RAM failing on new 6port router from AliExpress

[kevdog]

I just wanted to run something by the forum member here as I'm slightly confused. I recently purchased this new 6 port 2.5Gb Topton Like Router from AliExpress. I unfortunately only got around to testing the unit yesterday. Included was a single DIM 32 Gb Samsung RAM Module (Notice this RAM was included and this is simply an Amazon link showing the appropriate part). I ran a single pass Mem86 testing of the RAM with to bottom of the unit open and the RAM was noted to fail during the Test #13 Hammer Test.
I contacted the AliExpress vendor regarding the issue and surprisingly they did respond in less than 24 hours. Their response was my test was done under ideal condition. They indicated the RAM which reach 52C was a possible reason for the RAM failure and asked me to re-run the test over using a fan or other cooling mechanism. Indeed 52C is quite hot.

Is their suggestion to test the RAM with use of a fan legitimate? The unit was designed to be fanless with no included FAN in the chasis. I ran the test with the bottom of the unit off allowing for likely more airflow than if the unit was completely closed as designed. Is it likely the unit would ever really reach 52C under normal operating conditions (virtualized pfSense and Debian server)? This seems like either a case of bad RAM or actually poor thermal design.

***Note -- some may wonder why I haven't actually tested the RAM with a cooling fan over the unit. I unfortunately need to wait till next week to do this do to travel however it's just an issue I've been mulling over in my head.

 

[jei]

52C for RAM is totally fine.

 

[necr]

3.2 DMA Test (Test 14)
Errors The DMA test involves executing DMA-initiated memory transfers between the scratch partition of the MemTest86 USB flash drive and the system memory. As this introduces an additional component to the data flow, errors detected by MemTest86 may not be an issue with the memory or DMA subsystem, but with the USB flash drive itself. If MemTest86 detects errors during the DMA Test, there is a possibility of a fault in the USB flash drive. In such cases, it is recommended to re-run the tests using another known reliable USB flash drive. If the errors persist, there may still be a possibility of a UEFI bug with the device driver DMA implementation. This is usually indicated by a sudden increase in the number of errors. If you suspect this may be the case, please see Troubleshooting MemTest86 Problems.

^from the MemTest manual. Look for test #13, there's some interesting info there. https://www.memtest86.com/downloads/MemTest86_User_Guide_UEFI.pdf

Overall, from my experience, DDR5 is way more picky - have you tried relaxing timers? Does this tiny box do memory training during POST?

 

[WANg]

I just wanted to run something by the forum member here as I'm slightly confused. I recently purchased this new 6 port 2.5Gb Topton Like Router from AliExpress. I unfortunately only got around to testing the unit yesterday. Included was a single DIM 32 Gb Samsung RAM Module (Notice this RAM was included and this is simply an Amazon link showing the appropriate part). I ran a single pass Mem86 testing of the RAM with to bottom of the unit open and the RAM was noted to fail during the Test #13 Hammer Test. View attachment 35750
I contacted the AliExpress vendor regarding the issue and surprisingly they did respond in less than 24 hours. Their response was my test was done under ideal condition. They indicated the RAM which reach 52C was a possible reason for the RAM failure and asked me to re-run the test over using a fan or other cooling mechanism. Indeed 52C is quite hot.

Is their suggestion to test the RAM with use of a fan legitimate? The unit was designed to be fanless with no included FAN in the chasis. I ran the test with the bottom of the unit off allowing for likely more airflow than if the unit was completely closed as designed. Is it likely the unit would ever really reach 52C under normal operating conditions (virtualized pfSense and Debian server)? This seems like either a case of bad RAM or actually poor thermal design.

***Note -- some may wonder why I haven't actually tested the RAM with a cooling fan over the unit. I unfortunately need to wait till next week to do this do to travel however it's just an issue I've been mulling over in my head.

Eh, RAM shouldn't give you issues unless it's substantially above 70 C for an extended period of time. 52 C is slightly high as than operating temp, but should otherwise be still fine as the chips on the SODIMM module should be able to handle temps from 0 to 85 C.

Given the fact that this is meant as part of a passively cooled router there should be no expectation of using it in conjunction with a fan. Is there any thermal compound or a metallic block making contact with the SODIMM unit acting as a heatsink during normal operations?

 

[kevdog]

Eh, RAM shouldn't give you issues unless it's substantially above 70 C for an extended period of time. 52 C is slightly high as than operating temp, but should otherwise be still fine as the chips on the SODIMM module should be able to handle temps from 0 to 85 C.

