Best Safe Node hardware

Ah completely missed that one. Thanks! Yes I do think that the CPU and capability of the router is a larger bottleneck than upload/download speed. At least here in the Netherlands where glassfiber is common and 2,5Gbit/s 4Gbit/s and 10Gbit/s can be ordered as well now for consumers.


20 nodes with vdsl mikrotik hap at 100% all the time, waiting for mikrotik to die so i could replace it. :slightly_smiling_face:

Wow, that is nice. I don’t know about any ISP in Czech Republic offering more than 1 Gbit to home users.

1 Like

It’s ofcourse massively oversold, and you need quite some beefy hardware to be bottlenecked by a 1gbit link.

Software router, pfsense or opnsense.

Which provider do you have ?

So far do we have a good estimation of HW requirement to run one node?

I’m about to rebuild my retired HP Microserver with the following characteristics:

  • CPU : AMD Turion II Neo N54L (2.2Ghz)
  • RAM : 1*4GB (can be upgraded to 8GB)
  • HDD : 1 SDD 120GB for the OS ; 2*4TB for data
  • Internet speed : 1GB/s down / 500Mb/s up

What would be the limiting factor? CPU I guess?! How many nodes would I be able to run?

1 Like

Probably your router like I’ve just found out :joy:


That CPU is ancient, it will definitely be the limiting factor. My guess is 20 nodes maximum.


That’s very optimistic. That cpu has 853 score on

I’d say try 2 and cross your fingers…

1 Like

It’s old. But it’s not terrible either! I’d say it’s a bit like a Intel Core Duo in a 15 year old laptop. And I’m guessing it would be able to run at least 20 nodes. Pure guess though.

I have a N40L kicking around so maybe I’ll put it to the test. Funny thing is I have no idea where it is. You’d think I’d know with it being the size it is!

In my experience you need around 750 points per node on a fairly modern cpu.

1 Like

I see. Well I think there is something amiss because I can run 40 nodes no problem on a RPi4 with its ARM A72 4 Core which gets 834 on that same chart you see the Turion II having a score of 853. So I’m going to stick with my prediction of at least 20 nodes no problem.

I won’t be able to sensibly test more than 20 nodes with my terrible ADSL connection.


General use benchmarks won’t say much, safenodes are specific use which behaves differently. One of my machines has Celeron N3350 which is 300 pts above ARM A72 and chokes on 20 nodes.

Number of cores is very important for running safenodes. It won’t show in top but because we run more nodes than cores, it means OS has to switch between tasks and that costs significant amount of CPU time. More nodes per core translates to more “power” lost to hopping between tasks.

I don’t have HW to test it properly, but if you take two CPUs with same architecture, one 2core@4Ghz and one 4core@2Ghz, the second one will be much better in running nodes.


That’s a very good point! Yes, I’d forgotten that those Turions in the MicroServers are only 2 core and there could be a big effect from having more cores for this. That puts AMD CPUs at a big advantage as they tend to have more cores.

I still think 20 would be feasible on those Turions and it would be a lot more than just 1 or 2.

1 Like

I managed to get 30 nodes running on an old qx9650, but I had to turn it off cause of bandwidth issues on the router. I’m unable to get the launchpad working again now (to run lesser nodes), so will just wait for an updated release.


There’s a big difference though between having 20 node processes running on a quiet network and keeping them running when things get hot.

In previous testnets I saw my xeon 12 core puke on 150 nodes when things got hot and everyone was hammering the net with massive uploads.

AMD’s are faaaarrrrr better than intels. Even comparable cpu’s with the same core count.


In the future, is it planned for the node to be bigger than only 5GB? With my Turion, assuming I can only handle 20 nodes because of my CPU, does that mean that I will not ever be able to allocate more than 20*5GB space to Autonomi’ s network?


Probably not, small nodes are beneficial because this assists decentralisation.


I suggest spreading your node count (maybe as VM instance of your OS) across multiple IP addresses with multiport NICs and ensuring your egress and ingress router bandwidth is sufficient to handle the aggregate I/O , and have your ISP run that many external IP addressess matched 1:1 this way you scaleout node count on the same hardware across multiple IP addresses. Might be time to look into and price a colo 1/4 cage with an MSP/ISP that can set that up for you.

Although that is getting even further away from what the goal is supposed to be: getting lots of nodes into lots of people’s houses on spare resources. Not getting lots of nodes into one location or network using resources that have to be paid for.

I’ve looked at the options and it’s just about doable on my budget to set something up in a colo to run a few thousand nodes. It doesn’t really make sense economically but I would do it for giggles and if I thought it would be helping the project. But it wouldn’t really be. Just getting me up the leaderboard and getting me tokens and making it harder for others to get rewarded.

I think the better thing for me to do is focus on drumming up interest and getting some nodes running in people’s houses.