TR Forums

I commented in this recent article about 3800MHz SODIMMs having the same speeds, timings, and voltages as "full size" desktop sticks (I presume the >3800MHz chips are simply binned for desktop kits currently since that's pretty much the only place you can actually utilize those OC speeds). There are already some options for desktop mobos and SODIMM RAM so it got me thinking, wouldn't it make sense to just move the entire industry over to SODIMM only? That would consolidate products and reduce the chances/risk of unsold inventory at the very least.

Admittedly the article is a bit boring so it probably didn't attract many views, but I don't think my question got answered. Can anyone tell me what's the reason for not just going SODIMM for everything (desktop and laptop)?

I wholeheartedly agree. Only drawback I can think of is cost--SODIMMs are obviously slightly more expensive.

Going further, that'd leave PCIe as the last "big" pin interface. MXM still hasn't caught on really.

If I had to guess, 284 pins in a regular DIMM are spaced further apart than the 260 pins in a SODIMM so cheaper slots can be used and the PCBs themselves don't have to be made to quite such high standards to stay within tolerances. Packaging costs are non-trivial with RAM, and I'd also assume that the extra PCB surface area allows easier/cheaper BGA mounting, easier trace routing, and possibly this allows the use of PCB with fewer layers, which also reduces costs.

I'm speculating, of course, but if costs weren't a factor, more manufacturers would have switched to SODIMMs to simplify their supply chain and inventory management; They only care about their bottom line, so the combination of regular DIMMs and 284-pin slots must be significantly cheaper when bought in bulk by OEMs. What we see with retail price points on individual packs of consumer RAM bears little relevance to the "guaranteed order size of two million units over the next quarter" that big OEMs like HP, Dell, Lenovo are pulling down.

Duct Tape Dude wrote:

Only drawback I can think of is cost--SODIMMs are obviously slightly more expensive.

Possibly. Although it appears 8GB modules are pretty much the same cost. 16GB SODIMM modules are up to 10% more than desktop kits, but that makes me wonder if some of that is just volume. How many laptop owners are really upgrading/adding RAM considering they've only got 2 slots compared to desktop owners?

While this isn't a bad idea it would require a lot of motherboard redesigning, both for the smaller traces and the need for horizontal space where today the space taken up is vertical. Could be rough for 8-channel workstation boards.

I guess the slots could be redesigned as well to not need the retaining clips? That's even more work for motherboard OEMs though, and they hate that.

Chrispy_ wrote:

284 pins in a regular DIMM are spaced further apart than the 260 pins in a SODIMM so cheaper slots can be used

Makes sense. Cart + Horse situation.
Wouldn't the benefit to OEMs come in the fact that they would only need to order and stock SODIMMs though? Higher volume orders if everything is consolidated. Even the slot itself should become cheaper simply by economy of scale, although it remains to be seen whether that would offset the perceived cost difference or not...

Just spit-balling. Not trying to shoot anyone down.

Vhalidictes wrote:

Could be rough for 8-channel workstation boards.

As in that AsRock X299 board, the SODIMMS can stand vertically. I'm glad you brought up the size though, I forgot to comment on that. A SODIMM is almost half the length of a DIMM, so I was actually thinking it would be more conducive to multi-channel configurations going with the idea that you could arrange 8 sticks of SODIMMs (4 columns, 2 rows) in the same orientation as 4 sticks of DIMMs (4 columns, 1 row) and they would take up nearly the same amount of space on the mobo. Probably use a one-sided latch system with the fixed latch between rows 1 and 2.

How do you fit more than 16 (though I *guess* you could squeeze in two more for ECC) ICs on a SO-DIMM? There are DIMMs (though atypical) with 32.

And that's with those squarish FBGA packages, because with TSOP you really couldn't fit more than 8 on a SO-DIMM whereas you can easily get 16 (which is normal) on a regular DIMM.

