All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Sneak peek into MD platform testing and development (photos)
PulsedMedia
Member, Patron Provider
Testing of new components, needs some reorganizing but i'll guess good enough for a few sets!
This rack has been mostly emptied from old production and is being turned for only MD platform nodes slowly. 42RU and expecting to put 240 or so at this testing scale on this rack. (Final setup is planned to have ~320 per rack)
This was surprisingly quick design, from scratch to what you see on desk was less than 1hr. Helps to have high speed printers 
Market doesn't have what you need?
BUILD IT.
Rack Shelving and Yes 3D printed from PC-ABS which is fire retardant UL 94 certified (extinguishes itself in few seconds). Testing parts are printing for some destructive stress testing (load test, let's see how many kg can single wall, low infill take!)
We need about 1200 RU of 1RU rack shelving over the next few years, buying commercial option would be something like 60 000 to 75 000€. These we are approximating will cost "only" 14 000€.
Efficiency at every level!
Snap fit is so nice These are the non-connected, tertiary power monitors seen on the rack photo. This is more of a technician tool than actual monitoring, can immediately see if something is wrong, and check long term power usage for this particular set.
and destructive testing via percussive maintenance is always fun evening 
These darned conical off set + size adapting AC duct adapters cost an arm and leg (if you can find one you need), so let's just make them ourselves. Needed to adapt 200mm to 250mm ducting to use larger fan on smaller ducts (higher efficiency == cheaper cooling == lower price for you!)
Cost is about 15€, commercial (if you find right size) can be as high as 500€, sometimes you see someone dumping these for 80-120€ but didn't find the right size.
Second one will have mounting brackets too.
We got 800 x 800 x 1000mm printer coming later this year so we can print our own intake air filter holders, and DC intake/exhaust air vents (Have you ever checked what basic "grilles" cost when you need it in approximately 80 x 200cm size??)
Comments
Fun trivia, on the first photo you see the 8 nodes taking 77W. 2 nodes were actually broken and not in power state mode, so it was over using. Shut those down, and power draw for 6 was about 52W. This is idle, in power save mode as booted into Linux (before that idle draw is higher, only OS can tell the system to go for lower power state than what bios defaults).
6units, 52W idle = 8.67W per unit!
We wanted this power monitoring since we had in production conflicting data, some sets showing only 0.6AMPS so ~141W in production, while others 1.3A or 305.5W. We got ~235V at DC, fluctuates depending on the per phase/wiring/fuse/etc load between ~232 to 240V at use point.
Very interesting to hear these updates as someone with mini dedi. Performance is good, price is good, uptime not as good as some of my other machines but to be expected for an experimental platform.
Tempted to pickup a storage box to go with it actually.
Keep it up!
P.S: Do you ever see this platform supporting 10G ingress & 1G egress?
Still waiting for rdns functionality.
These kinds of posts are so cool, hope it encourages other providers to show off their setups! Props to you man
This is interesting. Didn't realize that those MD units are so small. !!
Thats nice! Whole things looks space and cost efficient.
What "MD" stands for?
Mini Dedicated?
Mind sharing which printer?
They already support 1G egress.
10G we were just yesterday looking at, but it's far off. Those nodes will be more along the lines of 200€+ per month.
2.5G wouldn't cause much changes, apart from cost of the node, something like 75-90€ price range expected.
Issue with bandwidth is; Make it too cheap and you get all the abusers, no one will have a good time and company makes a loss.
Yes started as Mini Dedicated, mITX sized boards... but when you can go for EPYC CPU + bunch of NVMe ... is it mini anymore?
Disclaimer: not tested, but theoretically possible to do that.
We have a plethora, small printing farm. 100x more capacity than we actually need tho.
We have CR-30 belt printers, Elegoo Neptune 4 MAX, a small army of Ender-3s, Sovol SV07Plus, Kingroon KLP1, Creality K1 Max, Tevo Flash, Folger FT-6, and a bunch old less than useful printers gathering dust; Including a original Prusa Mendel.
Sounds interesting, good luck to you
That's why we are putting so much effort on these!
Our mission is to make dedis with high performance accessible to everyone. Anyone who has a need or want for a private dedicated server should be able to get one!
In theory, eventually we can make dedis which have only ~2.50€/Month operating cost (+support!) ... You can easily see that might lead to sub 10€/Month final price in a decade as that hardware depreciates to nearly worthless.
We are hoping to massively over produce these eventually, with every config we can come up with, so there's a node fitting every possible need precisely.
We think this as a factory, make everything streamlined, mass produced and as quick as possible to manage, as little human effort involved as possible. This philosophy will be visible on every level.
