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
I noticed that my CPU has been changed from a 9275F to a 9475F.
Can you transfer it to me? I need such a configuration
update: well things are falling apart again. My last update didn't age well.
Disks and pings all over the place again and - this is new - STEALtime skyrocketing on some machines, while "just" going up on others.
At first I thought this is some SICK form of recreational YABSing on the weekend, but it still persists well into the new week. I have a very moderate usage and still notice the breaks.
idk what people are doing on those machines, but they definitely need more. sic
Ah, and BTW it seems like cores are getting put to sleep pretty aggressively, which is weird.
I had a feeling this was gonna end up too oversold when I noticed they were taking preorders via support ticket.
It's a pity that feeling seems to have become a reality.
Maybe they got unlucky and attracted the "wrong" kind of customers with that - if there's such a thing. I ordered with tons of headroom for usage (basically 10x) and am still getting squeezed out
Same here, but I opened a ticket and Berohost explained that they're currently performing some migrations during which the nodes temporarily have more guests than intended. My Steal Time has improved as of today (0 - 0,x)
Perhaps they could post that kind of information on the status page.
Didn't get the mail ether. Anyways I was still cautious, so not many customers on that cluster, but man things were going smooth before this anniversary surprise and the move to the new dc. Thats why I ordered - not that I was looking around for a new hobby.
bero ryzen 4cpu 12gb ram (12EUR/mo) review with yabs:
console is hard to use, spend 20+ minutes to figure out how to install proxmox
then spend another 20+ to install usual ubuntu back (it didnt work)
so spend another 10 to understand that it might be os related and tried debian 13
it did work, but only after I realized that I need to check new generated pass TWICE :P
as first didnt work
also got like 6-7 mails that I didnt requested in that hour
network is poor comparing to hetzner, and iops as well, only CPU is better
(on par with arm)
## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ##
Yet-Another-Bench-Script
v2026-10-07
https://github.com/masonr/yet-another-bench-script
## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ##
Tue Oct 20 13:27:23 UTC 2026
Basic System Information:
Uptime : 0 days, 0 hours, 3 minutes
Processor : AMD Ryzen 9 9950X 16-Core Processor
CPU cores : 4 @ 4291.948 MHz
AES-NI : ✔ Enabled
VM-x/AMD-V : ✔ Enabled
RAM : 11.7 GiB
Swap : 0.0 KiB
Disk : 98.3 GiB
Distro : Debian GNU/Linux 13 (trixie)
Kernel : 6.12.38+deb13-amd64
VM Type : KVM
IPv4/IPv6 : ✔ Online / ✔ Online
IPv6 Network Information:
ISP : ROETH \u0026 BECK GbR
ASN : AS44486 synlinq.de
Host : Rene Roeth trading as ROETH \u0026 BECK GbR
Location : Gelnhausen, Hesse (HE)
Country : Germany
fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/sda1):
iperf3 Network Speed Tests (IPv4):
iperf3 Network Speed Tests (IPv6):
Geekbench 6 Benchmark Test:
Test | Value
|
Single Core | 2486
Multi Core | 6720
Full Test | https://browser.geekbench.com/v6/cpu/19367534
YABS completed in 12 min 6 sec
@AleksZ to be fair afaik hetzner has become prohibitively expensive and a lot of chicken are jumping ship, picking around for something like bero.
They just had quite a fire sale, so maybe waiting on a bunch of machines to spread out the load a bit.
Of course the results are not great right now (see my comments above).
delete
up/up/update: low pings are possible:
@berohost The more I look into the ping issues, the more I get the impression that there are some serious configuration problems with the hosts:
Usually pings between servers hover now between 0.3 to 0.5ms which is unfortunate for machines next rack - this is more similar to inter data center traffic inside Frankfurt.
BUT when I put some serious load between them, pings actually DROP to pretty stable 0.15ms !!!
Sometimes I even see 0.12ms which is more like the default on the older cluster (~0.14 ms) and shows one thing: the hardware (network/hosts) is pretty much capable of doing so !
By serious load I don't mean some simple iperf test, but something that puts all cores to work full send.
Unfortunately I can't do that all the time, since my real world traffic is rather low.
It seems to me that the hosts are too aggressively tuned towards putting the cores to sleep for energy saving (which is ironic when I see steal times >10% on some of the vms;). Maybe you could check/prevent higher c-states in bios/proxmox and see if it helps. I get it that energy is expensive in Germany, but cmon.
Additionally on tcp dumps I see quite some unnecessary discovery traffic with lots of ARP triplets which hints to some classic lacp/circular routing reflection misconfig & lack of clear trunk separation. Yes, I've tried the new vlan feature but no change (just less arp but still triplets) - pretty cool btw.
Again no ticket, since this is not impacting just machine A or B but every host. As far as I can see my machines have been moved around several times now between hosts anyways.
Maybe you could have a look into this.
edit: I would actually pay for an upgrade to a small set of pristine machines with much less traffic and way much more restrictive cpu rules - same hardware, same switches, just a separate (sub)net - you have it all there already - just charge more for less traffic/cpu.
This is what I mean:
While the world is spinning towards the weekend and another round of recreational yabsing is gearing up, I am at my new hobby: bero maxing
@bero has actually put some amazing hardware together - it's just a pity that it doesn't live up to its true potential (yet)
So it can really ping. Everybody with 2 machines can try it. It's a bit of walking the line. As described in the last post, one needs to entertain the cpus without overloading them. So for example you can put Postgres on one machine and an API on the other and try to load each core to about 50% +. This way the host really lights up as many cores as your machines have and voila, you get the true pings that the hardware is capable of
)
I ran this on my 2 best machines with the clearest path - its definitely 2 different hosts.
