Hey Topix ...
100% agreement here .....
We need to know 3 things when it comes to DDR4 memory
1 - What brand / model # is known good
2 - What brand / model # is known bad (as is bad for these Gen 6 NUC's, but otherwise fine)
3 - For memory that falls in to "its not been tested so we don't know" ---- Are there any characteristics that could be used to identify memory that could be an issue?
Are there particular manufacturer's to stay away from when getting memory for a Skylake NUC?
These are really basic questions ....... please Intel, we need better information that "early DDR4"
I want to add that the requirements are, according to Intel:
- Supports two 1.2V DDR4 SO-DIMMs
- 2133 MHz
- Unbuffered, non-ECC
- Minimum recommended memory: 2 GB
- Maximum memory: 32 GB
- Serial presence detect (SPD)
Then there is a list with some parts that passed testing during development. However this list is useless, its only use is for a generic info, because:
- many modules (micron, hynix and samsung) are oem and so difficult to find everywhere
- none of them is a kit of two (so you risk if you buy them and want to use them in dual channel)
- corsair value is impossible to find (that code doesn't exist as a full name), Corsair produce two versions of each module and these versions aren't clearly indicated on external box product code so you don't what you have bought till you have it. There is lot more to say about these versions...
- As also Intel have indicated these part numbers might not be readily available throughout the product life cycle
- I add also that If someone do some research (I have done a lot...), it is easy to understand that there are different hynix, samsung and micron ICs, there are different revisions of the same ICand different revision of the sodimm itself
- Also most of the time the manufacturer insert a label on it that hide the IC code, or as Crucial, that delete the code from the ICs
Then there is a list with parts which have been reported as compatible by owners of the Intel® NUC Kits:
- In this list are reported Kingston KVR Value but not the Kingston HyperX that a lot of people have (I think that if we do a poll is the #1). So I don't know how this list was compiled, because I'm sure that there are a lot more people that have reported they use HyperX kits 2x8GB or 2x4GB or single HyperX than the user of one module of 16GB KVR (I remember probably one...)
After all of this, write "early DDR4" for identify a ram model that we have to avoid is useless, it doesn't mean nothing...we need to know WHAT PRODUCT CODE/BRAND HAVE THIS "EARLY DDR4".
- many modules (micron, hynix and samsung) are oem and so difficult to find everywhere
- none of them is a kit of two (so you risk if you bought these and want to use them in dual channel)
Crucial is Micron RAM. Crucial is a subsidiary of Micron and uses their modules. They are also available as a kit. I bought th 2x8GB Crucial kit and the product code on the modules is the MTxxx code that is on the list of supported RAM. I also didn't have any RAM related problems.
The problem with other (non-RAM chip) manufacturers' RAM modules is that you don't know what you get. Since these manufactures buy the RAM from one of the real RAM manufacturers (SK Hynix, Micron, Samsung), they also seem to switch as they please. I have seen pictures of the same model Corsair modules with different RAM chips on them. So it's almost impossible to predict how they will work in practice. This is one reason why I wouldn't recommend buying from Corsair. I don't know if Kingston is better in this regard. My HyperX Impact 2400 is using SK Hynix RAM and does work (even with latency optimization switched on and using CL13 timing).
Probably only CL13 timing os working OK. On all other timings (99% we are talking about CL15) mentioned 3 capacitors fails. They have low parameter so probably they are not able to supply power for CL15 modules for longer time. And those are "early DDR4" - all modules where NUC failed (freezing issue I think). Maybe Intel will create small SW which can test RAMs through those 3 capacitors . If test fails, NUC will be replaced. If it passes test, your NUC is OK (but only with current RAM modules whose are present in NUC). Once you change RAM modules, you should run that SW test once more and see the result.... I don't know, this is only idea...
So based on what you are saying what would be safer, lower or higher CL values? .... I would assume that higher latency values means slower memory, but given the way the NUCs seem to be behaving, is that better or worse ....
In my case the memory in use is: G.SKILL Ripjaws Series 16GB DDR4 2400MHz CL16 1.2V SODIMMs (F4-2400C16S-16GRS) and I have two visually identical modules.
And although it is rated for 2400Mhz, I am allowing the NUC to run it at 2133Mhz .... and would that also be a mitigating factor, using higher spec'd ram ?