Usually companies avoid such amount of choice -> Analysis Paralysis.
We'll probably fight that with having just highlight servers on top (Best value picks), and a click away for all the models.
We already have 18 different models, and we are just "testing and fooling around" level still.
This is despite concentrating on 1 hardware model, just 3 different generations of.
Any advice on preventing EMI from those long power cables?
I've been having all sorts of EMI issues from long power cables affecting similarly long data lines (HDMI) and possibly even laptops in the same room.
Fixed the HDMI issue with an expensive shielded cable as an AmazonBasics shielded one didn't cut it even though the power and data cables are routed on opposite sides of the room (antenna effect perhaps?).
Pavin.
This is awesome. Following the thread!
Ground plane.
The whole platform is grounded, and the systems are electrically connected to the ground plane.
Only cables there are DC Power + Network.
We have not noticed any EMI issues etc. but are worried over power ripple.
Need to find qualified electrical engineer willing to spend that time measuring the ripple to double check tho.
However, if we find ripple it's "easy fix" with filtering PCB.
We'll do some testing since we've had issues with Realtek NICs having randomly piss poor ingress performance, which logically seems absurd (higher latency might be faster, 5ms might bottleneck where 200ms is line rate etc etc.). Cuts average ingress performance to about half, but worst of all; some are absurdly slow and some are full line rate.
That might be an EMI issue.
But we found that issue also on a system not mounted to a plane and "far away" from EMI sources.
You can also use ferrite rings to lower EMI, this is both cheap and easy; such as: https://www.aliexpress.com/item/1005006245855842.html or https://www.aliexpress.com/item/1005006174055731.html
Considering you are having issues with even laptops in the same room, you might actually have power line issues, power being "dirty" or something bigger at play there.
Only time we've actually seen EMI related issues is with those stupid electronic locks, which still have physical key, but zero power / self powered (twisting of key powers it) which reads the RFID (presumably) on the key and opens the lock.
Those are the most stupid invention there is, they don't work even without EMI issues.
Thank you, lots of great info in there.
The laptop power supply has those cylinder looking things which google says are ferrite rings, so I think I'm covered there. I don't think the power being dirty could be a major factor, even if there were some voltage drop due to an aircon compressor turning on, this being a laptop with both mains and battery should be able to handle that just fine. A double conversion UPS would drive me insane with its constant humming noise.
Do you think those DC power cables are themselves shielded and grounded?
Or perhaps your whole setup being a metal rack plus the chunky metal ground plate acts as a Faraday cage.
The Realtek issue appears to be something with the board itself since it's happening when both grounded and not grounded. Can't imagine errant charges building up even when grounded.
Lots of respect for you hardware guys, I will go bald in a day doing it.
Pavin.
So DC in question? Since large one. But then you already got noise.
Small ones don't make any noise.
No shielding, without ground the system would not function. Need to close the loop.
and racks itself being grounded too, the ground moving from the plate to the rack, and also safety earthed on top. and DC- is on the ground plane too (standard industrial practice).
Tbh, i have never seen any EMI issues, i think i once suspected network cable crosstalk. Apart from those stupid locks that is, but they are bad enough that even in a faraday cage they probably fail 30% of the time.
I would suspect the chips and/or linux drivers. I think it was 8111 Rev 15 what those have, lspci doesn't show precisely the model.
Oh well, M.2 2.5G NICs to replace then, got many models en route or just arrived. time to design and make mounts etc.
Thanks!
One could have never guessed the years spent "hot rodding" (building, modifying, racing) cars, welding, fabricating etc. would come this handy in day to day datacenter ops.
Another business owner in the industry just said "I've heard of full stack, but this is something else ... deep stack!"
Somehow slowly things just keep getting vertically integrated.
Slight language barrier here, probs local saying meaning "go crazy" on the "will go bald" part?
Not in a DC, I work from home exclusively after the pandemic and a bad case of pneumonia. These are my personal machines
Perhaps depending on the brand, we have a 2kVA one that has its fans running all the time probably due to the 24/7 AC-DC-AC conversion.
Got it, no shielding for the power cables but everything else is in a metal cabinet and grounded.
If the issue with my laptop is in fact EMI (no scope to test, resorting to trial and error), I guess it could be because the plastic body is not grounded and it has some metallic parts like a heat shield that could build up errant charges.
Interesting background and "deep stack" is a fun way to describe the specialization 😂
Pulls hair out in frustration, no hair... 🥲
Pavin.
Bad efficiency. Most UPSs suck, even enterprise scale.
Modern, good efficiency UPS will waste only a few % -> No fans required.
