Mini PCs with 128 GB of RAM in Home Lab with new RAM kit!

preview_player
Показать описание
The time has come! 128 GB RAM kits have been officially released from Crucial and this is going to be a game changer for home lab enthusiasts running mini PCs in their environments. We have been stuck at 64, then 96 GB of memory for years now. Finally, we are getting to workstation levels of memory for powerful mini PCs that will help take home lab and home server environments to the next level for self-hosting with virtual machines, Docker, and Kubernetes.

List of compatible mini PCs that work with new 128 GB RAM kit:

Proxmox server build components list (amazon affiliate links I get a small commission):
BD795M:

Huge news with 128 GB of RAM now! - 0:00
Memory is the key in home lab - 0:53
Previous cap from 64 is now 128 GB - 1:15
Specs of the new 128 GB memory kit - 1:50
Compatibility - 2:49
Hurdle of seeing official support - 3:20
Testing and playing around with these new 128 GB RAM kits - 4:00
Models I have tested with the 128 GB RAM kit - 4:27
Minisforum MS-01 running 128 GB RAM - 4:57
ASUS NUC 14 Pro running Proxmox and 128 GB RAM - 6:13
Looking at the free -m output - 6:36
Trigkey S7 Pro with 128 GB of RAM - 7:20
Why this is a huge development for home labs - 8:12
Using this in conjunction with VMware NVMe memory tiering or Optane Swap in Proxmox - 8:45
Getting closer to having the level of enterprise server resources in min PCs - 9:08
Possibilities of upcoming hardware - 9:34
Let me know what you think and if you are excited about this development - 10:23
Subscribe and get updates! 10:42
Рекомендации по теме
Комментарии
Автор

Nice to see them work for the MS-01. Not-so-patiently waiting for the MS-A2's as well. I was just about to buy a 96GB for either my minisforum UM773 lite or the Precision 3260. Thanks for the info!

terryjohnson
Автор

Thanks for the 128GB Ram update, I bought a MS-01 after seeing your channel and added 96GB RAM, but I am now waiting for the Minisforum MS-A2 to add to my home lab setup, so will definitely consider adding 128GB RAM to that machine, thanks again for your very useful videos.

CraigMcIntosh
Автор

Three things:

1) Pricing
Your Amazon affliate link shows this 2x64GB Crucial kit at $365. The Trigkey S7 Pro, for example, starts at like $469. So that will push the total cost, per node, to $834.

In other words, whilst this is great that the 2x64 GB DDR5 SODIMM kit is now available, it gets quite expensive to actually deploy.

2) Homelab RAM usage
It REALLY depends on what you're doing and how.

VMs typically take more RAM than LXCs.

And if you're able to run your stuff with Docker containers inside a LXC, then you can jam more stuff into the relatively limited amount of RAM that you have vs. running a bunch of VMs to do the same thing.

(I had originally migrated away from using VMs unless I absolutely had to, in favour of LXCs. And now, thanks to Kasm Workspaces, I am starting the process of migrating what I can from LXCs to Docker containers by way of Kasm Workspaces WITHOUT running Kasm Workspace. (or rather, Kasm Workspace is optional because by default, Kasm Workspace wants you to carve out RAM for the Docker containers, if you go through their interface, but if you just use normal `sudo docker run...`, then you can skip carving out RAM explicitly, and let it dynamically grow/shrink as the container needs to, for RAM usage.)

I'm ALMOST a the point where I can have like a thin client open Firefox automatically on boot, and then I will just connect to the Kasm/Firefox or Kasm/Ubuntu-jammy-desktop docker container, and just do everything that I need from within that environment.

ALMOST.

3) Upgrading RAM
I just bought 16 more sticks of DDR4-2400 ECC Reg RAM for my "do-it-all" Proxmox server, as I was running into 231 GB out of the 256 GB installed limits.

(Hence why I am now trying to "downsize" the RAM profiles of the stuff that I am running.)

But, I'm also going up to 768 GB of total RAM installed (when the modules arrive), so I am very much looking forward to that.