Given the fact that this is meant as part of a passively cooled router there should be no expectation of using it in conjunction with a fan. Is there any thermal compound or a metallic block making contact with the SODIMM unit acting as a heatsink during normal operations?

No there is no heatsink in this unit. I was planning on putting a fan ontop of the unit during normal operation, however I was testing the RAM actually with the bottom of the case open. I usually always test RAM before installing the hypervisor software.

 

[Stereodude]

You could try clearing the CMOS so it will retrain the memory and see if that helps. You can also try a fan as a diagnostic step, but it should pass without a fan. However, they probably included a low quality memory module. You won't know without trying another DDR5 SODIMM (if retraining doesn't help) whether it's the memory or something else.

The general recommendation I've seen is to buy these barebones and buy your own quality RAM and SSD.

 

[kevdog]

^from the MemTest manual. Look for test #13, there's some interesting info there. https://www.memtest86.com/downloads/MemTest86_User_Guide_UEFI.pdf

Overall, from my experience, DDR5 is way more picky - have you tried relaxing timers? Does this tiny box do memory training during POST?

Relaxing Timers -- I'm not exactly sure how to do this. I went into my memtest86 test and found the following timings:



I'm looking at my BIOS however I don't really see an option how to adjust RAM timing (I admit I've never done this before):


Is there anything else I should be looking for?

 

[necr]

Doesn't seem like this basic UEFI will give you any advanced settings like timers. Anything you can change in CPU configuration that's related to memory? From the DDR5 standard, base frequency is 4000MHz, might be possible to set that in this UEFI.

But also, just please take a look at the test description again - it could be that you're trying hard to fix a problem that's not a huge issue, these tests are named experimental for a reason. I would test with stress-ng and a live OS, when in doubt. And if you're still unsure, just buy another DIMM (and return it if nothing changes).

 

[kevdog]

@necr - I'm posting some screenshots within the Bios in regards to CPU configuration, however unless you see something I don't see anything to adjust the timings. If you could take a look that would be great.

I looked at the test description in regards to errors on the Hammer Test: PassMark MemTest86 - Memory Diagnostic Tool - Troublingshooting Memory Errors. Interesting read however in the end it didn't say much stating -- is this error important to you?? -- ignore it -- and test with new RAM. I'll try contacting Ali Express in regards to the issue since I don't know what other RAM manufacturers or models would be equivalent. I don't have a spec sheet regarding the MB. The article also said try ECC RAM -- which I'm not opposed to -- just need to see if this is supported and what model I would need.

In terms of stress-ng -- let me look that up. Plan is to install a hypervisor - either xcp-ng or Proxmox and go from there. Intended use of this box is virtualized pfSense router.

 

[WANg]

@necr - I'm posting some screenshots within the Bios in regards to CPU configuration, however unless you see something I don't see anything to adjust the timings. If you could take a look that would be great.

I looked at the test description in regards to errors on the Hammer Test: PassMark MemTest86 - Memory Diagnostic Tool - Troublingshooting Memory Errors. Interesting read however in the end it didn't say much stating -- is this error important to you?? -- ignore it -- and test with new RAM. I'll try contacting Ali Express in regards to the issue since I don't know what other RAM manufacturers or models would be equivalent. I don't have a spec sheet regarding the MB. The article also said try ECC RAM -- which I'm not opposed to -- just need to see if this is supported and what model I would need.

In terms of stress-ng -- let me look that up. Plan is to install a hypervisor - either xcp-ng or Proxmox and go from there. Intended use of this box is virtualized pfSense router.

Eh, the SODIMM looks like run-of-the-mill stuff shipped with DDR5 notebooks nowadays - DDR5 4800 CL40 is pretty typical for Ryzen Rembrandt and Phoenix machines, or newer Intel 'Lake DDR5 machines. The interesting thing is that Intel tends to gimp the memory limit on Atom based machines in the past (in fact, for some reason the memory limit on the i3-N305 is said to be 16GB from Intel's own spec sheet), so I am slightly surprised that the memory controller play well with the 32GB stick out of the box. My guess is that since it's a Alder Lake die with only E-Cores (which is essentially Atom), it probably has the mainline Alder memory controller instead of whatever Intel ships with their Atom/embedded line. In terms of the memory test...eeeh, unless it fails on the mainline tests, I would not worry too much about it. I used to have DDR4 RAM freak out on me on memtest86 and would still be reliable using stress-ng during 12 hour core-and-heat-soak tests. if the RAM is truly bad it'll manifest its faults pretty fast anyways.