256 Cores in a Pizza Box! Half-Depth H242 Gigabyte Overview
Вставка
- Опубліковано 2 кві 2023
- Wendell takes a look at the Gigabyte H242-Z11, a small edge compute server with serious power!
www.gigabyte.com/Enterprise/E...
**********************************
Check us out online at the following places!
bio.link/level1techs
IMPORTANT Any email lacking “level1techs.com” should be ignored and immediately reported to Queries@level1techs.com.
-------------------------------------------------------------------------------------------------------------
Intro and Outro Music By: Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
creativecommons.org/licenses/b... - Наука та технологія
I'd love to see you do an episode or 3 on specing and buying used server equipment. The new stuff is great, but too expensive for a lot of use cases. Especially given the price of used stuff these days.
Just watch out for DELL vendor locked setups ⚠️
If you're curious about pricing, Silicon Mechanics has a configurator. Prices range from $11.5k to $75k for CPU and memory BEFORE adding storage and networking.
Don't understand 90 percent of this but I feel smarter listening to him.
Wendell, we need to talk about the size of US pizza boxes!
It's lilliputian!
That's the key word in the level 1 drinking game. Now we must all finish our drinks.
Love the added Star Trek clip... was waiting for another when you mentioned "assimilating" hadware into Azure :)
I would love to see some practical demonstrations comparing a big memory dual socket 128-core configuration against a 4x32 core hyperconverged cluster. It would be really interesting to see where the cluster performs surprisingly well, and where it falls down.
+++++++++
"There are still people rocking surprising large clusters on dual 10gigabit"
*looks at my little 32 core opteron NAS with dual 1gigabit...* I guess mine is not surprisingly large at least.
How much power is the opteron using?
I would love to see OpenShift running on this. Especially comparing Assisted Installer + Machine Configs vs Tinkerbell
Reminds me of the old SGi/Rackable half depth servers with the i/o in the front. A 2U half depth dual xeon and a half depth 16bay disk shelf was my first homelab. (both heavily modded for noise considerations)
I desperately want to see XCP-ng running on this! Great video as always.
The fans on that coffee table may make conversation a challenge.
Definitely wouldn't mind a quick look at Rancher and Harvester, taking a look at that vs VMware/IBM solutions.
I love this. So many things I could do. Too bad Dell and HP are the only servers on the approved product list and I still had to jump through hoops to get Milan instead of Ice Lake (silvers I think).
The video just hit UA-cam an hour ago and your comment is 4 days old? 😮 Are you a time traveling lizard person?
@@jonas000111 Patreon
@zepesh I prefer time‐traveling, lizard, magician.
@@pkt1213 what they don't know is that patreons also become time‐traveling lizard magicians.
I wish I could be around some of these servers. I want to check them out so bad lol.
Please make a proxmox cluster, and then show us a live node failure by removing an node/ or simulating a component failure!
That would be cool. Good idea!
Looks awesome, but no pricing is available anywhere I've searched thus far. What are we looking at for entry level spec hardware configs pricing wise?
I double dog dare Level1 to build a server in a pizza box.
I wonder which Kubernetes on bare metal OS would suit these best... K3OS, Talos or RancherOS... Perhaps Rancher due to available ZFS storage or Longhorn for workloads storage, and yet kind of simplified OS-level tasks with Kubernetes focused bare-metal deployment. And then run VMs inside Kubernetes, not the other way round so not VMs to comprise Kubernetes cluster... Regardless, if only I won a lottery!
Are nodes the new blade? How does something like compare to those HP Apollo system?
Woot!
Kinda curious what your opinion on Canonical's MAAS (Metal as a Service) and Juju stack line up fairs against Tinkerbell :P
can you run a single program, that needs parallel computing, and get the benefit of all 256 cores working together?
Man what an easy job would be to run sims at home if I had one of these bad boys haha
Please could someone explain the difference in what Floating Point Sixty Four or I n t Eight etc can and cannot do? I'm trying to properly compare a smartphone, Supercomputer, and gaming hardware and understand the difference in computing. Oh, and what is Matrix Floating Point computing?
What's power consumption like?
Yeah Booooiiii. You lost some weight. Im proud.
i was kind of thinking the same thing
What about using it to build a supercomputer cluster?
speaking of the OS to run on those things.... how about TALOS os then? Would that work well / compare favorably or not onto those things? (as a bare metal OS)
I can fit so many PBXs or PAC on this set up.
