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
Definetly, you can still find workable 20+ year old hardware on ebay, like SCSI drives.
I think i have a shelf of FIberChannel from NetAPP somewhere, with FiberChannel drives, 450GB at 15k as I remember. We did hook it up to a FC HBA at some point, did a ZFS raid10 on it, and was amazed at how good they behaved, for the 2Gbps link it had to the HBA.
So all in for ServerGrade
OK so now that I unanimously won the Server vs Consumer argument by virtue of stark righteousness, in the complacence of the moment we should sticky this thread, so that posts like this ("Oh, my crappy little Hetzner gaming computer, a.k.a. """""server""""", keeps crashing for absolutely no reason!! Why!!!") can be spared to the readers.
At the end of the day both work.
Advantage of consumer grade = Lower prices. However, when it comes to drives such as SSD's and NVMe's it's best not to cut these corners as a provider or user that needs heavy writes as the Enterprise drives last much longer.
Disadvantage of consumer grade equipment = No IPMI. Hosting servers without IPMI isn't a great way of doing business.
Example a cheap consumer AMD Ryzen mobo might be $129 while a Server board is around $400. That almost $300 price difference in just the board could really cut back on the costs of a server. However not having remote access requires more work from staff.
That being said the typical one is better than another in my mind mostly comes down to some of the extra features that our clients prefer to have.
A CPU might be more consumer but there are still server boards for those systems. Typically more consumer style of "servers" lack features such as higher ram support most max out at 128gb ram other than some Ryzens supporting up to 192. Along with a lack of more PCIe lanes.
That has been disproven already. There are IPMI boards out there but, admittedly, using those with EEC memory is not really consumer grade HW, more like a hybrid.
The point is that, given an "work order" somewhere east you could have whatever hybrid you would like in whatever kind of rack-able way you would like it with whatever HW inside you would want. Given enough spare parts included and an enforced warranty (leasing, maybe), that would last to the limit of 3-5 years before the next cycle of 5-3 nm would arrive. Customers would not expect the stability of a server, but you could chuck in some kind of a video card, for example, at costs vastly inferior to DC types out there.
That would only apply to dedis, any kind of commercial VPS would not be run on that HW, you absolutely need a serious beast for that in order to be able to compete.
You are absolutely correct.
Our new MD series is based on consumer gradeish hardware. It has taken years of development, we tried this first time more than a decade ago. When they worked, they were absolutely brilliant, but it was a mess back then, the tools etc. to create something nice instead of janky didn't just exist and prices were ridiculous. We abandoned it, it was just so much easier to do regular servers and we got a brilliant deal at the time for few racks of nodes to replace the consumer boards.
This MD series now is a culmination of pooling all that knowledge, testing, new tooling, new subcontractors, new hardware available etc. To make it viable, it needs all kinds of stuff.
But now that we got it running, they run absolutely awesome, still some constant things to debug, and each version is slightly better than previous (like it should be), but i think by end of '24 we have the concept nailed down and last of the annoyances solved. We are near enough there that, and confident enough this works at scale, we have right now a stack of 100+ nodes to build at warehouse.
Sample size of one does not prove anything.
We got a management PC which is Core2Quad, rock solid, always ran hot, and didn't even have proper thermal paste for like 5 years. open case, always abused etc.
Does this now prove that PC hardware lasts forever and ever?
While we have servers of similar age which required extensive maintenance over the years due to instability -> Thermal sensor offset was incorrect, and system integrator cheaped out and used cheapest aluminium basic heatsinks where copper higher surface area was specced, combined with the cheap thermal paste + non-temperature durable pins to install them (they snapped) ... Yeah, tons of issues. Finland ran out of 40mm fans in retail stores anywhere near the capital area when we needed hundreds of 40mm fans by yesterday lol.
Nope it does not.
There's just too many factors that go into it, and lot of it is model specific. Other models are brilliant, no matter the target audience, others suck even if targeting the enterprise ultra reliability crowd.
Sorry the argument was already won a few hours ago.
There's no legal way to reopen a closed case now.
Moderators, archive please.
Not disproven at all. Those are server boards for consumer CPU's. There is a big difference there.
Maybe a little misunderstanding or better wording could have been used on my part.
@PulsedMedia
Off topic, in what country are you exactly colocated with infrastructure?
>
Own Data Center
@Calin , Bro, I got that one since they use E5 series ( LGA1366 ) Xeon and mini PC CPU's, I was curios in witch Country.
EDIT: Located with Data-center?, not colocated, my expression was wrong.
Finland
OK THX!
Just to pour some Gas on the fire, as it flamed out:
https://lowendtalk.com/discussion/189365/hetzner-ax52-constantly-crashing#latest
And yes, now we can close
Do not close until a comment is issued by @Francisco .
The one, the first lord of Ryzen servers.
Hardly.
