# Staking on Power Efficient and Low Cost Hardware: From ARM64 to RISC-V Boards | Devcon SEA

- Channel: [Devcon](https://streameth.org/devcon)
- Date: 2025-10-09
- Duration: 25:13
- Topics: Science & Technology
- Watch: https://streameth.org/watch/yt-owzXocC1biE
- YouTube: https://www.youtube.com/watch?v=owzXocC1biE

## Description

The entry barrier to staking on Ethereum got lower, as ARM boards, the tooling and OS support have improved massively. We show the current landscape of hardware options and the software stack to go along with it. 
As a glimpse into the future we will talk about RISC-V, an open CPU architecture, present the current state of RISC-V based single board computers. We will discuss the progress we have made to run Ethereum nodes on these boards and the road ahead to optimize clients.

Speaker(s): aavegotch1.eth, haurog
Skill level: Intermediate
Track: Core Protocol
Keywords: Validator Experience, Home staking, Decentralization, optimization, hardware

Follow us: https://twitter.com/efdevcon, https://twitter.com/ethereum, https://warpcast.com/devcon
Learn more about devcon: https://www.devcon.org/
Learn more about ethereum: https://ethereum.org/ 

Visit the https://archive.devcon.org/ to gain access to the entire library of Devcon talks with the ease of filtering, playlists, personalized suggestions, decentralized access on Swarm, IPFS and more.

Devcon is the Ethereum conference for developers, researchers, thinkers, and makers. 
Devcon SEA was held in Bangkok, Thailand on Nov 12 - Nov 15, 2024.
Devcon is organized and presented by the Ethereum Foundation. To find out more, please visit https://ethereum.foundation/

