ASRock X570D4I-2T

Their XMP rating is trash and basically fake news, I got a partial refund on amazon due to that crap (priced down to what it actually hit)

1 Like

I was afraid of that. Although from the BIOS screenshot in @dandanio’s post earlier in this thread it does look like at least some people might have been able to get their 2933 Nemix kit to hit its rated speed. I did poke around the overclocking menus a bit and did not see 2933MHz as an option (only 2900MHz and 2950MHz) so I’m not sure how I would even test running at 2933MHz. Perhaps @dandanio would have some suggestions.

I bought my kit direct from Nemix to save a few dollars vs Amazon. The 2400MHz kit goes for $11 less than the 2933MHz kit on their website. Not sure how open they would be to a refund for me but I don’t think I’ll go through the trouble for a Hamilton and a Washington.

2 Likes

Its one thing to not hit the rated speed, but to have issues being in dual channel, thats a bit much, probably try to do an exchange

yeah seems like 32gb ecc are not a thing, any reason to not go 4x16 (you need more memory then that?)

That’s a really good point. As much as I’d like to put the machine into service I’m going try running in dual channel mode again and if the problems continue to crop up I’ll start an RMA.

As for why I went with 2x32 instead of 4x16, I was just trying to build for the future. Given my current plans for the server, I don’t think I need 128GB of memory right now but thought it would be nice to be halfway there if I were to decide it is necessary down the line rather than needing to buy 4 new sticks.

I moved the memory back into A1+B1. Interestingly, dmidecode is now reporting that the kit is running at its rated speed!!

Handle 0x0016, DMI type 17, 84 bytes
Memory Device
Array Handle: 0x000C
Error Information Handle: 0x0015
Total Width: 128 bits
Data Width: 64 bits
Size: 32 GB
Form Factor: SODIMM
Set: None
Locator: DIMM 1
Bank Locator: P0 CHANNEL A
Type: DDR4
Type Detail: Synchronous Unbuffered (Unregistered)
Speed: 2933 MT/s

Rank: 2
Configured Memory Speed: 2933 MT/s
Minimum Voltage: 1.2 V
Maximum Voltage: 1.2 V
Configured Voltage: 1.2 V
Memory Technology: DRAM
Memory Operating Mode Capability: Volatile memory

Non-Volatile Size: None
Volatile Size: 32 GB
Cache Size: None
Logical Size: None

Handle 0x001B, DMI type 17, 84 bytes
Memory Device
Array Handle: 0x000C
Error Information Handle: 0x001A
Total Width: 128 bits
Data Width: 64 bits
Size: 32 GB
Form Factor: SODIMM
Set: None
Locator: DIMM 1
Bank Locator: P0 CHANNEL B
Type: DDR4
Type Detail: Synchronous Unbuffered (Unregistered)
Speed: 2933 MT/s

Rank: 2
Configured Memory Speed: 2933 MT/s
Minimum Voltage: 1.2 V
Maximum Voltage: 1.2 V
Configured Voltage: 1.2 V
Memory Technology: DRAM
Memory Operating Mode Capability: Volatile memory

Non-Volatile Size: None
Volatile Size: 32 GB

I’ve initiated the same operation that originally triggered the ECC errors and will monitor the system for the next day or two and post a final update here for those who might be interested.

I don’t want to turn this discussion about the X570D4I-2T into a thread about debugging my Nemix RAM but if folks have more questions on this aspect of my system feel free to DM me.

1 Like

Nemix RAM Update: After a day running at 2933MHz in dual channel mode the frequent MCEs have returned. I tried manually lowering the frequency down to 2400MHz in the BIOS while still in dual channel mode and the machine wouldn’t even POST. I’ve emailed Nemix and initiated an RMA. Here are a few snippets of the machine check events in case there is anyone here who has more in-depth experience with ECC who is interested in the specific errors I’ve experienced with this kit:

Sep  8 11:07:25 myserver kernel: mce: [Hardware Error]: Machine check events logged
Sep  8 11:07:25 myserver kernel: [Hardware Error]: Corrected error, no action required.
Sep  8 11:07:25 myserver kernel: [Hardware Error]: CPU:0 (17:71:0) MC18_STATUS[-|CE|MiscV|-|AddrV|-|-|SyndV|-|CECC]: 0x9c2040000000011b
Sep  8 11:07:25 myserver kernel: [Hardware Error]: Error Addr: 0x00000007ba156880
Sep  8 11:07:25 myserver kernel: [Hardware Error]: IPID: 0x0000009600150f00, Syndrome: 0x182f00400a800902
Sep  8 11:07:25 myserver kernel: [Hardware Error]: Unified Memory Controller Extended Error Code: 0
Sep  8 11:07:25 myserver kernel: [Hardware Error]: Unified Memory Controller Error: DRAM ECC error.
Sep  8 11:07:25 myserver kernel: [Hardware Error]: cache level: L3/GEN, tx: GEN, mem-tx: RD    

