Why avoid generic PCBs for DDR?

[Muse]

It seems to me that the quality of the chips on a stick of memory is 1000 times as important as the PCB they are seated on. The PCB is just a printed circuit board, or is it? I'm sure that they are completely speced out for the chips that go on them. As long as they are made of good materials, with excellent conductive properties, I'd think that they'd work fine. What am I missing here? I've seen a lot of posts from people saying that you'll get better results if you don't use generic PCBs for DDR.

I just bought Samsung chips on an SAtech's Excelerate PCB: 512 MB PC2700 non-ECC.
Why would this behave differently from "original Samsung?" What IS
original Samsung?

 

[rimshaker]

Some PCB's are manufactured or designed in-house. Specific brand name boards might be tested for tighter tolerances for example. Generic boards are not bad quality, they're just manufactured with a wider ranger of specs, so that anyone can pick them up and start building on them.

 

[Carrot44]

Guess answer
The inhouse PCB's are built to the parameters of the chips used

 

[Sunny129]

memory chip makers who also make their PCB in-house (such as Samsung Original) have designed a reference board specific to their chips. its like nVidia and their video cards...they create a GPU and place it on their own reference board. then they test it and send it to hardware reveiwers to be tested. the same goes for many chipset makers, like SiS, Intel, and VIA. all their chipsets are first placed on a reference board in-house and tested. and if you've ever noticed the trend, the reference board (video card, mobo, or whatever...) almost always performs better than any 3rd party manufactured board performs. so in a sense, when you buy Samsung Original, you are getting a "reference board" b/c it is their chips on their PCB. so that is why people speak of in-house PCB as being better than 3rd party manufactured PCB when it comes to memory...one of the reasons anyway...

 

[Muse]

Originally posted by: Sunny129
memory chip makers who also make their PCB in-house (such as Samsung Original) have designed a reference board specific to their chips. its like nVidia and their video cards...they create a GPU and place it on their own reference board. then they test it and send it to hardware reveiwers to be tested. the same goes for many chipset makers, like SiS, Intel, and VIA. all their chipsets are first placed on a reference board in-house and tested. and if you've ever noticed the trend, the reference board (video card, mobo, or whatever...) almost always performs better than any 3rd party manufactured board performs. so in a sense, when you buy Samsung Original, you are getting a "reference board" b/c it is their chips on their PCB. so that is why people speak of in-house PCB as being better than 3rd party manufactured PCB when it comes to memory...one of the reasons anyway...

So, I'm wondering if some generic PCBs would work (for all "practical purposes") as well as the _reference boards_, i.e. the chip manufacturer's own PCBs. I guess there's a lot more to the DDR PCBs than I thought. It's not just a PCB - there's actually some componentry on the PCBs.

 

[Sunny129]

So, I'm wondering if some generic PCBs would work (for all "practical purposes") as well as the _reference boards_, i.e. the chip manufacturer's own PCBs. I guess there's a lot more to the DDR PCBs than I thought. It's not just a PCB - there's actually some componentry on the PCBs.

i suppose there would be cases in which 3rd party PCB's perform as well as (and maybe even better than) reference design in the memory industry. look at the motherboard market for a moment. ASUS is one of the best 3rd party board makers, if not the best, when it comes to quality, stability, and reliability of products. now they make several different mobos using several different chipsets, so somewhere in there i'm willing to bet that some of their boards follow the chipset maker's reference design almost down to the last detail, while other boards use ASUS's own PCB design. either way, ASUS does a tremendous job of putting out a high quality product, thus justifying their slightly higher prices and industry lead. so basically a 3rd party PCB design may perform as well as a reference board b/c either the 3rd party board maker used the chipmaker's reference design, or their own design just happens to perform as well, and in some cases better. more often than not though a 3rd party PCB design doesn't perform as well as the reference design. this is mostly b/c the chipmaker knows exactly how to opsition the traces, capacitors, inductors, MOSFETs, jumpers, etc. to allow for optimal chip performance (whether it be a CPU, GPU, northbridge or southbridge, FireWire controller, memory chip, or any other kind of chip). companies like ASUS are few and far between, and there are many more average to sub-par manufacturers out there than there are high quality manufacturers. equivalents of ASUS in the memory industry would be Samsung, Micron, Corsair, and Mushkin. but of course you know there are several other memory manufacturers out there...some recognizable names, but mostly no-name companies you've never heard of before. and they all tend to use memory chips from the reputable companies like Samsung and Corsair, but their PCB is the difference. so if you arent getting memory directly from the chipmaker on the reference board, or arent getting memory from a 3rd party manufacturer who follows the referecne design, or arent buying from a 3rd party manufacturer known for their high quality even if they dont follow the reference design, then the memory most likely will not perform as well.

