren't dense (except in weight!) at all.

 

[larrymoencurly]

I've read that new chip designs tend to be more overclockable because manufacturers are more conservative in binning chips at first, and their fastest rated chips may be the most overclockable because new designs get only slower ratings initially.

From what I've read, density like 28nm vs. 50nm doesn't matter, but chip manufacturers now get more consistency in a run of chips than they did in the past, so they don't need anywhere as much headroom as they once did. For example, some of the very early DRAM chips had to be built so the bits averaged 3x the speed specs of the chip because otherwise too many of the slowest bits would fall short of specs.

OTOH I don't see how to tell the overclocking headroom of a module unless you know the actual specs of the chips beneath its heatsinks. But it wouldn't surprise me if there's less headroom because module companies are squeezed on costs.

 

[magellan]

When I meant higher density, I meant more ranks. For example, my 2GiB DIMM's are single rank and I've been able to take them from their rated 1600Mhz spec up to over 2000Mhz and still keep them within their maximum rated voltage. I've read that if I had high density DIMM's that had two ranks or more (quad rank I think is the max) I wouldn't be able to overclock nearly that much.

The IBM System x3550 M3, IBM Redbooks Product Guide has a section which refers to the maximum frequency at which the memory can be run and tying it to the number of ranks that
a DIMM has -- the higher the number of ranks, the slower the memory must be run. I remember
that some chipsets in the past had limitations on the number of ranks that a DIMM could have as
well (I think the Nforce2 chipset had these limitations).

 

[jawadesign]

magellan,

Not sure if this is helpful or not but...
http://www.oempcworld.com/support/SinglevsDualRAM.html

On a64‐bit (non‐ECC) DIMM made with two ranks, there would be two sets of DRAM that could be accessed at different times. At the end, it states some DIMMS are being manufactured with up to 4-ranks!?

 

[ivanlabrie]

Is this fact? Or 'net rumor?

What gen?

Talking ddr3, you can't oc newer modules as much as older sticks because of the types of ic's used on them. The memory chips that are produced nowadays tend to be better for loose timings and high mhz vs the older tight timings and high mhz (but not as high).

More sticks put more strain on the memory controller, so that contributes to a lower max oc, but higher density chips by themselves don't warrant that.

 

[magellan]

What gen?

Talking ddr3, you can't oc newer modules as much as older sticks because of the types of ic's used on them. The memory chips that are produced nowadays tend to be better for loose timings and high mhz vs the older tight timings and high mhz (but not as high).

More sticks put more strain on the memory controller, so that contributes to a lower max oc, but higher density chips by themselves don't warrant that.

My question isn't so much about the types of IC's used, as much as how they're organized. As jawadesign's last reply indicated I'm more interested in how the number of ranks on a DIMM affect its overclocking potential. I think it's usually the case that higher density/capacity DIMM's have more ranks. I don't think there are any 8GiB DIMM's that have a single rank. And you've answered that. Thanks.

 

[jawadesign]

megellan,

cross posting, but Kingston does make a 8GB DIMM that's single rank... Intel compatibility tested. It's expensive though, $400+ on Amazon for a 4-pack (32GB)
http://www.kingston.com/dataSheets/KVR16R11S4K4_32I.pdf

shopping for RAM again, lol!



...

 

[keny]

Single sided sticks usually can clock higher, especially the newer hynix sticks

 

[magellan]

megellan,

cross posting, but Kingston does make a 8GB DIMM that's single rank... Intel compatibility tested. It's expensive though, $400+ on Amazon for a 4-pack (32GB)
http://www.kingston.com/dataSheets/KVR16R11S4K4_32I.pdf

shopping for RAM again, lol!

...

Those are registered. I wonder if there are any UDIMM's like that?

I still hope you're going to do your memory bandwidth tests, I'd be
interested to see if having two DIMM's per channel with rank interleaving
gets you anything over having one DIMM per channel, that's if you
get single rank DIMM's. From what I understand you can do rank
interleaving w/a single DIMM of > 1 rank.

I looked up your CPU vs. the older Westmeres and your newer, better
LGA 2011 is cheaper than the older 6 core Westmeres. But why?

 

[jawadesign]

megellan,

I don't know, I've only been looking up info on RDIMM type RAM. My last order for RAM got canceled, so I just went with 4 sticks of 4GB|RDIMM|ECC|1600MHz (single rank). Dumb lucky, but the second go around will yield me the fastest module for my system, so long as I have my information correct

You know way more than me about messing with and configuring RAM, I'm just learning and trying to put my info together. I hope what I bought works (ie compatible with my S30), LOL! Fingers crossed, I hope there is some type of bench test out there for a systems RAM configuration, filling channels, filling half the channels, and so on.

 

[ivanlabrie]

Interesting stuff indeed...I'd be curious to see more conclusive testing on the bank interleaving thing, it's been heavily debated in benching circles. Most xtremesystems guys agree that you get a performance increase with 4 sticks vs 2 sticks (double sided ic's).

 

[magellan]

Interesting stuff indeed...I'd be curious to see more conclusive testing on the bank interleaving thing, it's been heavily debated in benching circles. Most xtremesystems guys agree that you get a performance increase with 4 sticks vs 2 sticks (double sided ic's).

So that configuration would get you channel interleaving?

I would be interested to see what kind of difference rank interleaving makes (which I can't do because I have low density [2GiB], single rank sticks). The IBM whitepaper I read seemed to indicate having at least 2-rank interleaving was best.

My basic question was whether or not the number of ranks
in a DIMM affects the overclocking margins.

 

[EarthDog]

I'd help, but I do not want to add to the guessing that is already going on. My suggestion is to PM Woomack to have him come check out this thread and get things straightened out.