Please take the below with a grain of salt as these were very rough tests that compared 1 minute to 1 minute. The actual video recordings are posted under the Chia channel of our discord.
Methodology: To test I used 1203 K32 Plots spread out over 14x10TB Seagate Exos x16 Enterprise HDD’s rated at 5W idle using a SAS expander card. The PSU is a HX850 Platnium and is rated at 89% efficiency at 100watts. I used a wattage meter at the plug to measure and the PSU was only connected to the 14 HDD’s, 1 SAS expander card, and one fan. Standby mode was turned off for each HDD and the test was on Ubuntu. Chia version 1.2.11 and Flexfarmer 2.0.3 were used. I purchased this rig from a member of this forum. Only the power usage of the HDD’s (plus the fan and expander card) was tested not the cpu+other misc parts.
Data:
Wattages: (Chia GUI+Flexfarmer closed) ~95.76 Watts
Chia GUI: High 108.50 Low 102.13 Approx avg based on eyeballing the video 103-104=7.392 Watts per 10TB drive
Flexfarmer: High 104.04 Low: 98.8 Approx avg based on eyeballing the video 101-102=7.25 Watts per 10TB
Conclusion:
There is around a 6 Watt difference between the high and low of both farmers.
There is approximately a 1.92% decrease in power usage per HDD based on the eyeballed average.
These results are not conclusive. This is one test with each only measured for 1 minute. I encourage other Flexpool farmers to run their own tests.
Explanation:
Alex states that the difference is likely due to the Proof of Space and Time Flexfarmer uses. He stated that the standard Chia PoST library lacks some optimizations and that the most recent benchmark comparing the time for execution of loops filled with quality and full proof lookups for random input the original PoST in C takes 52 seconds and our improved version in Go takes 26 seconds.
4Wh * 24 h = 96 Wh/day * 30 days = 288 Wh/month = 0.3 kWh/month * $0.2 / kW = $0.06 saving per month per 140 TB of plots
Is that really where the Flex farmer advantage is, where Alex should be spending his time? Sure, it is a good checkmark to have, though.
Just glancing at chia’s start_farmer / start_harvester processes, they take about 10-30% of chia’s CPU usage. Maybe a better power saving is on the CPU side. Also, it may be not just power saving in general, but rather leaving more CPU cycles for the start_full_node processes, while under stress (dust storms).
Idle is 5W and at 90% efficiency that’s 5.5W. It’s unknown how much of power usage was from the fan and expander. I’m guessing it’s around 7W per drive farming. Around a 0.142W reduction per drive isn’t huge but it adds up. It’s not just a 2% reduction in costs but it’s also likely a 2%+ increase in lifetime because the drive is working less and isn’t as hot.
Flexfarmer does a lot more as it provides one click farming, reduces iops, increases speed of lookups, reduces Cpu and ram usage and yes provides large reductions in the main rigs power and resource use. But this test was just focused on per HDD benefits.
I got some data on reads and writes but I need to review it and I’m not sure I should post it as it’s rough.
I think that is what makes me interested in Flex farmer.
One thing that I would add to that is that on this forum people are discussing problems with their setups, including harvesters / farmers. I cannot say that I have seen one person writing about problems with Flex farmer. Although it may be that those people go to Flex discord channel.
By the way, are Flex farmer logs compatible with chia logs? Can monitors like Farmr, ChiaDog, etc. be used with it?
There have been issues, most of them small, an update is usually released in 1-2 days after a report. Because its so fast and simple to use if we test it ourselves and it works for us it usually works for others with only small bugs remaining. But yes, please go to the flexfarmer channel in our discord if you have issues.
Flexfarmer has an api and people have setup prometheus metrics. I don’t believe its compatible with those programs.
shit son now you’re making it hard for me dismiss your claims because this time you bring hard evidence.
I think here are flirting on the edge of margins of error. How many partials were picked up in that time? I think the last leg of your calculations is dividing the power usage by partials in that time period to get a watt-per-partial figure, as its impossible to control for how many lookups each client will do in the test period.
1 minute test isn’t enough, this is not a statistically significant difference. It feels like splitting hair to be honest, I don’t think there’s much to gain in opex with a different farmer, we’re still talking about chia and not gpu mining afterall
This is really interesting data. I think one key thing to note here is that the power consumption was measured only for the HDDs. I would like to see the power usage comparison of the whole machine (including CPU) while using Chia GUI and Flexfarmer. Since Flexfarmer uses so much less CPU resources, the power consumption difference should be a lot more than 4 watts.
I might even do this test comparison on my own setup to see the difference.
Completely agree but can’t just stand infront of a wattage meter watching it for a hour. I encourage everyone to do the test for themselves.
Yep. I didn’t track partials well but I assume there were a few during the minute given diff 1.
Highs and lows give some indication. Feel free to repeat for yourself. Alex says it’s not enough to prove anything too.
I would say power savings aren’t that important unless you have many PB’s. But I would assume the reducing in power and heat would increase HDD lifespan.
For me I see the full node using 3 Watts. Flexfarmer uses practically nothing but 3 Watts is $3 per year so that’s not why I use it. Flexfarmer is just better and easier than having multiple harvesters
Yeah I know it’s better. I’ve used Flexfarmer since the beta and haven’t gone back to the original farmer. Well I still farm my OG plots on the official software. But I’m just curious on how much the difference will be when the whole farmer machine is taken into consideration and not just the HDDs.
If I got time next week I’ll do a test on my setup, can post the results here.
Hard to compare power use of the main rig as everyone’s setup is different. I think flexfarmer can go down to 1-2 watts on a pi zero. I just focused my test on the hdds themselves.
Sure, but wouldn’t the wattage difference be bigger on any setup if FF uses less cpu resources to farm?
Just saying I think the amount of wattage saved is probably more than that 4 watts if you take into consideration using less cpu power as well.
