New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Today's AliExpress Roulette
(sans copper nvme heatsinks)
Some stuff already arriving for testing this round, surprisingly fast. One neesd hundreds of odd little pieces here and there, and volume too, just a few is not enough.
for picopsu's angle adapters ordered couple models for testing. Who knows how many weeks until we get to test and finish the first prototype set.
Good luck, this looks like a fun project that will have many benefits.
How are you handling putting them in a rack, are you hoping you can fit 2-4 units in 1U of space? Or do you have a custom 'blade' type enclosure planned?
The N100 is a crazy efficient CPU do you see an option to have an upgrade N150 version too?
Milk V Duo S test completed
cheap relay boards controllable, USB works so we can use the expensive pre-existing units.
The USB Power adapters are jank tho, supposed to be 24V (so with tolerance ~29V) but Overvoltage protection kicks in at ~19V. So no good!
Well, we'll dispatch the first set with POE Power Splitters probably then.
We'll use these to "upgrade" (more like sidegrade) the current power control. We are using a rat's nest of USB wires and hubs. That was deemed too unreliable, shame it would've been cost effective. But with these the cost is ~same (for the final version), but we get some local logic! ... got some ideas for the final version
For now tho, this is all added cost.
Soon off to design the software architecture on this.
Thinking all identical SD cards (these are with emmc, so a bit slow to flash ...), DHCP and MAC as UUID, so the SD card can just be replaced when it inevitably dies in a few years.
No local configs etc. all fetched remotely after boot, and all sensor data dumped over the network so conserve the write cycles on the SD Cards, and log rotate like hell to force the SD card to change blocks (well EXT4 extents), no journaling etc.
Before that tho, need to design the enclosures and decide how to retrofit the old units, there's no room really, so ... we might have to mount the first units on rack instead of the mplate, use external PSU Etc. so we can upgrade them to these without any downtime.
Sad thing is, all these will be then doing is log temperature + control the relays, using an expensive relay board.
BUT, that stops the flakyness of the current method (one system controlling few dozen relay boards) and is fast.
Current BOM for the sidegrades:
Total ~35€ // ~4.4€ per node
Not too shabby, cost saved back in a few months by less headaches. The most expensive part is probably POE switch rack space consumption.
Thanks! Yea it is, and quite demanding. When you deploy a lot, you do not have room for weakness. everything needs to be robust.
The rack mount has already been on previous photos, 2x per 1RU on rails.
We'll later add similar form factor higher end CPU options too, but we'll start here. N100 efficiency means lowest thermals, cheapest price to work out the kinks on HW etc. easiest to make work in other words, once we know the thermals and power consumption precisely we know how much room we got.
In theory we can put some Ryzen X3D CPUs in there tho ....
The power meters we use have test pads and chip has UART .... Wonder what we'll find from those points
Would be awesome if we can tap into those and collect the data from AC side for logging.
I want to buy a PT server, but your unlimited traffic server is too expensive.
Its not that efficient because most manufacture tend to have those N100's idling at 10 a 11W. When they can actually idle at 4W. Same with the load, where your often seeing jumps up to 30W, when proper n100 implementation does 18W. That is rather rare!
Its rather powerful for the price your paying, for this level of device. But there are now competitors like the AMD 5700U or 5825U's, as in older laptop CPUs that are now being dumped at cheap prices by AMD.
When you compare that to a AMD 5825U, that also idles at 4.5W, and with the same 30W power budget, kicks out 3x the performance... Anyway, its a different discussion.
What is PT server?
Oh you are talking about those 10G nodes? Yea true 10G costs quite a bit.
Yea wondering how the power consumption metrics for these will turn out!
We only measured with the worst PicoPSU -- which happesn to be original "official" picopsu, bougth from them direct like 12 years ago (tech advances) 1.42A @ 12V on these motherboards at worst. We'll get actual numbers don't the line.
AMD E350 was insanely efficient too, we had gigabyte boards around 12 years ago. Well, actually still got a few left over.
That info is actually very interesting, i'm kinda AMD fanboy and own a lot of their stock, but always choose for what works best and that "fanboyism" only gets so far (the ZEN MiniDedi lineup, only few dozen ever got made!)
Been wondering about those motherboards too, for example; https://www.aliexpress.com/item/1005008706625310.html
I'll try to find some videos or something about those. We could set those up. Anyone up for another presale?
The N100 tho is super cheap and the encoders are gold for those doing video transcoding/streaming. Both N100 and AMD 5700U/5825U are good!
That's why we want to setup a plethora of options and huge amounts of different configs, there are different needs for everything. We have NFI how to price them tho, so hence the auction pricing. Speaking of which, just noticed some crazy low prices as we put so much at once available.
Oh while we want to try all kinds of stuff, we do have a backlog of like 300 more of the Prodesk G3 lineup that needs to be put online :O
This 6xHDD thing is a bit sidestep taking time from that production, but we need these cranking out ASAP too, as we have some plans for this type of setups.