(I actually already have 7 server compute nodes where each node has 128 GB of RAM, but they take up way more power than the Mini PCs.)

But yes, RAM is now my biggest limiting factor.

ewenchan
Автор

Excellent Video! I will be putting this new kit in my Incus servers.

scottibyte
Автор

Highly highly request that you throw the 64GB SODIMMs into any Alder Lake-N based mini PCs you have laying around.
I've been running a 48 GB SODIMM with zero issues in both my N100 and N305 based Beelinks for a long time now. Verified in BIOS, Windows, and Proxmox.

ReQuiem_
Автор

Just because the BIOS detects the hardware you need to be sure it can actually utilize it reliably. I would run it through a few series of memtest to make sure it can do all the normal sequential and random pattern bit read/write operations to every address register without error. If your BIOS is set to fast boot, it skips memory counting and reads the hardware's reported values and takes it on faith.

eltreum
Автор

Sure that’s a huge step forward to use 128 gigs. I use my system for audio production and now the setup becomes very handy. 💻

lordasriel
Автор

Excellent presentation. Thank you. I along with many others, cant wait to try these in an MS-A2, maybe with a bifurcation card on the x16 slot. The crazy thing is that a 3-4 unit cluster of MS-A2's running Proxmox, filled with x4 NVMe drives and 128g of ram with dual 10GbE - could run everything at 90% of the businesses in country. Would really like to see you test a 3-4 node, with bonded 10GbE, of them with Ceph all on a 10GbE switch to see how disk performance is with VM's. I would really like to know if Ceph performance improves with bonded 10GbE on a fast box like these.

markmonroe
Автор

Need to see this plus the other nvme ram stats together

kristopherleslie
Автор

can you test by configuring vms to run and address/consume the additional RAM? There is a huge difference between it being recognized, and it being addressed...you might just find corruption and errors.

druxpack
Автор

I saw a bunch of posts regarding the NUC requiring BIOS updates in order to have more than 64GB RAM. Have you had any issues with that?
I currently have a NUC 13 Pro with 64GB, but if i can properly use 128GB i'll definitely get the 14 pro as well.

MrSpamcho
Автор

That is very good news, also people can try KSM

ertanerbek
Автор

Nice to see the extra capacity, I wonder if these work in the MS-01 ?

Mat--jv
Автор

For home lab should I go with the ms-01 or the ms-a2 and can I add 25gb network card to it

arturorubio-wm
Автор

Looks nice, but after major dataloss I don't use non-ECC memory in my servers.

StepanTezyunichev
Автор

Is ECC support not necessary for a home lab virt env? Is there even a current model mini pc with ECC supported memory? I would love to build a small (3 device) cluster, but the lack of ECC support is off putting.

fahrersicht
Автор

Pro:0, con: £343.73 -double con: Usually dispatched within 1 to 3 months-
DDR5 RAM 48GB (1x48GB) £103.99 available delivery tomorrow.

paulmaydaynight
Автор

I wonder if N100/150/300/etc support it. Not sure if it's worth it but N100 supports 96gb so makes me wonder.

tbard
Автор

I changed xmp profile 1 from disabled.. Cause my ram is 5600mhz and it was only using 4800mhz.. After changing that xmp my ram was still limited to 4800mhz cause in memory section there is "PCH limit set to 4800" it's limiting ram speed xmp setting.. What do i need to do i tried many options in OC but that PCH limit isn't changing.? No single info or video on YouTube..

jaidart
Автор

Yeah, I ran unofficial 32 gig on an N100 and there were memory corruption issues. Same sticks in another PC that officially supports 32 gig and no problem. So I’d be real shy of sticking 128 gig into something that doesn’t officially support it because you might be getting silent corruption. Remember this stuff is not ECC. I was getting these random hangs on certain software setups, very sporadic, ran memtest86 and sure enough errors started to show up. Yes it’s not mission critical stuff if it’s for a home lab but still you want it to work.

volfactory
join shbcf.ru