So far, I am not having issues, but my NUC is less than 2 weeks old, and I am running bios 039 ..... from 036 (immediately on first boot), from 028 .... I have not had WHEA errors ...... yet ...
I don't know exactly My NUC is also working well, but it's only 1 month old...maybe I get WHEA tomorrow... Overall regading RAM modules - as people say, better is higher frequency than lower CL. In real life is CL13 or CL15 almost unmeasurable difference. I think that for NUC is better lower CL13. I saw somewhere table with timings/speeds, but concretely for Kingston XyperX was quicker CL15 than 13 And because of JEDEC profiles, it doesn't matter if you use 2400MHz RAMs, system sets automaticaly highest possible speed (of course in NUC it's 2133 CL16 for you). If you wanna see all possible profiles, run SiSoft Sandra / System / Motherboard, it shows also these profiles, ideal profile and curently set profile.
But I think that all NUCs are defective and it really doesn't matter which RAM we use. We have to wait for new NUCs distribution
Yeah, I would tend to agree that all of the current skylake i3 and i5's are time bombs, even if ours are working for the moment.
I hope that Intel offers a simple swap for them via a cross ship.
I have a gen 4 NUC that about a year & 1/2 in developed a problem with the SATA connector, I was offered the option of a cross ship (as opposed to ship to them, wait for diag, send replace/replacement) where they charge me for the replacement NUC + $25.00US (to Canada) for expedited cross ship, and send it ASAP where the box is then used with a mailer label to return the defective unit .... and when they get the unit credit my card for the amount charged minus the shipping.
What I would like to see for us here, is we register for the replacement giving all the pertinent info on our model & they ship out the replacement, without charge as it is a manf defect, and then we use the box & mailer label sent to return defective unit. I would even accept the temporary charge on a CC for the replacement, until defective was returned in order to ensure that folks do return the defective ones. But I would not accept a $25.00 charge to facilitate the cross ship, again, as this is a manf defect as opposed to a one off.
I don't think that Intel will make simple swap...defective are all NUCs accross the world ! I think they ask resellers to return current NUCs back a they will change them by new model 1:1 ...and if end user has some problem, he will receive new NUC under warranty process
Well, hopefully mine lasts long enough for them to build the newer revision ......
Yes yes. We have 2 years warranty (or maybe 1 year), we can still get new NUC and new NUC... :-) And after 3rd we can ask for money back or choose different product (for example older but stable 5i3 or something like that...). We can sell DDR4s and buy DDR3s (it'd not big difference in speed in real world). Yeah, I'm not playing nrw games so I don't need powerful GPU... But I hope that Intel will solve this WHEA in this year :-)
Probably only CL13 timing os working OK. On all other timings (99% we are talking about CL15) mentioned 3 capacitors fails.
Not correct. As I wrote, I have two NUCs. One of them is with "standard" Crucial 2133MT/s, CL15 RAM and it works flawlessly. Please stop speculating. The increased capacitance seems to be only needed for some RAM. Most RAM seems to work just fine. In fact I have only read about Corsair RAM failing.
Please note, that this (probably) has nothing to do with WHEA errors / MCEs. They are two different things. The PCN is about making the non-working RAM (early Corsair modules?) work reliably.
Hehe , now I see that in my shop from where I bought NUC, I have 3 years warranty for NUC, 5 years for RAMs, and 10 years for SAMSUNG SSD :-) I don't have to go to this forum anymore....But I want !
You are right. I think too that RAM modules itself are OK, and PCN solves only compatibility for some of RAMs. And this PCN has nothing to do with WHEA
Crucial is Micron RAM. Crucial is a subsidiary of Micron and uses their modules.
I know this but MTA16ATF1G64HZ-2G1 and MTA8ATF51264HZ-2G1 are micron branded oem modules part numbers not Crucial part numbers.
On the Crucial there are the same chips as on the Micron oem? probably because they are a subsidiary, and in your case this is true, but this doesn't mean nothing...it is not this the point, as even the HyperX have on it the exactly the same chips as the Hynix branded that are on the compatibility list.
They are also available as a kit.
Even the Corsair ones on that list are available as a kit but again this means nothing... usually on a compatibility list there is clearly written that a module is tested and compatible in single or dual channel configuration...on that list there isn't this indication.