So maybe it's less relevant now (TSOP seems deprecated or even obsolete for memory ICs), but unless servers are willing to forgo quad-rank for maximum density and low-end desktops for possible efficiency (twice as many low density chips can potentially be cheaper than half as many normal density ones) DIMMs are still going to stick around.

Regular sized DIMMs have the room to hold enough chips for quad rank, registered ECC memory for ultra high capacity. I think off hand that SO-DIMMs only have enough room for dual rank and registered with a bit of extended height. So in that regard, they wouldn't work in the service space where memory capacity is a selling point (and that ignores that octo rank is permitted in the DDR4 spec too). I think the larger DIMMs also are allocated more power within the spec too so that it can drive the additional memory chips.

I've also noticed that SO-DIMMs top-out at lower speeds, and are usually 1-2 CAS cycles slower than similar standard DIMMs. I couldn't tell you whether this is due to a physical limitation or just that they're aimed at different markets (mobile isn't as performance-oriented), but it's been the case since at least DDR1.

Glorious wrote:

(TSOP seems deprecated or even obsolete for memory ICs)

I believe it is a signal integrity issue. At the bus speeds used by modern DRAM chips, the additional inductance and/or capacitance of the TSOP leads is enough to cause problems.

Here's where I'm going with this. 4 SODIMM sticks on a mITX mobo.
Here's the original image. The SODIMM slots actually fit perfectly end-to-end on top of the DIMM slots!!

I believe there are some mitx boards with SO-DIMM slots instead of DIMM slots.

I'll have to Google around and drop in a link here as an edit if I can find them.

Edit:

http://www.asrock.com/mb/Intel/H81TM-ITX%20R2.0/

There we go.

So: lower performance and lower capacity, but at least they're more expensive. Let's do this.

UberGerbil wrote:

So: lower performance and lower capacity, but at least they're more expensive. Let's do this.

I really don't see the point unless the application is mobile or at least SFF.

UberGerbil wrote:

So: lower performance and lower capacity, but at least they're more expensive. Let's do this.

I have to admit this is exactly what I thought too. I understand one or more of those might be mitigated at scale, but I just don't see the point of doing it for desktops/servers. /shrug

Glorious wrote:

So maybe it's less relevant now (TSOP seems deprecated or even obsolete for memory ICs), but unless servers are willing to forgo quad-rank for maximum density and low-end desktops for possible efficiency (twice as many low density chips can potentially be cheaper than half as many normal density ones) DIMMs are still going to stick around.

BGA mounting was a requirement for DDR2 and beyond. Some high end DDR1 chips back in the day would use BGA packaging so this style of packaging has been around for a long time.

While not obvious externally, some of the high end DDR4 chips are using TSV to increase capacity. So memory packaging technology is continuing to evolve. So far each stack is seen as a single rank, though I'm not sure if this is a hard requirement in the DDR4 spec or not.

UberGerbil wrote:

So: lower performance and lower capacity, but at least they're more expensive.

DancinJack wrote:

I have to admit this is exactly what I thought too.

Lower performance: I already discussed this and is most of what sparked this thread. Please read OP. Also as stated above, clearly it's possible to manufacture high speed SODIMMs, but with them limited to laptops which have no ability to OC RAM, there's no incentive.
Lower capacity: Both DIMMs and SODIMMs offer up to 16GB per module (stated above)
More expensive: 8GB SODIMMs are mostly on par with DIMM prices. 16GB modules are only about 10% more (also already stated above)

Except they are lower performance most of the time as things stand right now. In spite of the recent 3800MHz modules that TR posted about, which don't seem to be on sale yet, almost every single other SO-DIMM kit is slower than what you can get on the DIMM side for desktop/server at the top end. That's just how it is right now.

Oh and here is a 64GB module. There are a number of them on newegg alone. Some 32GB modules too. Can you fit that into a SO-DIMM? Please, show me one? https://www.newegg.com/Product/Product. ... 6ZP6HA9393 Newegg currently carries 0 32GB or 64GB SO-DIMM modules.

