Efficient Harvesters (Compressed Plots)

Would be nice to start a thread with what people are doing for the harvester aspect of compressed plots.

Post here what your set up is with total harvester wattage, type of compressed plots, and how much.

I think it will help people find the best set up for their situation.



My current set up is,

ZimaBoard 832 (6W TDP)

Planning on converting 142 TBs to C4 plots.

We’ll see how it does once fully replotted.

Waiting for full support for OpenCL and what others are doing before finding an optimal harvesting solution for my size farm…

Still missing a lot of info on farming requirements, at least from Chia’s own version.

But as far as I can tell, higher compression is only interesting when your kwh price is low.
The first compression step so far seems to be the one that is most efficient in terms of %increase versus additional power usage. And up to level 4 the power increase should be minimal.

I have an i3-6100 that is only running at minimal load for 250 TiB, so I expect it will have no trouble farming up to at least C4 for this size farm. Think the CPU is using 5-10 watts now

But there should be a tool coming out at some point that you can use to test the actual increase in power for each compression level. I will wait for that until I make any changes to hardware for farming.


Here is Chia’s estimate on watts increase per level of compression (Comparing to 500TB of non-compressed plots)

Found here, Plot compression TCO public - Google Drive


Personally I’m waiting for Chia’s compressed plot solution as I have ARM64 SoCs (RK3399, 7W) for harvesters which looks like are not supported by current MadMax solution or at least I did not see any downloads a part from x86.

1 Like

FlexFarmer is an option, it’s compatible with ARM64

Not a fan of centralization when it comes to blockchain so was not even considering it. Plus, I solo farm.

You can mine to your own node now with Flex Farmer…

Chia Network is awarding a grant to a light farmer like FF being built by Evergreen so looks like they support it, plus you can connect to your own node. Running 6 nodes on 6 different computers or 1 they all connect to doesn’t change much.

Also what makes you think Chia’s farming solution will come to ARM anytime soon? GPU farming is 6+ months out and likely Windows/ARM support will come quite a bit after Linux support. Adding more than Linux was hell for max and for us even with his help. We may have ARM support for Chia’s compressed plots before they do.

If your farming for charity then feel free to continue, if your here to make money then do the math and choose the best path.

That being said FlexFarmer Gigahorse for ARM isn’t out yet, still being bugfixed.

I realize I’ve been quite frank and may have come off as rude, if so I apologize. At Flexpool our goal is to make the benefits of being with us far outweigh the pool fee. I think we accomplish that. If your farming/mining to make money then hopefully we’re the obvious choice. You only pay 0.6125% of your income to us in return for the software, customer support, and website tracking we provide. FF will cut power costs enough to balance out our fee, in addition the reduced wear on your hardware may extend its lifespan.

You also save 0.125% off the Gigahorse fee so our fee is actually 0.4875%, 33% less than most other pools.

Since Chia solution is and will be open-source I can compile for whatever platform I’m running on (ex: ARM) just like I’m doing it today on my Linux boxes.

We have totally different goals! I’m not driven by purely profits or immediate returns and you should not assume that every one thinks the same as you do.


Unfortunately you will not be able to simply compile the source and run it on ARM.
I think the proof process will be too slow and the GPU code might be cuda only

Why did you delete? What you said seemed sensible

I wasn’t 100% on your situation… which why I deleted…

But I believe Compression level C6 is too high for the device you are trying to farm with. If it is a device like a Raspberry Pi, then the max compression level would be C4, but C4 would be pushing it at its max with 150 TB of C4 plots… And with the likely hood of the plot filter being reduced next year to 256 will further make C4 plots harder for a Raspberry Pi like device.

If using a Raspberry Pi, stick with C3 or lower, I personally would recommend C2…

There are two possible routes for this (in the future).

The first is “decompression” on a different box. Some people run a “naked” farmer, and all HDs are under remote harvesters (e.g., RPis). Max has already announced that he will have remote decompression module. In this case, all those RPis could piggy-back to that module (running on a farmer, or a separate box). This is maybe the most needed feature to be added to gigahorse, as no one wants to run a GPU in each remote harvester. However, I am not sure whether there is an ETA for it or now. (Not sure how powerful CUDA implementation is on Jetson boards, maybe that would be a low power option for RPi like setups to run such module.)

The second is that RPi has quite decent iGPU. I am not that familiar with RPis, but looks like CUDA may not be available on RPis, though. Still, if enough people ask about it, maybe Max will consider at least looking into this.

Ive grepped for “too much” in the flexfarmer log so this just shows the worst.
I don’t know if you can see but one plot looks like it gets hit twice so I’m not sure if each needs 4gb of memory or if it runs in series
I chose C6 as I am waiting for a gtx1650 to arrive. I don’t know if it will be enough either.

This is on i7 3770. The pi4 is just for ssh to the flexfarmer

10TB of C 6

I would like to have a go but Ive no idea where to start. Unless someone makes a binary that just works of course!

RPi may have a decent GPU… but is it better than the RPi’s CPU? To earn more when you have to pay the 3% GPU fee instead of the 1.5% CPU fee?

Will have to wait and see I suppose…

I picked up 3 Nvidia Tesla P4s. I got one for $50 and the others for just under $100.

Two to plot with and one to farm with. At some point, I will take the second one from the plotter and move it to the farmer when I hit near 1PB of compressed plots - sooner if I start getting stales.

Do not buy anything you dont need. The Tesla P4s are $100 and a great deal if you have a server


How many plots you have right now, what compression level, and maybe you know your avg lookup times.

Also, what are your plotting times and RAM size?

Well my third and final Tesla P4 arrives tomorrow, according to Canada Post, so who knows when it will actually show up. But right now using I’m getting ~7-8 minute plot times completely in ram (256GB DDR3) for C8 with the one Tesla P4. I hope to get down to sub-5 minutes with the second card.

I only have 40TB of compressed c8 plots out of 1.4TB so far and am using flexfarmer. My farmer is a HP ProLiant DL120 G7 which is hilarious but it’s farming 5 jbods no issues and takes the Tesla P4 like a champ. I do have a 12-bay server that I had retired that I’m going to bring back online that will let me fit the second Tesla card, when I need it.