## Transcript

[Music] hello hello it's nice to see some friendly faces here I'm HCK I'm a physicist turned software engineer and now full-time in ethereum but not associated with any project hello I am aot youan I'm a software engineer as well I'm an ethereum hobbyist and for the past few years I've been trying to make ethereum work on various and exotic Hardware that we will be talking about thank you for being here so when when we did this presentation prepared it we were thinking okay there's two kind of audiences here it's home stakers they might be interested into power efficient Hardware how to get their setup started but there's maybe also some software Engineers here who are interested in in the risk 5 platform so we will'll talk about that as well and how this came about in the in last summer in July aoi made a call hey who wants to bring consensus clients and execution clients onto a risk five platform I knew him from before in Discord um I joined him and since since summer we started um compiling clients running them on these platforms and then we saw that Al Al the ethereum one arm team is working on this possibility so we teamed up with them a bit so it's Diego and Fernando from the ethereum on arm team uh there's also Leonardo Aras he's associated with flash Bots who who was part of it and then it's a two of us who are independent of any project so let's get started with arm boards why would anyone want to use Arm boards actually they're quite nice because they're all in all on in on one board so you have the RAM on it you have everything on it it's just one simple uh device they're low power consumption and it's a well balanced Hardware if you choose the right one so it's a good CPU enough RAM and you can get running the problem is a bit it's a steeper learning curve getting Linux to run on them it's a software support and generally is not on par compared to Intel Nooks or anything like that and sometimes the a ability is a bit difficult so if a new barard comes out it might be sold out for weeks or even months most people think about the Raspberry Pi it's a nice armboard it's good it's available in almost all the shops it's cheap and has very good Linux support the problem is they don't have enough RAM the CPU is just barely fast enough and you need an additional board to attach an nvme SSD so it's possible to run a node on them but not the best choice nowadays the good ones are the ones we use and also ethereum on the arm suggest to use it's they have all the same RK 3588 CPU they have either 16 or 32 GB of RAM it's orange pi5 Plus nanop pc6 or the rock 5B somewhere between $ 120 and $220 depending uh where you get them important thing always an official power supply because the ones you have at home they most probably won't work and that's a pain um you'll have to add an SSD which has the right size and please also use a Linux specifically made for these sports like the Ubuntu from Joshua Rees or ethereum on arm um CPU performance wise you see a small comparison here we have Intel and AMD CPUs which you can might have at home at the bottom then you have the orange pi plus in the middle which is about two to three times slower lower performance than these high power ones the Raspberry Pi 5 is another factor of two slower and the Rasberry Pi 4 yeah as you can see we probably shouldn't use that one we I ran a validator on it a validated node on ethereum Main net without the validators we saw so the no node used about S 5 to 8 watt the CPU usage was about 25% at Peaks and yeah Ram was about 50% filled so it was 8 gab was filled when I directed a few notes a few validators to it uh power usage was slightly high higher but more or less the same CPU usage increased to about 40% and RAM was pretty much the same interestingly when I looked at at a ation efficiency over then the day which I used it it was pretty much the same as on my high power uh node I had before so you can totally run a ethereum full node on one of these devices so with this I give it over to aoi thank you so um some tips and tricks um first of all the arm boards and also the r five boards are fragile they will always come without case you need a case for to protect against dust and other unwanted electrical short circuits like your hands um an active cooling it's always required because the CPU will Spike most of the time and uh you'll get better performance and actually better power usage including the uh small power usage of the actual active uh fan so you'll not losing power performance um also an important part is the nvme uh for these small boards just take the good enough um nvme SSD do not go overbo overboard always check that the uh PCI controller um is um on the exact PCI controller that the mvme supports if you buy a I don't know a PCI Express um um version for NVM and you put it in a PCI Express 2 controller it's it's a it's a waste of money also note that the PCI Express 4 and five are uh very uh let's say power hungry they um need a better Cooling and also note that if you have a 4 terabytes in the future that also will need better cooling because the chip surface it's directly proportional so it's twice also um related to overclocking the CPU it's worth it's worth it but just make sure that most of the time if you increase the frequency by 5% you will basically increase your power consumption by a factor of two maybe 10% maybe more so just make sure that uh you take that in into consideration also an important part um the arm boards uh have CPUs that use the big little architecture so uh for example the orange Pie has uh eight cores but four are performance for are low uh low performance SL efficiency so in some sense in some cases you might want to pin the consensus EXE the consensus client on the more performant CPUs because that one will be used more than the execution layer uh day to operations okay I have it all running all installed what to do next so it's important to to choose a very good distribution like Ubuntu or Debian based uh also to have a well-maintained staking framework like dap node um eth e Docker is the is my favorite choice because most of the time the um base OS that comes from the manufacturer uh it's very has very