Question Thoughts on this nemix ram?

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
I noticed this on the nemix website after seeing the 2933 on newegg.
Apparently it has hynix mem chips in a 9 (cause of ecc) on each side configuration.
It claims to be 2Rx8 just like my 16GB with hynix chips and the same 9 chips per side.

The odd thing is the chips in the picture clearly say "AFR", I'm guessing it's the identical chips and manufacturing process ?
Maybe each chip has 2 dies in them, and just a higher speed binning ?
At first I thought maybe the picture was wrong, and it would have 19 chips per side...
This appears to be it, just no mention of "AFR" ... D-die ?

This is the first time i've seen 32GB dimms in this configuration for ecc udimm that are readily available.
These would be perfect for a ryzen build, and cost 1/2 the price of the hard to find samsung dimms.
Still 21nm hynix stuff ?

Here are some PDFs i've collected on the subject
 
Last edited:

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
From the older 16GB configuration:

# of Row Addressing Bits: 16
# of Column Addressing Bits: 10
# of Banks: 16
# of Ranks: 2
Device Width in Bits: 8
Bus Width in Bits: 64
Module Voltage: 1.2V
DRAM Stepping: 255
DRAM Manufacture: SK Hynix
SDRAM Package Type: Monolithic, 1 die, Single load stack

You think the new stuff is 32 Banks ?
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
I found some Micron stuff
This appears to be it, listed as "Die Revision B" instead of "E" like the micron 16GB modules ...
You think this is 20nm ?
So i'm uncertain if that micron is 20nm or if it's older/newer than the E-die used in 16GB modules

Looks like the 32GB modules are starting to hit the market now.
You guys think those micron chips are better than the hynix chips ?
 
Last edited:

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
From my question above:
# of Row Addressing Bits: 16
This should double going to 32GB modules, the banks/ranks/Column will stay the same.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
After running this mem for a full day I noticed some ecc errors, so I rebooted and ran some memtest86.

verified 1 dimm is bad (same addresses over and over), 3 dimms are good.
I tried speeds down to 2933 with the same results. Also I tried different dimm slots with the same results.
Not sure how I missed this in my initial memtest86 run, I was in a hurry and only did 1 pass (can't even remember). I was too eager to boot into the os and get on with my life I guess...

bummer ...
So now I guess I have a headache as it was purchased in a kit. Hopefully they'll let me just return/replace the single bad dimm and not all 4.
 
Last edited:

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
I just initiated a return/refund. They aren't responding to my email and I found a competitor product with the same chips for 20% cheaper.
 

Ryan Koller

Junior Member
Apr 26, 2017
7
3
51
I just received this Nemix 128GB kit. Amazon has delayed my 3970X delivery until April 21st unfortunately so I won't be able to test it until then. As it's essentially the same module that you've ordered, I'm interested in what you find to be optimal overclock settings. I'd like to get it to 3600 MHz with as tight of timings as possible with as little voltage as needed to do that.

I'm using the system as a virtual machine host with multiple user desktop VM's using multiple GPUs with VFIO passthrough. I wanted the ECC for this use case for stability. I'm hoping I haven't given up too much speed to get it.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
I returned the sticks for a full refund. They had dropped the price 20% ... and the crucial memory from shopblt was 30% cheaper for the same specs/chips.

I ordered the shopblt memory, but the ETA was changed to over 2.5 months, so I had to cancel it.

I did do some extensive timings testing on the nemix before I returned it, it seemed like 3400MHz was optimal with tight timings.

These settings seemed to be what I was dialing in:

One stick was bad(at any settings), and another stick wouldn't do the 4-4-4 rrd_s/rrd_l/wtr_s
tCKE needed to be 10, 8 was unstable.
Most the timings were worse, NS wise, compared to the hynix stuff at 3000MHz with the exception of the rrd/wtr and rfc
With ryzen the 3400 speed would be way better because of the infinity fabric, even if timings were higher.
 
Last edited:

Ryan Koller

Junior Member
Apr 26, 2017
7
3
51
The rest of my stuff finally is being delivered today. My plan has always been to take it from 1.2V to 1.35, increase it to 3600 and then drop my CAS timings as low as I can get them. I'm just hoping I don't have any bad modules like you did. I'll keep you posted.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
To be fair, nemix did offer to memtest86 them and send good modules back to me. Unfortunately they didn't return my phone calls or emails untill I had already initiated the RMA and sent them back to them. At that point prices had already dropped 20% and I was out shipping both ways.
I guess the lifetime warranty and replacement is their biggest perk.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
also I think these micron chips are ~17nm instead of 20nm+ like many of the older stuff, that's why i'd be hesitant to move the voltages over 1.35v.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
I had read that micron e-die has multiple batches. Some are 19nm and some 16nm? Which may actually be 20nm and 17nm.

That could be so, I only know the TechInsights imaged/measured the 8Gb chip (1Y) and said they were 17nm. The 16Gb could very well be a little smaller.
They also have 1x and 1xs
I think all these 16Gb ones just now hitting the market are atleast 17nm, manufacture dates of basically right now.

There might be some older "dual die" packages using the 20nm stuff on some server dimms. I dunno.
 

Ryan Koller

Junior Member
Apr 26, 2017
7
3
51
So memtest86 doesn't recognize the memory as ECC. It shows up as ECC enabled using info64. I ran it through two full tests of memtest without any errors, but if the ECC is fixing things I wouldn't know it. I've got to dig around to find another way to test it.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
Only the payed version of memtest86 recognizes the ecc (which I purchaed).
In linux you can run anything (ie: compiling code) and then use edac-utils -vr to report anything, dmesg will also show the errors.

These nemix dimms (and most ecc dimms), also have temperature sensors on them. These are also exposed in linux.
 
Last edited:

Ryan Koller

Junior Member
Apr 26, 2017
7
3
51
According to the memtest86 website, ECC detection should be included in the free version. Only ecc error injection requires the paid version. As mine doesn't even detect ecc, I'm hesitant to purchase it. I'm not very confident that general use will saturate enough of the memory to really test it. Compiling something like chromium likely isn't going to use all 128GB of memory. I'd rather run a memory test that gives me some confidence that everything checks out. I'll contact the memtest86 guys and see what's up with the TRX40 and 3970X in regards to ECC detection.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
Yes, you're right "injection" support is the pro feature.

It's not going to detect ecc if you don't actually have ecc mem installed.
 

Ryan Koller

Junior Member
Apr 26, 2017
7
3
51
Yes, you're right "injection" support is the pro feature.

It's not going to detect ecc if you don't actually have ecc mem installed.

I have the ECC Nemix chips installed; it still doesn't detect it. I've confirmed it the best I can using info64 and dmidecode -t memory that ecc is enabled. memtest86 still doesn't detect it.