Nodes with all network cards open is fun with render farm. A lot of power even if AMD think they efficient they not that much mostly SP3 compared to the new ones. Doesn't look like but splitting the heatsinks is better than splitting the CCX with a bit of OC😎
This makes my C6100 really look like the relic that it is...
that is super cool
@13:25 Infrastructure As eXcel? I don't know who'll sue first, MS or Hashi?
ah yes, chat
Indeed. Chat.
How loud is it? :P
I was really hoping this was going to be about the dual processor 128 core Ampere Altra Max... Still interesting though.
Looks almost exactly like a Rubrik Cluster
Hyperconverged Proxmox or XCP-NG would be nice
The Broadcom VmWare deal did not go through yet, and its looking like it never will.
Can I run a Kubernetes datacenter with them, and have DNS server containers to be running off the Kubernetes cluster that is in the edge with these things? geographically this hardware would be in the Philippines for my thought experiment. Philippines is a nightmare for datacenter stuff...
Speaking from personal experience, yeah running a k3s cluster and having an external DNS *with a static IP* external to it works great. Definitely don't be using your DNS to resolve your DNS, that can get a bit iffy
@@IAmPattycakes please explain. i am trying to set up a internal and externally pain membership Paas-SaaS.
Ngl you guys either need a gimbal or to stick to static cam shots. Just my person preference
"Cepth at home" video would be nice
If only I had that kind of money for that stuff😅
Assimilate & Borg reference in one video? 🙂
Usb 5?
I never stop wondering when I see this expensive server racks, can it run Crysis?
Seen Crysis running on a quantum computer would be the day...
Can you imagine if you wanted to put Windows Server on this, what the license cost would be!
I need 4 of those
I could do it with one, fully stocked example maxed RAM.
@@Stopinvadingmyhardware remember if you have just one you have none 😂😉
quoram witness these are require for windows clustering... You don't need a quoram witness for VMware cluster ..even you can run VMware cluster in a single Datastore
"what's their minimum specification" 🎉
I don't love 200ns synchronization latency
It's funny that Oracle is anal about enforcing it's legal policies and yet they themselves do fuck all to comply with FCC rules regarding spam email.
I don't understand how half depth servers are useful or needed, because the rack is still full depth so surely you'll just have empty space at back of your rack? I feel like I am missing something here...
Does it run Crysis?
Costs a bit more than a pizza though. Hell, I could order a pizza from New York, get it flown to the UK for less.
If course, I'd destroy the pizza in 20 minutes, this might last a bit longer.
I wouldn't call 2u a pizza box
family size. 🙄
Chicago-style deep dish :)
They didn't package it with a PSU I hope? 😂
Perfect for a Lack Rack
This is where Gigabyte's focus is. The enterprise. Zero support and backing for their consumer products. They sit on and never return their esupport messages. A big Eff U to consumers.
Render Farm!
Remember when you could chose between PCIe slots with 1 sicket, and 2 CPU sockets without PCIe slots on these 4 node systems?
Just saying gigabyte, maybe take a note out of the Dell CloudEdge C6100 book, drop the full sized PCIe slots for 512 cores in a 2U chassis, you could still have 24 U.2 drives in the front, or... what 64 of those E1.S SSDs
Heck, if you didnt need front IO, you could fit ~72 E1.S drives in the front, make the chassis as long as the monster that was the cloud edge C2100 and you could fit 30+TB E1.L drives in there
oddly the C2100 was only a single node despite being the longest standard server i've ever seen, i have seen non-standard servers that were too wide to fit in a normal rack that were longer, 1U with 4 non-swappable nodes, but the C2100 i bought (used) required a special adjustable 4 post rack and special extended rails for all my other servers, If i had the thing still i'd measure it but there is no way the internet is correct in saying it is 28 inches deep, i remember struggling to get my hands to both ends when lifting it into the recycle so i'd say it is 60+ inches deep
What i'm going to do is measure the depth of that adjustable rack, because i know it did stick out behind the actual posts. If those posts are only 25 inches apart then i'll know i'm just remembering the wrong server.
Great video! I bet this video will not perform very well but I still appreciate it. This is one of the few channels that could answer this highly technical question for me. Is it possible to install a AMD inside sticker on to a Gordon Moore coffin?
I want to see how well it will handle virtio-fs in Proxmox