@MikeA is one of the first on ryzen I think.
Francisco
At enterprise scale like your business?
@MikeA any comment on Ryzen as servers?
In many enterprise grade setups, a single server going down won’t impact the whole service
Seems like the fact that redundacy exists is forgotten
My opinions aren't useful, everyone has their own opinion. People like Francisco (or Radic for dedicated servers) would probably be more useful since he's cycled through much more hardware.
People (like me) started using Ryzen in servers right at the start because you could get much better performance for the cost. It's the reason why many smaller VPS hosts started using them, unless density was more of a concern and less worry about cost, then Epyc was always the clear winner (or the god forsaken dual E5's).
I've never had a problem with Ryzen, any "problems" were self inflicted out of stupid mistakes (like not doing proper burn in tests) or BIOS. I've only used the server boards for my Ryzen systems except for one consumer board for game servers right when Ryzen 1000 series released. If you run proper server boards and don't use cheap drives there is really never going to be a problem. Same for ECC RAM. 95%+ of people using common VPS providers would never know the difference and reliability is going to be the same with quality brands. The server boards for Ryzen are "mature" now and the board is the main thing imo that matters. The ones that do worry are likely working in a corporate setting in IT and would be using AWS, Azure, etc. anyway.
I'll probably never go away from Ryzen unless I needed something more dense and then I'd just go for Epyc. But with the newer AM5 server boards support 192GB RAM there's not going to be a need really for me.
Keep in mind many laaarrrgee companies swore off Ryzen and literally laughed at people many years ago when they asked about Ryzen in servers. Now almost all of those big providers are offering Ryzen, the ones that still don't you simply don't hear about anymore.
Helsinki, Finland.
This field is full of people who think they know, and just want the imaginary things, data be screwed. It does not matter what the data says, or what is logical, but if it provides imaginary benefit, you just have to do it like that, doesn't matter if costs increase 10x that's irrelevant for those people. It's religious at times.
Enterprise-grade setups vs hobbyst-grade setups is a completely different discussion than Enterprise-grade HW vs Consumer-grade HW.
Redundancy can be achieved with any kind of hardware, RAID means Redundant Array of INEXPENSIVE Disks which implies it gives redundancy at a low cost. If you want exponentially higher safety, you just keep adding cheap disks/servers at a vastly inferior cost than adding ONE big ass expensive server and arguably having less redundancy and higher risks overall.
I have a RAIS-Redundant Array of Inexpensive Servers with many providers. Arguably, my RAIS is more resilient than any one setup which relies exclusively on one provider, be it Azure or AWS. My costs are also pennies and, while the quality is obviously not the same, the resilience is undoubtedly better.
I'm saying Consumer-grade HW + well thought-out redundancy = Enterprise-grade setup
Edit: Many enterprises obviously has deals with manufacturers, so take it with a pinch of salt, but the above could work for something that "shouldn't go down"
The why big companies spend premium on Enterprise hardware? They rather lay off workers than to use Tenda, TP-LINK, EDIMAX, PC Desktops
The big companies, as said, usually have "their" manufacturer that they use and get support from
Ryzen's etc have previously been "impossible" or close to
With Supermicro starting to offer them, some of them likely will start to use "consumer-grade HW"
OK,
So, take this scenario:
They have been working with DELL/HP for 10+ years. Their internal support team has undergone in-house training for these systems, and outside training to debug them. for the past 5 years of those 10 they gain experience with day-to-day problems and manage to fix them fast. Now, in the next 5 years you say they will start from scratch?
@host_c
What?
No.
Take this scenario:
They have been working with Supermicro for 10+ years. Their internal support team has undergone in-house training for these systems, and outside training to debug them. for the past 5 years of those 10 they gain experience with day-to-day problems and manage to fix them fast. Now, in the next 5 years you say they will start to use Ryzen as it's now offered by Supermicro.
@emgh
Let's see some other opinions, before I respond tot this.
Edit:
End let's make it more general so basic question, do you think that large companies would rather/start use PC based hardware rather thane extremely expensive enterprise grade, alto whatever combination of Ryzen / I9 + supermicro / Asus etc... have same or better performance and better Bang for the Buck spent.
If Supermicro didn't see a real need for Ryzen's in their clientele, they wouldn't have launched the product. The line between "grades" is getting thin
LET( get it LET
) see what others have to say.
Depends on the drawbacks.
If it means dumping their supplier and not getting any support, hell no.
If it means being able to continue working with their existing supplier, keep getting the same support and Supermicro saying that it'll be stable, sure - why not.
If there's good redundancy in place, provided support from their existing supplier is the same, the real question is likely: Will there be any savings after any increased man-hours as a consequence of working with "consumer-grade" HW?
@emgh , do not get so attached to 1 brand. keep an open-mind for others also.
I will give you my answer on this, but I wish to see some other opinions.