KVR13R9D4K3/48 and KVR13R9D4/16 please help me to spot the differences

UnicornU

Distinguished
Jul 10, 2013
91
0
18,640
I recently bought these two items from Kingston
http://www.kingston.com/dataSheets/KVR13R9D4K3_48.pdf
and
http://www.kingston.com/dataSheets/KVR13R9D4_16.pdf

They have identical specs but someone at Kingston told me that they might not work together. Before I decide to return them I wanted to check this out with you guys:

The 48GB kit (3x16G) is made with these chipsets


DDR3 SDRAM
H5TC4G43AFR
The H5TC4G43AFR-xxA, H5TC4G83AFR-xxA and H5TC4G63AFR-xxA are a 4Gb low power Double Data Rate III (DDR3L) Synchronous DRAM, ideally suited for the main memory applications which requires large memory density, high bandwidth and low power operation at 1.35V. DDR3L SDRAM provides backward compatibility with the 1.5V DDR3 based environment without any changes. (Please refer to the SPD information for details.)
SK hynix 4Gb DDR3L SDRAMs offer fully synchronous operations referenced to both rising and falling edges of the clock. While all addresses and control inputs are latched on the rising edges of the CK (falling edges of the CK), Data, Data strobes and Write data masks inputs are sampled on both rising and falling edges of it. The data paths are internally pipelined and 8-bit prefetched to achieve very high bandwidth.
Features

  • VDD=VDDQ=1.35V + 0.100 / - 0.067V
    Fully differential clock inputs (CK, CK) operation
    Differential Data Strobe (DQS, DQS)
    On chip DLL align DQ, DQS and DQS transition with CK transition
    DM masks write data-in at the both rising and falling edges of the data strobe
    All addresses and control inputs except data, data strobes and data masks latched on the rising edges of the clock
    Programmable CAS latency 5, 6, 7, 8, 9, 10, 11, 13 supported
    Programmable additive latency 0, CL-1, and CL-2 supported
    Programmable CAS Write latency (CWL) = 5, 6, 7, 8, 9
    Programmable burst length 4/8 with both nibble sequential and interleave mode
    BL switch on the fly
    8banks
    Average Refresh Cycle (Tcase of 0 oC~ 95oC)
    - 7.8 µs at 0oC ~ 85 oC
    - 3.9 µs at 85oC ~ 95 oC
    JEDEC standard 78ball FBGA(x4/x8), 96ball FBGA(x16)
    Driver strength selected by EMRS
    Dynamic On Die Termination supported
    Asynchronous RESET pin supported
    ZQ calibration supported
    TDQS (Termination Data Strobe) supported (x8 only)
    Write Levelization supported
    8 bit pre-fetch
    This product in compliance with the RoHS directive.

The other one (1x16GB) is made with this memory chip:

DDR3 SDRAM
H5TC4G43MFR
The H5TC4G43MFR-xxA, H5TC4G83MFR-xxA and H5TC4G63MFR-xxA are a 4Gb low power Double Data Rate III (DDR3L) Synchronous DRAM, ideally suited for the main memory applications which requires large memory density, high bandwidth and low power operation at 1.35V. SK hynix DDR3L SDRAM provides back-ward compatibility with the 1.5V DDR3 based environment without any changes. (Please refer to the SPD information for details.)
SK hynix 4Gb DDR3L SDRAMs offer fully synchronous operations referenced to both rising and falling edges of the clock. While all addresses and control inputs are latched on the rising edges of the clock (falling edges of the clock), data, data strobes and write data masks inputs are sampled on both rising and falling edges of it. The data paths are internally pipelined and 8-bit prefetched to achieve very high bandwidth.
Features

  • VDD=VDDQ=1.35V + 0.100 / - 0.067V
    Fully differential clock inputs (CK, /CK) operation
    Differential Data Strobe (DQS, /DQS)
    On chip DLL align DQ, DQS and /DQS transition with CK transition
    DM masks write data-in at the both rising and falling edges of the data strobe
    All addresses and control inputs except data, data strobes and data masks latched on the rising edges of the clock
    Programmable CAS latency 6, 7, 8, 9, 10 and 11, 13 supported
    Programmable additive latency 0, CL-1, and CL-2 supported
    Programmable CAS Write latency (CWL) = 5, 6, 7, 8
    Programmable burst length 4/8 with both nibble sequential and interleave mode
    BL switch on the fly
    8banks
    Average Refresh Cycle (Tcase of 0 oC~ 95oC)
    - 7.8 µs at 0oC ~ 85 oC
    - 3.9 µs at 85oC ~ 95 oC
    Auto Self Refresh supported
    JEDEC standard 78ball FBGA(x4/x8), 96ball FBGA (x16)
    Driver strength selected by EMRS
    Dynamic On Die Termination supported
    Asynchronous RESET pin supported
    ZQ calibration supported
    TDQS (Termination Data Strobe) supported (x8 only)
    Write Levelization supported
    8 bit pre-fetch
    This product in compliance with the RoHS directive

I have marked the difference with red


Complete specs here
https://www.skhynix.com/inc/pdfDownload.jsp?path=/datasheet/pdf/dram/Computing_DDR3L_H5TC4G4(8_6)3MFR(Rev1.0).pdf
https://www.skhynix.com/inc/pdfDownload.jsp?path=/datasheet/pdf/dram/Computing_DDR3L_H5TC4G4(8_6)3AFR(Rev1.0).pdf


Will they work together on my MB (KGP-D16 with two processors installed)?