It also works on a more janky machine with pings <0.15ms but with hard spikes due to host overload (steal time anybody?
This combo is a beast and could easily do even better when configured accordingly, instead of forcing it artificially to perform well.
And that could be had essentially for free - well maybe at the expense of some time for tuning the config and perhaps a couple of extra watts.
Almost no natural load could create these pings right now, so fingers crossed that bero will actually put some effort in.
There are also some interesting findings regarding the net - I've been throwing some dumps at Claude to see where the journey with the LAN party goes.
Maybe I'll drop a separate post on that later.
And no, I am not applying for a job at bero - this is strictly a hobby;) , although I am overall a fan of what they have put together and achieved so far.
Hopefully things will work out fine.
Happy yabsing
Interesting stuff. I only have one box at bero (Turin, Anniversary M), so I tried a poor man's version and pinged the gateway instead of a second VM:
ping, 1s interval (30 pings):
rtt min/avg/max = 0.842 / 2.914 / 11.847 ms
ping -i 0.01 (500 pings):
rtt min/avg/max = 0.711 / 2.792 / 18.472 ms
next level analysis. thanks. all let member should like this.
The 2 machines need to be able to talk directly to each other without a hop through a router - like same subnet/direct mapping/vlan - and you need some real world load between them that wakes up all the cores assigned to your vps on the host, so not just iperf3 or ping, which the hypervisor can easily handle on a portion of a single real cpu thread. -- Thats the problem, that most of the server is dormant and not ready to pick up your requests instantly (even if you see steal time under load).
Makes sense, thanks for explaining. So the router was adding its own delay, and a ping is too small to wake up the server. With only one box I can't test it properly then. Will keep following.
Consider me a total newbie, but do you actually run the API on one box and the DB on the other in production?
yes, I run the DB and a failover replica on separate servers then the APIs.
It's considered a good practice for production - separation of concerns (botched updates/ fails etc..) - on a single server its rather binary: either it works or it doesn't. And I hate pager duty at 3 a.m.:)
Thus I am so concerned about internal pings and tail latency.
.. on a single server I would rather use sqlite or turso and call it a day
Not sure if this is the right thread to ask, but it's been 20 days or so and I'm still on the fence about moving my SaaS from GreenCloud to Bero. The YABS/GB6 variation makes me hesitate.
Do you actually host production on Bero? I'm fairly new to this market and don't know much about them yet, so I'd value hearing from someone who does.
@ahson
... I've described my experience in several posts in this thread. In essence the TL;DR is: pre anniversary top notch, post not so happy. So for my customers in eu-cenrtal its a mixed bag right now.
also routing to other services in frankfurt got worse in my opinion - not sure how that happened.
Maybe some more lucky guys could chime in with their story ..
@berohost
Puh sorry, meant @berohost of course in the ping screenshot post before
Nothing has changed in the routing or network equipment compared to before the promotion. As mentioned earlier: Please open a ticket. We can only help you if you open a ticket.
@berohost thx! Did you have a look at the screenshot I posted earlier & do you have any ideas about that?
To me it looks like you are trying to force the provider 'fix' an issue that doesn't affect most of the customers just because it's not fit for your use case.
Perhaps look for a new provider if you think 1 ms extra latency cause issues for your setup.
Create a fucking ticket, put the screenshots instead of pinging provider in this thread and see if they can fix it or not. How hard can that be?
shouldn't one consider that ICMP is not reliable for this kind of analysis at all, because QoS and such.
Hi, the same drop in latency is also measurable inside the application inside the established tcp connection and correlates with the icmp results (its just more stable). This is actually how I found out about it during an integration test.
A ping series print is just more easy for demo purposes.
Did you create the ticket?
Hi @barbaros,
yes I get it that most people put everything on a single machine and call it a day. And thats fine for quite some projects. Have been there, know it.
Just asked the other way around - what is particularly wrong with my modus operandi separating DB, failover and API on different machines? This is pretty standard practice and I explained above why I do it. And maybe I am not the only one.
Furthermore what harm does it do if I share my findings here that might bring an overall improvement in internal latency - even if not relevant to every use case.
Some others complain strongly about Geek bench results. To everybody his own needs.
Not yet, since it affects all hosts and VMs get reshuffled amongst them periodically (at least some of mine did). So I wanted to find out if there is a general solution for this.
I do not expect to get my own corner of bero via ticket;)
You are just bitching on this thread about your pings from the 1st day you bought those servers. Whatever you are aiming to do isn't doing anything good to provider.
You are shit talking about the provider bcs their ping values are inconsistent inside the DC. When repeatedly berohost asked you to create a fucking ticket so they can check the issue.
I am ordinary customer like many people on LET, so I am not going to defend any provider. But Andreas is good with dealing issues (yeah I created tickets about issues I had with node) and he was really quick to respond.
So overall I am happy with berohost.
Yet you are being a whiny twat, just posting random ping values occasionally, claiming there is an issue, when provider asked you to create a ticket again and again.
Do you see anyone else complaining about same issues that you are having? So perhaps the issue is you and not the provider?
and this just confirms that you are not seeking a solution, but just creating nuisance in the thread.
Create a fucking ticket. Let them investigate it, if you are not happy with their answer or solution, share it with us and just move to another provider.
How thick is your skull to understand this?
Edit: The dude has 38 comments on LET and 37 of them is about Berohost issues, since he bought them on 5th of October.