Key Features
1. Intel® Atom® processor C3558,
Single socket FCBGA 1310
2. System on Chip
3. 2x 2.5" 7mm fixed drive bay
4. Up to 256GB registered ECC,
up to 64GB unbuf. ECC/non-ECC,
DDR4-2133MHz, in 4 DIMM slots
5. 1 M.2 M Key 2242/2280 (PCI-E 3.0 x2)
6. 4 GbE LAN ports w/ Intel C3000 SoC
7. 2 USB 2.0, 1 VGA via BMC
8. Fanless
9. Lockable 12V DC 60W power adapter
I am currently with a dell i5 optiplex 3050 thin pc right now, plenty of power for a farmer.
The processor is rated for 35 watts but the system power consumption is not tested.
I’m still up to actual power measurements on the wall but do not want to shutdown my node currently. Probably ill do that when upgrading my Node to 1.3
Pros of the super micro system would be:
registered ecc ram and 2 slots for nvme.
fanless design (unfortunately very dusty location)
amd embedded also seem very copmpelling to me. (epyc is at 30 watts though)
Let’s see. The full node job is to basically sync the blockchain db.
So, if you don’t care about that part (only need wallet and farmer/harvester), the most efficient node is FlexFarmer. It draws 0W for the full node part. I mean, FlexFarmer is not running node at all
My understanding is that you cannot start just farmer, you do need to start the full node, so it needs to be synced. Although, I would try it. In the other thread, there was a discussion about harvester, and the standard way is to also start it through a daemon, but actually it can be started directly (no daemon or other processes needed). Maybe farmer is the same thing (as it communicates with the pool via REST requests). Although, the challenges may be coming from the full node (therefore the blockchain sync is needed).
As far as FlexFarmer, that is a combined farmer / harvester that talks to the hosted modified full node. So, you would run it instead of your harvesters, and basically by done with that whole red box. I use it kind of in emergencies (when working on the full node, to not be time pressured), and it works great. The main problem for me are logs, as those are not compatible with chia logs, so that makes FF basically worthless if you are using any chia monitoring tools (log analyzers).
The newer generation as in your link, might not be too bad at all.
The downside is they’re kind of pricy, and if it breaks you have to bin it all as the cpu is soldered on.
The old Atoms like the n550 were really crap so don’t be tempted to buy an older version to save money I had one and it struggled to run windows 8 alone.
You can get the low power ‘T’ versions of Intel cpus. But the motherboards typically have lots of things on-board that also consume power like sound chips, and other controller ic’s.
Regular PSU’s probably aren’t going to be that efficient at such low wattages either.
The atoms can run off a small 12 or 19v adaptor I think. Then you have the saving from the fans too.
You can get fanless heatsinks for standard cpu sockets just like oversized coolers don’t forget.
Update us if you try it, like to know how well it works.
Saving pennies on the CPU+Memory energy consumption? It is a wrong thing to measure. Most of your power consumption would come from HDDs and their enclosures, wouldn’t it?
For the same reason I think debate about FlexFarmer saving energy is not worth of a doughnut hole. As far as I am concerned, the only practical advantage of FlexFarmer is that you don’t have to maintain full database, thus saving space/endurance on a disk. Harvesting plots would use the same amount of energy whether you use vanilla or Flex harvester.
Assuming you are done plotting, the most operating costs come from sustaining the HDDs and the JBODs (or USB enclosures) they are in.
I agree with you, about the whole thing about the energy (to some extent). Although, for chia setup, the full node power requirements are driven by CPU horsepower and efficiency to keep up with the network. As we see, those requirements keep growing (slower db with size increases, more dust storms), so trying to “optimize” power side is rather pointless.
My point about FF was rather that FF doesn’t have the full node (as you stated no db, but also immune to dust storms (to some extent, as actually the pool’s farmer takes the hit, but this part can potentially better scale) and sync issues - no syncing at all). To some extent it also makes a farm be more resilient to failures if there are multiple harvesters (there is no full node that can bring the whole farm down).
Saying that, I am Flex pool member, but as stated not using FF (permanently) because logs are not compatible with chia’s. If not those logs, I would switch today (especially, as v1.3.x has now wallet mode).
Anything ARM will be crazy efficient. If you want to spend more bucks, try the new apple products : their new cpu is crazy efficient and you can run anything on them with ease. If you want something lower budget, try RPI. If you’re in the middle, any NUC with a SSD and low ram stick count will have low power consumption and a ton more power.
It’s my understanding that you can run a full node farmer on RPIs with some tweaks (increase swap, reduce node connection count, add overclocking with active cooling). Maybe having two Pis set up in redundancy, with slightly different specs and parameters, would work best for the time and money spent + power efficiency?
For now I will stay away from non-official farming software… Also I disagree with flexpools political entrepreneurship.
I am not intending to open up yet another operating system box right now and stick with Ubuntu server. If you can suggest another arm box running unix server, Id be glad to take the suggestion.
I have seen some micro boxes with snapdragon processor which are quite appealing to me. Thank you for that suggestion. I have not looked much into that area.
20W power saving equals 20~60 bucks yearly. Not really much but still, I try to keep everything as efficient and maintenance free as possible. If you consider 30 watts energy savings equals 3 more disks you can run in your Office, you would even get to 430.68$ per year.
As far as I have seen, the ryzen soc’s are ripping these atoms apart power/performance wise.
Correct, this is another page for optimisations. I am now evaluationg how many drives I can run on a single cpu/board for my next office to open. Right now my JBODS consume roughly 5-10W per 10 disks.
You are right, it is not possble to upgrade/replace the CPU on those boards. However, I do not expect the performance requirements of the node to raise dramatically as we aproach maxing out the blockchain speed with the dust-storms. System memory can easily be upgraded with these boards and if you were to upgrade the cpu in 3-4 years, you can throw your components (mainboard, ram) anyways as Technology and Sockets moved on. I have been running some low power SoC systems for over 5 Years now without any issue and they are running my Node fine (replaced thermal paste recently ans cleaned the cooling sink).