Sep  8 11:12:53 myserver kernel: mce: [Hardware Error]: Machine check events logged
Sep  8 11:12:53 myserver kernel: [Hardware Error]: Corrected error, no action required.
Sep  8 11:12:53 myserver kernel: [Hardware Error]: CPU:0 (17:71:0) MC18_STATUS[-|CE|MiscV|-|AddrV|-|-|SyndV|-|CECC]: 0x9c2040000000011b
Sep  8 11:12:53 myserver kernel: [Hardware Error]: Error Addr: 0x00000007b8823d00
Sep  8 11:12:53 myserver kernel: [Hardware Error]: IPID: 0x0000009600150f00, Syndrome: 0x182f00400a800903
Sep  8 11:12:53 myserver kernel: [Hardware Error]: Unified Memory Controller Extended Error Code: 0
Sep  8 11:12:53 myserver kernel: [Hardware Error]: Unified Memory Controller Error: DRAM ECC error.
Sep  8 11:12:53 myserver kernel: [Hardware Error]: cache level: L3/GEN, tx: GEN, mem-tx: RD

so it does down clock to 2400mhz? Kington has traditionally always had an honored their liftime warranty.

ah so maybe it isn’t the board or cpu but, crappy ram?

i usually do Kingston ram unless i can get tighter timings for more money from a reputable brand. i don’t really see much 2666mhz ECC sodimm with less than CL19

i wonder if an APU overrides the BMC on board GPU with the iGPU??? i was lookign at the GE series with it’s supposed 35 watts but, it’s got an vega onboard.

Yes, I suspect it was bad RAM. I packed it up and sent it back today.

As described above, the kit ran at 2400MHz by default when in single channel mode (with no errors) and 2933MHz by default when in dual channel mode (with frequent errors). I wasn’t able to get my system to POST when running the memory in dual channel mode at 2400MHz but I’ve never underclocked memory before and didn’t try tweaking any settings other than the frequency itself; someone with more expertise might have better luck.

I started out looking for sticks from the top brands (Crucial, Kingston, etc) but sadly couldn’t find any x32GB ECC SODIMMs in stock from anyone aside from Nemix.

1 Like

Just a thought on the issues with this board. A buyer comment on Newegg brought up a good point.

!!!PSA #2!!!
They use a 24pin to 4 pin motherboard power and you NEED to have a powersupply that can at least do 40 amps on the +12v to make the motherboard boots! that will leave a lot of lower powered SFX/ Flex ATX power supply out, a 850W power-supply can usually supply 70A on +12V, but a 400W can maybe do 30, the lowest I tested are 500W @ 24A+24A and it boots, a 400W 28A on the +12V fails to boot but IPMI is able to run.

I’m current running a 3950x in this boards with 4x 32GB Nemix ECC Dimms (can’t get ECC to work but that’s another issue). I’ve been running into random kernel panicks that are always memory related. I ran MemTest86 and it would always eventually fail during the first pass at stock 2666mhz or even fail during the 2nd or 3rd pass even at 2133Mhz. When it would fail, it would fail spectacularly, throwing hundreds of errors instantly. I am currently Memtesting each stick individually, but I think my issue may actually be PSU related. I’m using a questionable 500w Athena Flex ATX PSU, as I thought that would be plenty with no GPU in the system. But it turns out that this PSU has 2 seperate 12v rails at only 20A each. That’s only 240W on each rail, which I could see a 3950x pulling for than that at peak wattage. Plus it’s a no-name band anyways. the other reason I’m suspecting PSU is the issue is that it seems to crash guarenteed during any moment of dirty power, like during any bad weather. While no other machines go down when this happens, this one always does. So it seems like it’s just on the cusp, and if the PSU is worked too hard, a memory bit flips and the system crashes. I’ve ordered an 850W PSU with a single 70A 12v Rail, and hoping that will fix my issue.

1 Like

what ram sticks are you running?

So got the mobo in, i will have to piece meal this. bought use a U-NAS on ebay while back for WAAAYYYYY cheaper than new. wish i had bought a few!

please forgive the mess and crappy angles and focus.

I’ll be hard pressed to get a REAL raid card in here for my Raid 10 or 5. maybe an external PSU? but ,the post above says you need a 40amp PSU!!! (ffuuuuuuu)

That’s just what a reviewer on Newegg said. I’m using Nemix 128GB (4x32) M474A4G43MB1-CTD. That guy couldn’t get his board to boot, but mine boots just fine. It’s just the huge memory instability that is annoying.

look over the last 10 replies, the Nemix ram seems to be of poor quality. at least others here are having issue with the Nemix RAM. also make sure it’s installed properly the slot layout is side by side not every other slot.

All 4 DIMM’s are populated so the layout is irrelevant. I know it’s poor quality but I’m passed any sort of return to Newegg and this memory was very expensive. So I’m going to have to pin down if any 1 stick is bad.