You talk about the prices yourself. SO-DIMMs, are on average, more expensive for their exact counterparts on the DIMM side. That's just how it is now.

I read the OP, and your suggestion that I didn't isn't appropriate. You also didn't include the part of my post where I said that if SO-DIMMs were adopted large-scale, then a lot of these issues would be mitigated. Let's not get it twisted here DPete.


Edit: here is a good example. G.Skill 3200MHz 16GB (8GBx2 kit). I think that's a pretty "mainstream" kit for a lot of PC builders these days.

SO-DIMM (only a SINGLE option from G.Skill) - 249.99, and it has higher timings than the desktop version(s).
DIMM - 192.99, and as stated above, it has tighter timings which help quite a bit depending on the application and CPU generation these days. I admit this is higher voltage, but 1.2V vs 1.35 isn't that huge of a deal and there is precisely zero other options for G.Skill SO-DIMM kits here. This is the most expensive G.Skill 16GB 3200MHz kit on newegg.

I don't think this comparison is a poor representation either. It's not the exception, but rather the rule of the economics of DIMM vs SO-DIMM right now. Unless there is some serious momentum happening for SO-DIMM for some reason, I just don't see them taking over.

DPete27 wrote:

Lower capacity: Both DIMMs and SODIMMs offer up to 16GB per module (stated above)

But that's simply wrong.

You can buy 64 GB DIMMs. http://www.crucial.com/usa/en/ct2k64g4lfq424a

Will they work in your desktop? No, of course not. But since your entire point here is based on the concept of benefiting from the economy of scale inherent to standardization, well, it isn't going to happen. Plenty of servers use the additional PCB space available on those DIMMS, and they're not going to stop doing it.

And, as I also said, it's not impossible for desktops to use similar concepts when it's cheaper to buy 2 less density chips than 1 higher density chip. I have no idea how prevalent this is on the low-end, but I used to see stuff like it.

DancinJack wrote:

almost every single other SO-DIMM kit is slower than what you can get on the DIMM side for desktop/server at the top end. That's just how it is right now.

Sure, but that's almost certainly because faster = hotter and it's not a reasonable trade-off for mobile or SFF. So there's currently not much market pressure for that.

However, we're still undermining the original point behind Dpete27's pitch:

DPete27 wrote:

hat would consolidate products and reduce the chances/risk of unsold inventory at the very least.

I mean, it might, on the generic side of things, but not on the performance side of things. We'd still have different SKUs, and if you use a performance SO-DIMM for a laptop you're either wasting money or lighting your lap on fire.

Glorious wrote:

DancinJack wrote:

almost every single other SO-DIMM kit is slower than what you can get on the DIMM side for desktop/server at the top end. That's just how it is right now.

Sure, but that's almost certainly because faster = hotter and it's not a reasonable trade-off for mobile or SFF. So there's currently not much market pressure for that.

Point taken and conceded. So let's take an even more mainstream kit to task. Let's talk about 2666MHz 16GB (8GBx2) kits.

SO-DIMM: Again, a single option from G.Skill. 210.88 with timings of 18-18-18-43 at 1.2V.
DIMM: Multiple options. Highest of which is priced at 162.88. These kits have tighter timings, a lower price, the same operating voltage, and more selection.

This is just the way things are right now. With SO-DIMMs, you either get lower performance, lower capacity, or higher price the majority of the time. Don't forget the lack of options in a lot of cases for SO-DIMMs.

edit: I maintain that you'd see minimization of these issues with mainstream adoption of SO-DIMMs in desktops, but I just don't see where the pressure comes for that. It'll cost everyone(OEMs) a bunch of money to transition for virtually no benefit.

Glorious wrote:

But that's simply wrong. You can buy 64 GB DIMMs.