as far as componentry goes, it works just like the mobo analogy i used above...although by looking at a memory module, you see very little other than some traces and the chips themselves (unlike mobos, which are littered w/ components). but rest assured the components are there, even if they arent in plain sight. and their size, position, and direction is all key in making the chips perform as best they can.

 

[Dulanic]

Another big factor is the amount of PCB layers, alot of generic RAM uses 4 layer PCBs vs the 6 layer PCBs brand name stuff uses.

 

[LanEvoVI]

Originally posted by: Dulanic
Another big factor is the amount of PCB layers, alot of generic RAM uses 4 layer PCBs vs the 6 layer PCBs brand name stuff uses.

What is the advantage of having more layers?

 

[Sunny129]

What is the advantage of having more layers?

the traces we see on the surface of a PCB aren't the only traces on the board. in order for two traces to get to their destination, it might require them to "cross pathes." but if you know anything about circuitry, you'll know that two traces cannot touch, or one or both circuits will be shorted. a trace is simply part of a ciruit that connects one component to another. data flow that needs to move from one component to another would be split and corrupted if another traces were to come in contact with the desired data path. so in order to allow for traces to "cross," one trace will be built underneath the visible ones we see on top. a motherbooard is a great example b/c it has so many system components integrated onto the board, like AGP and PCI slots, the northbridge and southbirdge, CPU socket, memory slots, etc. when you look at the massive amount of traces leaving and entering the north and south bridges, notice that they are parallel and dont touch each other, and then they spread out from there. but somewhere in the layers underneath, you can bet that there are traces running perpendicular to these parallel traces leaving the chipset. 4 layers doesn't quite allow for as much traffic, thus a 6-layer PCB allows for more traces, and therefore more data flow, and is typically referred to as higher quality PCB for mainly that reason. its just like a busy interstate junction. if there weren't bridges going over and under each other, and they all crossed one one level (or layer in the case of a PCB), we would have several automobile accidents. so traces are built like bridges, only you can only see the top layer of traces b/c the lower ones are covered by an entire layer of (usually green) fiber glass PCB.

 

[Dulanic]

Actually Im sorry most Im thinking old school, JEDEC has had 6 Layer PCB as part of the spec since PC133 I think. I believe there is a thread in here that can explain the differences much better then I can. Im a bad teacher

 

[SuperTool]

Hmm. It feels like it's been years since I took that packaging class.
But the more layers, the further you can space out leads, and you can some or parts of layers as noise shields to elminate cross-talk. I personally don't think the expensive stuff is worth the premium, but that's just me. I think if it works, it works. The difference between the top stuff and the generic stuff might justify 20% markup, but 50% or 100% no way.

 

[Muse]

Originally posted by: SuperTool
Hmm. It feels like it's been years since I took that packaging class.
But the more layers, the further you can space out leads, and you can some or parts of layers as noise shields to elminate cross-talk. I personally don't think the expensive stuff is worth the premium, but that's just me. I think if it works, it works. The difference between the top stuff and the generic stuff might justify 20% markup, but 50% or 100% no way.