Right now it takes ~8hrs per mPlate from production start to online and inventoried. This does not include all the infrastructure work to make the rackspace etc. available.
~4hrs is the mechanical assembly, ~4hrs is QA, racking, cabling, documenting etc.
Skilled coder and interested? Contact us, we need more hands on the coding side.
PT websites, such as this website iptorrents.com, I installed vertex and qBittorrent on a server with a large hard disk and bandwidth to perform flow and seed preservation tasks, and I am engaged in upgrading these websites.
We have designed and decided how we deploy all the various distros. Some basic code has already been done.
We will most likely open source everything, including the tools so people can make their own templates.
Here's the basic how (which we can accomplish now for Debian/Ubuntu immediately for the MD 4TB models, with just 15mins of work for all!): Just make a local VM which can be accessed as root over ssh.
That's it. done
We will add some scripts, tools etc. you just rsync/untar into / later on.
But that's the basic gist of it.
The idea is that people can easily make their own templates and clone them. So, say if you need 15 web servers for your site, you just install one, and launch the templating script and host the template somewhere over HTTP.
Same with any Linux distro.
We'll figure out *BSD and Windows later on, those have to be deployed as images i think (That too is actually kind of trivial ...)
Presale Units; Waiting for parts to finish the first prototypes and rack them.
New DC got the first grid feed installed today!
Meanwhile, mass producing some of the MD 7th/8th gen i5 series. Mostly 32G RAM + 2TB or 4TB NVMe.
And this is why you need pressure on glue jigs, even tho straight from printer, parts aren't exactly straight, and the mPlate bends too etc. you need pressure when you glue things up. See the light under the parts. (Also new jigs are printing,. still using temporary ones. All printers on 24/7 100% utilization ... until they break lol):
some angle iron and clamps will take care of this better than just stacking weight on these
New even better NVMe heatsink glue jig bottom plate test, 5mm thick, ~100% fill and didn't remain straight. Also notice the print issues? This is Sovol SV08, shame it's too close to a Voron, you need to calibrate etc. it. It's not like Creality or Bambus, take out of the box and keep on printing. Also Elegoo Centauri Carbon is very nice, slight calibration issues tho, but same kind as K1 Max had originally; Z offset too close, they are probably afraid of blob of death; But also centauri carbon has the best handling of blob of death i've seen, even better than K1 Max.
Anyways, next version is not 100% fill, it's 10mm thick and has some slotting to make certain it remains flatter by default. This will have top piece to hold the heatsinks and leverage is added straight on the heatsinks.
Also speaking of heatsinks sigh, i think we need to make our own molds and just get the profiles extruded locally. Then again, listen to podcast and unwrap these. Next to the heatsinks is 500x USB Type A to C 25cm cables.
and all sorted
this is why you need many printers. Not quite satisfied with the SV08s. Each unit has had issues, we bought on the launch event 6 of them. This unit keeps losing it's Z-offset now. That's now 3rd hotend in row with only 1 succesfull test print done.
Think going to scrap this, take a few spare parts (like the custom SSR which keeps failing on these!) and bin the rest.
Our needs are very bursty tho, sometimes these machines sit ~idle for 2months straight, then 24/7 utilization for a month straight.
Ironically the most reliable printers we've ever had by print hours was Ender-3s, but by Kilos consumed the Sovol SV07Plus. But SV07+ try to get bed level, i dare ya. It will stay level for a few moments, then you battle hours. The ABL on it is worth jack shit no matter what you configure, it will take out maybe 0.1mm of deviance. Maybe a buggy version of klipper?
Also the SV07+ with our tuned profile is faster than any of these CoreXY machines with default profiles. ~13min 30second benchy, mechanically stock machine.
Anyways, despite fully working, retiring it for that constant bed leveling headache. Maybe it's just a faulty unit, who knows. It was awesome tho for the price, over 50kg of filament has gone through that machine alone and it still works. Time to give it to someone else
Future per mPlate control boards, these will be monitoring environmentals, controlling power etc.
they are so tiny.
Any PM-A possible?
I am craving to buy one but I want something really cheap u know 
PM-A?
Just like KS-A [$5] from OVH... PM-A from PulseMedia?
https://eco.ovhcloud.com/en/kimsufi/ks-a/ this?
we already make a better offer 🤷🤷
the $5/m one [change location and see the magic
]
oh showed ~30$ for me, i see under location.
There is a reason it's always sold out .... As in it's not real apart from the few samples/ loss leaders. The electricity alone costs more.
yup, I get it...it's quite rare and bots are the one who generally grab them
[I wasn't totally serious when I said it, but its not always bad to ask right? 
]
BTW you're doing great work and I do tend to read it sometimes when I pass by and see the thread above
Do you have a plans to make few nodes now or in future with more ram and less storage? For customers like me that use it not storing much or even anything.
Less storage - a little less price :]
Something like:
N100
32/64GB
1xNVMe(small for Os)
Same possibilities for network bandwidth upgrades.