You can also buy "power conditioners", they are more like protection devices, may contain a capacitor inside (another lengthy topic, caps + AC)
If the issue with my laptop is in fact EMI (no scope to test, resorting to trial and error), I guess it could be because the plastic body is not grounded and it has some metallic parts like a heat shield that could build up errant charges.
It is grounded via the DC- pole, otherwise it would not power on.
Typically they go to great pains to ensure minimal EMI, since no signals would get through otherwise. Especially PCI-E signals are very very sensitive. Longer the PCB traces the more EMI gathered. DRAM is probably 100x more sensitive than PCI-E too.
Issue you are facing, is most likely something completely other than EMI tho.
Wow, another nugget of useful information which lines up with my own experience with this laptop. Its iGPU uses a reserved portion of normal RAM (8G stock, soldered-in) as its VRAM which lead to frequent kernel panics on adding a 16G module (1 free slot), problem went away after switching to an 8G module.
I tend to lean towards the EMI theory because there were severe RAM issues on an older gaming laptop that I used prior to this, memtest showed its RAM had lots of bad addresses even after a hard reset. A few days later I checked with a new stick of RAM and it too showed issues. Concluded at that time the soldered-in RAM had gone bad but a few weeks later I checked and it was alright. So strange, that laptop is working fine now after some cleanup and re-pasting everything. At the time of the issue I was operating it with its back cover off due to heating issues, maybe it was dust and not EMI 🤔
I could bug you with silly questions all day LOL.
My current troubles are less frequent but I can't get the kernel to dump (LMDE 6, downstream distro woes). Will wipe and install Suse and see what the kdump has to say.
P.S. Sorry for hijacking your thread!
Pavin.
hmm, zyltech vs elegoo black -- quite a big difference. (still printing AC ducts, need 4 or 5 of these adapters)
That memory module was faulty or you were running it too fast. Desktop ram doesn't have ECC so errors can go through when running too fast, this is exactly what "faulty vram" looks like, it's memory corruption. Just little enough and your whole system won't crash immediately ...
See above. Faulty module most likely.
If either laptop is Zen 1, Zen 1+ or Zen2, those had quite a bit of memory issues. I think Zen 3 finally got all the rough edges smoothed out.
Writing this on a Zen1 which has DDR4-3200 but because i use 4 modules i had to drop it all the way down to DDR4-2600. 3200 was non bootable, windows couldn't load. 3000 got to login screen once, 2800 looked like functioning but bizarre crash and finally 2600 is working. I always jumped 2 steps down, borderline stable is no use for me.
Temperature. RAM is super sensitive for temperature.
We have one Gen1 Epyc server, i think we might've bought it before Zen2 launch even; It has this type of RAM: from cold boot it will generate a lot of ECC errors, same if it's been idle for days and you start to push it, a lot of errors.
High load constantly? Not a single issue. RAM has heated up, or the traces to RAM, or the pads to CPU which are responsible for RAM. Something has.
I think it was on 1 or 2 modules.
Only effect if it was full metal cover, most laptops these days have plastic covers. That doesn't affect EMI at all.
Also motherboard is one gigantic groundplane -- for this reason. The motherboard has insane amounts of filtering going on, and that motherboard would generate more EMI than anythign else on your room at that distance.
Not sure, but i think EMI, like any magnetic field, weakens at the rate of square root (2x distance, 1/4th strength), BUT do not quote me on this; I have nfi what i'm talking about.
@PulsedMedia
Have you seen this? Strip it down and make a custom snap-in tray that mount it vertically would make great density. With PoE it would only require 1 RJ45 each.
What is that?
Sorry, I forgot to add it's the new MINISFORUM S100-N100. Not yet sold, it's have an N100 and 2.5Gbps PoE, in a thin and long form perfect for clustering. I'll be keeping an eye on when it's available as I will need a small test cluster soon(TM).
Interesting. Couldn't find dimensions for it tho. Guesstimating 3 of them would fit into space of mITX.
The MS01 is very interesting to us, but will present some power challenges.
There's a lot of variables to which matter, and right now we only concentrate on very narrow band of models to get everything else sorted out. Supporting odd form factors would be very much side quest stuff.
Good news, just double checked; Helsinki record temperature ever was 31.6C
https://en.ilmatieteenlaitos.fi/seasons-in-finland
This is measured from the shade, you see your temperature gauge going above this many many times, but actual air temperature record is 31.6C in Helsinki.
This means our new DC might not actually even need any AC.
Even HDDs are 100% fine for upto 40C ambient (server intake temp) according to google research. Solid state even higher, CPUs, RAM etc.
Looks sexy...following it up...
and this is what the flanged hvac ducting adapter looks finished. ~10€ a piece in materials.
There's some restocking;