1 Like

when the NVMe is installed i assume i drops the PCIe form x16 ot x8? or do certain CPUs cause it go to x4?

Checking the block diagram on page 11 of the manual from the asrock site (can’t link), looks like the NVMe is driven directly by PCIe lanes off of the CPU.

1 Like

I’ve been lurking on this thread for a few months and I want to say a huge thank you to the people who have shared information on their experience with this motherboard and particularly those who shared detailed build information. It helped me a lot with my own build.

Like many others I wanted to build a small headless homelab server + future NAS, and the selection of this specific motherboard was primarily driven by the fact that I needed to fit everything into a 450mm depth network cabinet I have installed in my garage. This unfortunate past decision meant I basically had to use mITX if I wanted a Matisse CPU. I chose AMD Zen2 for PCIe 4 support and this motherboard also allows me to drop in Zen3 in the future. I don’t need a graphics card and so the PCIe slot will likely house an ASUS Hyper M.2 x16 Gen 4 Card or similar in the future when the 2nd gen PCIe 4 NVMe drives come out. Built-in 10GbE with the X550 is great because the X550 has a long track record in enterprise servers and good Linux support. The IPMI w/ graphics means I never needed to have a graphics card to get up and running and can use KVM if I really screw something up.

I piggybacked off many other’s part choices from this thread once I had settled on the motherboard.

Case: Fractal Node 304
Motherboard: ASRock Rack X570D4I-2T
CPU: Ryzen 5 3600
Cooler: Noctua NH-L9i
RAM: Nemix 64GB Kit 2x32GB DDR4-2933 PC4-23400 ECC SODIMM 2Rx8
NVMe SSD: WD Black SN750 500GB
PSU: EVGA SuperNOVA 650 GA
OS: Ubuntu

WD Black chosen because it was relatively cheap and will hold me over until the better PCIe 4 SSD come out. No HDD for now, but the case can fit up to 6x 3.5” drives. I’ll probably grab some Seagate Exos 8TB later this year.

Thanks to those who provided detailed information on the challenges mounting CPU coolers on this motherboard. I did buy some M3x20mm screws ahead of time but it turned out these are actually too long for the NH-L9i. When I tried to install them they ended up hitting the heatsink on the cooler while still protruding from the backplate. So I ran out to my local small hardware store and came back with M3x10mm screws (all they had were 10 or 20). These then had the opposite problem of not really threading all the way through the feet of the cooler so I felt like it wasn’t tightly attached, but still decently secure. As kcpants noted there is a gap between the feet of the cooler and the motherboard, which I just let the screw go across without any support. Regardless, I was kind of impatient and just wanted to boot the thing up, so I left it as is. The node 304 case is pretty tight inside, but since I had removed all of the HDD sleds and planned to keep them out for now, it wasn’t too bad and I could mess with the cooler again without too much disassembly.

Running the system like this, I noticed CPU idle temps around 70C which is too high, so I went to a different store and got some M3x14mm screws and swapped them for the 10mm ones. These are the best size for this cooler/MB combination and it seems to have helped as idle temps are now around 45-50C.

Running prime95 it still pegs at 95C so I think I need to totally re-do the cooler mounting solution at some point. Since that torture test is pretty much worst case scenario, I’m not too worried for now. Also, me and my family are the only users of this system, so I can take it out of service whenever I want. I’ll probably fix this when I start adding in more heat generating components like HDD and a PCIe card.

I too have the Nemix memory kit others have mentioned. I’m running them in the A1 + B1 configuration at 2933. I did run 5x passes of memtester without any mce showing up and nothing in the logs otherwise so far, so I’m gonna leave it as-is and keep an eye on it. Perhaps they will show up later.

Thanks again for all the helpful information and discussion on the thread!

3 Likes

what kind of CPUs are you guys rocking with the NH-L9i and what kind of fire shoots out of the back? i do not see a low wattage 6 core

EDIT: CHEAPER THAN Nemix junk!

https://www.provantage.com/kingston-technology-ksm26sed8-16me~7KINM25W.htm

grabbed 2 for $186 including tax and shipping which is cheaper than nemix kit! might have been the same price on the egg or mazon if either hadn’t charged shipping or tax. but, hey at least provtanage is likley an approved seller, the egg has this kit by 3rd party sellers.

PSA: neobits has Samsung M474A4G43MB1-CTD 32GB ECC SO-DIMMs for sale at a reasonable price.

3 Likes

Hi,
Got this board with 4x16 kingston 2400 ecc, 3700x and a 480 gb kingston datacenter nvme and 2x16tb ironwolf pros. Building a box to replace my current truenas box (3400g, 16gb non ecc, and 2x6tb wd red).
Want to use proxmox with one ubuntu vm and truenas. How much ram should I give to truenas? The 2 ironwolfs will be used only for truenas. Truenas has 2 plugins only transmission and plex. I am thinkning to give 8 cores and 32-48gb (?) ram to truenas.
What do you think?
Thanks

1 Like