N100 only officially supports 16GB tho, Google says the people that got 32GB running are the minority
not a "NAS" in that setup, but yes, N100 setups with variety of configs are coming, no worries about that
We'll probably do the same as with the previous models; Make a lot of different models, and see which are in demand and then make more of those models.
True, these boards here what we have booted with 48GB. Didn't test to OS, but it's pretty much certain it works when it even posts and goes to bios
We'll be trying other models with 48G too.
Which begs the question, which is the closest to N100 / i3-n305 level but dual memory channel ... Could try 96GB
which then begs the question, which mITX sized system can run with the most amount of RAM ...
Hi,
i just hope that this whole alibaba custom stuff wont burn down your rack/shelf/datacenter/world
But the idea is in general great, except that i do not really understand the business/price calculation behind that, but thats of course something you are to decide.
In any way, enjoy your DIY project. Really cool!
i still remember their retard migrate,no start date no end date,didn't tell is migrate finished or not.and only gives 3days to migrate(although migrate by them,I don't need to do anything),after the old server complete destroy,I found not all my files has been migrated,so I lost ton of data.they say next time ask for migrate process eariler
one time I sent a htop screenshot in Discord, and a stupid staff pointed at the main process and subthreads in htop, threatening me saying I was running multiple torrent software simultaneously, and wanted to ban my seedbox
imagine maintaining arrogance toward unfamiliar matters, and not apologizing even when others point out the actual situation.
Most electronics comes from China anyways
Fire safety was a huge concern, and we took it seriously. We’ve got staggered safeguards throughout. The PSUs for example has over temperature protection to cut off, all 3D Printed parts are either black or white for fire resistance etc.
Most of the stuff we got is brandname stuff from the likes of HP, AsRock, Dell etc. no more riskier than any other DC.
Ironically, many "enterprise-grade" setups have mind-bending single points of failure. Risk is everywhere, we just acknowledge it and engineer around it.
I would say we know few things about safety and reliability more than the average junior tech working at enterprise DCs these days.
We built our first "stacks" of mITX servers back in 2012-2013. Damn those were janky, but rock solid performers too. They were literal stacks on threaded rods + M6 extension nuts as spacers. They worked solid, but lack of internal storage was their main issue, we used network storage for them back then and that storage was not stable and reliable. Partly because of the ST3000D00M series of HDDs.
The business side of things ... Volume and Scale:) It will take us a bit of time to get there, but these run already profitable, but time to recoup the R&D costs is mind boggling until we get to scale up. This here is just one of the R&D steps we have to take.
Node profit is solid now; at full density it prints money. Layer on top automation etc. and it's easy to see these will eventually be the main profit driver for Pulsed Media. Not this 6bay N100 itself, but this type of setup.
I don't see others running nodes in this density, anyone else seen?
So your data was migrated, but you failed to double check?
If an server fails, like it sounds it was because you were migrated, all bets are off. Backups are always your responsibility. We merely do the best we can within the budget and setup possible.
Like everyone, you were told about the migration, probably an extra month service time to boot, probably given instructions to double check etc.
Depends on the situation which template was sent to you.
We do not promise backups of your data or data safety, and if you were on one of the V1000/V10G series which is RAID0 ... With zero data redundancy.
If you got only 3 days, it was emergency migration due to failing hardware; and we still tried to recover your data. Can most providers say that?
It probably failed during migration. We cannot help if HW fails, and compared to many big players; At least we attempted datarecovery for you.
Migrations almost never happens unless a server is failing. I do not remember any migration in a while which did not involve hardware failure.
If you lost data, that sucks. But you got to do your own part. Backups are your responsibility ALWAYS. Data loss always sucks. But if your only copy is on a €7 box with RAID0... you’re gambling. And we all know how that ends.
One htop screenshot and wild claims? Probably misread threads vs forks. But let's be clear:
Abuse the infra or the staff? We're not your host. Plenty of summerhosts out there.
Respect the metal, respect the humans -- and you’ll love it here.
🧪 R&D chaos, precision engineering, and profits measured in rack density.
Welcome to the lab.
Made a thing. Guess size and Purpose.
While waiting for parts for these NAS units to arrive so we can finish the first ones, built quite a few regular MD units, new glue up jigs etc.
OrangeStorm Giga toolhead had a bad day with first a blob of death, then with fire.
Got a bunch of notifications from freight, so hopefully we got all the accessories we need by next week.
Also found 90degree PicoPSU units in Ali; https://www.aliexpress.com/item/1005004160983520.html
Ordered 4 to test them out. This is why so much stuff comes from there, this kind of specialty stuff is Unobtainium in the west. And even if you find it, probably comes from the same factory.
Parts are starting to trickle for the 6x HDD versions. Hopefully we get first light next week! Maybe even first delivery within 3 weeks.
Meanwhile ... readying a batch of 64 or 72 regular MDs ...
Mounted 40 today on mPlates and finished first of 5 wiring.
Hi, is it possible to have the 2.5gbps addon on the MD series?
Thanks!
That's a hardware thing, it either has 2.5G or it doesn't.