low um capabilities low um low set of libraries so if you have Docker you basically borrow everything from the community via Docker so that's why I'm very a fan of e Docker also make sure every time that you firewall proof your system if you have an at home um staking gig uh because you might have a public IP without knowing it so if you actually have a public IP you you need to find out if you have it and fire all it making sure that it it works also I have there ker live patching thing that is very important to sleep well at night uh for example Ubuntu support it supports it is free for up to five machines also uh VPN for management wire guard it's a very cool Linux technology um it can it it can create very fast very efficient uh peer to-peer tunnels also tail scale tail scale or openvpn is good if you have enough R monitoring with uh e Docker makes sense um also if you have a lot of boards like I have uh it's worth to invest in a small UPS also if you're electrical grid is uh failing you it's it's a worth U uh the investment is worth so the summary on short um I uh I'm using it also as a solo home validator on arm64 but I would suggest to start first using them as a fallback node uh normally eum node you can run multiple test Nets on on them it's just a matter of ram sometimes and of course play with Hardware before investing in a very big staking rig it's best for that let's go to the risk five boards so um risk five has a quite a let's say young history it was um uh it's a young alternative to x64 um arm 64 instruct action set um the the standards are already in place and the most important part is that is free open source so everything there uh comes royaltyfree um it's collaborative um and in theory it should achieve a way u a way faster Innovation rate which for us means uh lower prices in the future lower prices better efficiency so we are somewhere in the middle at the moment um the what's hard has been already done with the risk 5 uh foundation and all the standards that are already in place and important is that uh we are at Big effort so a lot of Hardware software uh companies need to um come together and um make it um as um as performance as efficient as available as the arm 64 and then the x64 um the mainstream usage will come at a stage when you will be able to move any workload from x64 to arm 64 you can almost do it for arm 64 x64 right now and then to risk five so basically you'll have three Alternatives that would be the goal okay so we started with the LI Pi 38 to uh this year uh banana pif3 16 gab of ram ballpark $150 and then okay can I run a staking node on this board can I the answer uh will be given in a moment but you can see that there's a long way till um it uh will will see parity So currently these are the most uh Advanced boards on the market or the advanced CPUs which are affordable thank you perfect so then I continue what we did we we uh tried and built all the clients that are available I don't go into all the details but I I pick a few it's like gu the execution client is just beautiful you build it on risk five it works it runs no code change is necessary what we did with the con sensus client like Nimbus it was good to see we just had to change five lines of code maybe uh or added a Target it compiled it ran code is now Upstream to the main branch and it's in there now for two months so anyone can build nimus on risk 5 boards another one is Lighthouse a lot of libraries are not compatible with risk 5 but we made them work with a few hacks which we have in Loc local Forks of the repository we didn't Upstream it back yet all of them but um we can run it locally the stuff but so as you see many other clients are just in different states of non-working so there's still a lot of work to do but the nice thing is there is progress for example last Friday P to cogi uh tweeted out that g has now a Target added so they build a Docker image for risk five um now regularly then we ran the clients on various networks we used ethereum main net heski seola test net for ethereum and also the chiato test net which which is a test net for gnosis chain we first ran the clients on both clients on one board like nimus G and Lighthouse gath and as you can see an ethereum main net and heski it's kind of they can't syn up they might syn up to the tip of the chain for maybe a few minutes lose syn again and then not being nothing syn for half an hour and then syn back up again nice thing is seoa Works sinks up is stable perfect unfortunately we can't run validators on there it's a permission set so we went to chiato test net and tested it there we ran a validator it was stable enough we can see later on when we did the split client so like only one client on the risk five board to see if if it runs better if you only run the consensus client on the risk five board you see in the the the lower part of the table it's actually the same picture so it seems like the consensus client is a bit limiting but the nice thing is if you run G on the risk five board uh you can easily sync ethereum main net when the consensus client is on another board so it looks like consensus clients are the limiting ctor especially on large networks so if you have the validator validator set is huge these boards are limited and here is an example for validator and the chiato network what you see here in these graphs it's the CPU performance CPU usage over time on the left side is just normal operation of a validator node you have some spikes which are most probably Epoch boundaries and if you have a sync committee you have the CPU usage goes up to about 50% it's very stable and it uses about 5 gigabyte of RAM and overall it's about four to five watt of power used for this operation so as you can see it can do all the duties and the most beautiful part that's a From the Block proposals we did on the CH Network continuous block proposals no outages nothing and if I look at the rewards we get they are comparable to to what other no validators get so the testation efficiency is great on these boards on the testnet but what we need we need better Hardware company's working on that in the next one or two years we'll get where there where we needed to be we