Since I have a generic stick (with Samsung chips, which don't all have the same IDs, BTW), how can I go about determining how good it is? I know, this can be like chasing a bear into the woods, but I'd really like to get an idea. Thanks.

 

[RaynorWolfcastle]

Originally posted by: Muse

Originally posted by: SuperTool
Hmm. It feels like it's been years since I took that packaging class.
But the more layers, the further you can space out leads, and you can some or parts of layers as noise shields to elminate cross-talk. I personally don't think the expensive stuff is worth the premium, but that's just me. I think if it works, it works. The difference between the top stuff and the generic stuff might justify 20% markup, but 50% or 100% no way.

Since I have a generic stick (with Samsung chips, which don't all have the same IDs, BTW), how can I go about determining how good it is? I know, this can be like chasing a bear into the woods, but I'd really like to get an idea. Thanks.

Memtest86, if it passes these tests at whatever setting it's supposed to, you're OK. I ordered Samsung PC2700 on generic from googlegear, the damn thing is unstable at 166 MHz :|. I'm RMA'ing it and getting a refund, serves me right for straying from Corsair and Crucial (those two have worked better than advertised for me. every time)

-Ice

 

[Muse]

Originally posted by: icecool83

Originally posted by: Muse

Originally posted by: SuperTool
Hmm. It feels like it's been years since I took that packaging class.
But the more layers, the further you can space out leads, and you can some or parts of layers as noise shields to elminate cross-talk. I personally don't think the expensive stuff is worth the premium, but that's just me. I think if it works, it works. The difference between the top stuff and the generic stuff might justify 20% markup, but 50% or 100% no way.

Since I have a generic stick (with Samsung chips, which don't all have the same IDs, BTW), how can I go about determining how good it is? I know, this can be like chasing a bear into the woods, but I'd really like to get an idea. Thanks.

Memtest86, if it passes these tests at whatever setting it's supposed to, you're OK. I ordered Samsung PC2700 on generic from googlegear, the damn thing is unstable at 166 MHz :|. I'm RMA'ing it and getting a refund, serves me right for straying from Corsair and Crucial (those two have worked better than advertised for me. every time)

-Ice

Interesting you say this because I was just about to run memtest86 on this stick. When I booted my system less than an hour ago I immediately played an MP3 I wanted to hear. It had distortion in places, and I know it's a good file. Other MP3s sound muddy at times too. I'm testing on my "old" system (Epox 8K7A, T-Bird 1.2) because the HSF hasn't arrived yet for my new MSI KT3 Ultra2 MB system. Never had that problem with this box. Now I did swap my cheapie PSU yesterday for a new Antec True430, but I doubt that that's the problem. Shortly after I noticed the problem an hour ago, the whole system locked up and I had to reset. I'm back in Win2000, and the MP3 sounded muddy again, but no lock up so far. I'm not overclocked. In fact, I dropped my CPU back down to default 1.2 GHz yesterday although I've never had any problems at 1.4. At the time (yesterday) I wasn't aware of a problem. I bought the stick for another system that I'm going to assemble next week but decided to test on my present system because the people I bought this stick from had the audacity to put this statement on the receipt (didn't notice before):

ALL SALES ARE FINAL. NO REFUNDS.
Credit or exchange only within 7 days of purchase

At the same time they claim that they extend lifetime warranty, which seems to contradict the statements above. I called them before I even installed the stick because I'd decided that the extra $24 for Samsung Original was definitely worth not having to go through all this, but they said I should install the stick and see if it worked OK, and I did.

I really don't know how to test this DIMM. I assume that to test at 166 MHz you had to increase your CPU clock from 133 to 166 MHz. I couldn't get over 140 overclocked at 1.4 (got lockups) and that's why I dropped it down to 1.2, but I didn't try boosting the CPU clock after that. Maybe I should. Is this the way to test?

My Epox 8K7A doesn't officially support DDR333, although I've seen posts from people saying they've had success with them. My new MB is an MSI KT3 Ultra2 with KT333 chipset.