The server grade arena wasn't my intended focus, but it can't be ignored either since I DID say "for everything." Whereas consumer tier RAM doesn't seem to fully utilize the physical size of DIMMs, server tier stuff certainly can/does pack a huge amount of chips on each stick. Admittedly, I don't have an answer for this argument. You could separate DIMMs for ECC and SODIMMs for non-ECC, but the "consumer tier" CPUs and chipsets that support both ECC and non-ECC muddies those waters.

DPete27 wrote:

Glorious wrote:

But that's simply wrong. You can buy 64 GB DIMMs.

The server grade arena wasn't my intended focus, but it can't be ignored either since I DID say "for everything." Whereas consumer tier RAM doesn't seem to fully utilize the physical size of DIMMs, server tier stuff certainly can/does pack a huge amount of chips on each stick. Admittedly, I don't have an answer for this argument. You could separate DIMMs for ECC and SODIMMs for non-ECC, but the "consumer tier" CPUs and chipsets that support both ECC and non-ECC muddies those waters.

You don't need DIMMs for ECC, there just isn't any market for ECC memory in laptops right now. (since laptops are virtually the only place SO-DIMMs are right now). If SO-DIMMS were the mainstream RAM package you would see a lot more.

DancinJack wrote:

You don't need DIMMs for ECC

I know, but people are using high-density ECC DIMMs to burn my balls, so I'm having to admit defeat on the server tier side of things (even though, as shown above, you could stuff twice as many SODIMMs into the same size server mobo)

DPete27 wrote:

I know, but people are using high-density ECC DIMMs to burn my balls, so I'm having to admit defeat on the server tier side of things (even though, as shown above, you could stuff twice as many SODIMMs into the same size server mobo)

Look, the industry isn't going to change because you made an irrefutable argument on techreport.com that they should be using SO-DIMMs.

So there is no sense in taking this personally, much less our arguments over it. I mean, this isn't just theoretical, we're all just trying to practically explain what we empirically see, right?

That's the thing: Regardless of whether or not we can even identify it here, the assumption should be that there is a reason that the industry persists in using DIMMs. Especially since we know it can't just be historical inertia, as every time there is new memory standard (every ~5 years) the entire issue is revisited by JEDEC.

I mean, I'm sort of curious now. Weren't you asking an earnest question?

Was it meant to be rhetorical?

Another point is that SO-DIMMs have slightly tighter pin spacing than regular DIMMs. In addition to the more expensive layering and trace routing that results, it also reduces the wear-and-tear tolerance of the socket. In a laptop where RAM is commonly upgraded once or never, no big deal. In a desktop or especially a server where RAM may be replaced or upgraded multiple times, it's more of a big deal.

DPete27 wrote:

DancinJack wrote:

You don't need DIMMs for ECC

I know, but people are using high-density ECC DIMMs to burn my balls, so I'm having to admit defeat on the server tier side of things (even though, as shown above, you could stuff twice as many SODIMMs into the same size server mobo)

You'd need to have twice as many memory channels or twice as many modules per channel to support the same memory capacity though. That doesn't come for free. To add channels you'd need to redesign the memory controllers; and doubling the number of modules per channel would likely result in a speed penalty (lower clock speeds due to additional bus loading and trace length). Having twice as many memory sockets is also likely to pose some PCB layout challenges; if additional PCB layers are needed that'll drive costs up. So it's not just a simple "we can substitute two of these for one of those" deal.

I know very little about limitations in the server arena and if all RAM channels are being utilized. I wasn't suggesting "use two of these to equal one of those". That would be silly

DancinJack wrote:

You don't need DIMMs for ECC, there just isn't any market for ECC memory in laptops right now. (since laptops are virtually the only place SO-DIMMs are right now). If SO-DIMMS were the mainstream RAM package you would see a lot more.

Intel is doing a push with mobile Xeons in the 25W space. These do support ECC. They're not exceedingly common but they do exist. Example

Similarly, the blade and high density server environments which have occasionally used SO-DIMMs to shave off physical space to increase density. These use the same ECC DIMMs as the Xeon laptops.