need to get the clients ready we need to set up node setup tool like eth doer and S ready but obviously they're only working on that once the clients are there and more importantly for software Engineers we need to get some of the cryptography libraries implemented in assembly for risk 5 so which is a nice task if you like this kind of stuff because there it's not very often in your lifetime as a software engineer that you can get the chance to start from scratch in a in in this kind of assembly code but a little bit further in the future what could be amazing is like ckvm they're mostly designed to run on risk 5 instruction set architecture so depending what P kind of path we take in the future one could IM imagine that these kind of ckvm run natively on the risk 5 CPU which is just would be amazing so to summarize for the risk five boards if you want to run an ethereum mainnet node don't do that it doesn't work yet if you want to run a testnet note it kind of works maybe not the best choice but if you want to play with Hardware if you want to improve the clients if you want to write code for risk 5 or maybe have a glimpse into the future it's amazing to play with them thank you so um here is a ethereum future that might affect the small validators the small boards that we are trying to to make work so the 4 terabytes requirements might be coming sooner than later so that might um you know 4 terabyte means double the price almost the price of the board itself um also veral from the discussions as this conference I understood that verle might put um strain on the CPU so the uh CPUs that we already have that are already working with veral might not work anymore we need a lot of testing to do uh to actually confirm or infirm this U thing also if if and uh when the staking requirement will become smaller that will be a boon for this kind of um uh home validators so it will be great also if the maximum um eum balance for a validator goes up um then it will be less chatter on the network again they will work better um as a conclusion for now there is quite a large um um set of boards that work uh they are working in production they are already working for one year two years already and um some of them um might fit a home Staker needs today um other we are waiting for the next iteration of arm64 and also risk 5 to see where this goes um again uh contributions are welcome for uh for the work that needs to be done mostly in the um assembly implementation of the crypto Primitives that's uh in our uh opinion that it's uh why the risk five cannot um uh cannot sync ethereum um on all notes um and of course uh it would be great to verify more risk five uh boards thank you very much um we have a set of links um um on the presentation if you download the presentation okay cool I'm going to push you guys this way so I can join you on stage and we can handle questions okay thank you again so much uh we'll start from the top um again QR code here you can also collect a card to indicate that you've shown up here but um first one how much disk is necessary to contain the ethereum blockchain for the next year um So currently is borderline uh on uh two terabytes of course it depends on uh on each client slash um um so execution SL consensus layer client so I think without sorry I'm getting asked to push you guys more so we we without the verical implementation we'll expect it in one year or two um maybe faster it depends also again on the ethereum usage right okay perfect in your experience what's the biggest factor for missed attestations Network SSD or CPU um I I can continue on that I saw a a direct example example on uh on arm 64 is the CPU um basically if um I overclocked a little bit the CPU uh it worked perfectly so um yeah I from from my experience CPU obviously depends on the Node setup you have when when I support um node operators most issue these people have with is with the SSD which then yeah influences attestation rates but sometimes in a weird way so it's hard to track down but it really depends on the setup where the bottleneck might be yeah if if if you have any a good enough um nvme SL Network SL CPU that's no no issue at all depends where the borderline is so okay how important is 32 gigabyte versus 16 shall I take it okay 16 is perfectly enough for normal operations if the network is non-f finalizing and you run the ethereum mainnet node it's your main node for staking you might be part of the problem because your node might go down as a fallback node that's most of the time no issue so 16 gab might be enough depending on what exactly you do otherwise it go to 32 so um I mentioned also in the presentation if you want to have a performance monitoring and other let's say small utility stuff there uh 32 is U is is way better because you'll have a lot of more space to to put more more applications that can help you on the long term okay what possible Avenues can one explore if they want to contribute to eth compatibility on arm or risk five platforms that is a very open-ended question it really depends on your skill set what you want to do and uh in which part you want to contribute like there's all the clients most of them have issues so it's different programming Lang language different kind of each issues for some it's as far as we understand issues with the Linux kernel which needs to be changed specifically for these sports like for WS R you has have to do that for other ones it's probably implementing assembly code stuff if you like to do that one can do that yeah there are also more powerful boards but they are prohibitive in price they're like like $1 1.5k uh that might be able to run it at this moment like um making a obstruction of that they there's no assembly code for the crypto Primitives so if someone can um has access to that um kind of Hardware it would be great to know the results the code already works okay I just got a sign that said times up so thank you guys so much for joining if you have there are additional questions here so feel free to meet them on the side there's actually also a function on this to give speaker feedback so do that and watch out for that too perfect thank you so much thank you thank you thank you