 
Solution
I'd be hesitant to try mixing16GB sticks can be finicky and it appears these are made using different Hynix chips, with your mobo you'd be wanting to run run 2 sticks to serve each CPU, so one CPU will be using different sticks, if it throws things off with that CPU it can throw load balance off on the whole rig. If they do play, since it's ECC DRAM you might run into one 'side' lagging from doing a lot of error corrections, which again can throw off load balance....As far as buying individual sticks - I'd check and see what Kingston says, since it is ECC and all sticks will be the same they may say it will be fine - to be honest I wouldn't want to step on the ledge and say that... Provantage.com sells a good assortment of ECC DRAM in...
Ok, so I contacted our resident memory guru, and hopefully he'll be along to see about helping answer these questions as I can't positively answer one way or the other on this. I know for one thing he's going to say that quite often even with IDENTICAL modules that all came off the assembly line in the same batch, some of them won't work together in multiple channel modes, but he should have more solid explanations of why and if those SHOULD work together or not.

Pretty much, whatever he tells you, you can take to the bank.

The only difference I see is a possible lower latency potential on one model number, but if they're manually set I don't see what the issue would be.
 

UnicornU

Distinguished
Jul 10, 2013
91
0
18,640
Code:
I see. Thanks again for doing this.
So what is the deal with these memory kits, are they guaranteed to work together?

My problem is this, one of the sellers (the one who sold me the 1x16GB) does not accept a return since this is a special order (item had to be ordered)
The other one, NewEgg (great guys) accepted the return and my option would be to return the kit of three 16G modules and to by from the other vendor other three 1x16GB. Now from what you are saying I might have problems that way too although it would be really stupid ...
 
It's like this. GENERALLY, modules with identical or at least very, very similar specifications, when set to identical settings including speed, voltage, timings, etc., stand a chance of working together in dual, triple or quad channel modes. But sometimes they don't. They just don't.

Modules with IDENTICAL part numbers have a higher chance of working together, as all the characteristics of the module are the same and so there is less chance of incompatibility but even so, sometimes they won't run in multiple channels together. Why, I don't know. There could be any number of reasons and I don't have solid answers to this question but it might have to do with slight chip differences or maybe somebody was smoking their lunch that day. Your guess is as good as mine.

Matched sets have been verified to run together in multiple channels and unless something is physically damaged or there is a manufacturing defect that doesn't present itself until the module has been under a voltage load for a while or actual stress, there should be no reason why they do not. As we all know, sometimes brand new hardware fails right out of the box or two months later. Again, it just happens. But with matched sets there should be no question of compatibility, but even then there could be issues regarding the hardware's fitness for use.
 

UnicornU

Distinguished
Jul 10, 2013
91
0
18,640

So what happens if you have 4 modules with exactly the same specs but they don't work in 4 channels ..do they step back to 3,2,1 channel or they don't work at all?
How does the memory controller (or who else does that) decide what mode to run in? Is it testing them to see what is the best configuration?


PS: I have to go to sleep now, I will be able to answer more questions tomorrow. Thanks a lot for helping me with this
 

Tradesman1

Legenda in Aeternum
I'd be hesitant to try mixing16GB sticks can be finicky and it appears these are made using different Hynix chips, with your mobo you'd be wanting to run run 2 sticks to serve each CPU, so one CPU will be using different sticks, if it throws things off with that CPU it can throw load balance off on the whole rig. If they do play, since it's ECC DRAM you might run into one 'side' lagging from doing a lot of error corrections, which again can throw off load balance....As far as buying individual sticks - I'd check and see what Kingston says, since it is ECC and all sticks will be the same they may say it will be fine - to be honest I wouldn't want to step on the ledge and say that... Provantage.com sells a good assortment of ECC DRAM in large (4x16GB) sets, and actually so does Amazon. Both also cater to Kingston selections
 
Solution
He's in Canada and apparently the shipping and customs taxes are cost prohibitive so he'd need to find them through a Canadian retailer. I found him what looks to be a decent matched set but it seems he's in a bit of a bind since he already has those modules and can't apparently return the single module. I'd think if he bought three other identical modules he's probably ok, but as you say, I'd want to verify that with Kingston rather than go out on a limb and blatantly say it's good and then find out it's not.
 

Tradesman1

Legenda in Aeternum

__________

+1 Sorry missed the Canada part, think Provantage deals in Canada, though not positive...sorry :(
 

UnicornU

Distinguished
Jul 10, 2013
91
0
18,640
@tradesman1 ..I forgot why I started this thread,
Can you spot any notable differences between the two chips mentioned in the first post in this thread? (see the spec PDFs for more info)
The only notable difference is this
Programmable CAS latency 5, 6, 7, 8, 9, 10, 11, 13 supported

One of the chips supports CAS latency 5 the other doesnt
How important is that?

 

UnicornU

Distinguished
Jul 10, 2013
91
0
18,640
They told me that prior to me seeing what chips were on the 1x16GB that I was waiting for..
Iam now trying to weight my risks if I go with the current modules
It will cost me 60 bucks more to get (HOPEFULLY!!) other three identical with the ones that I have

I emailed Kingston and they admitted that the chipset differences are slight and it is likely going to work in channel mode
Now I want to learn what is going to happen if they will not. See my newest thread on this