Adventures in HDD mining

[TechElucidation]

Since the ‘SSDs/HDDs’ forum was empty, I figured I would put together my adventures in this area.

My original intent is to build up storage to use with SCPrime. In short this is a distributed storage solution, where you are paid for providing storage to the network. However, that project is new, and has currently has 50,000 times the capacity than the usage, so there is very little chance of renting our space there right now.

So while I wait for that SCPrime to (hopefully) pick up, I figured I could use the unused space for Chia. If SCPrime picks up I could remove plots to make space for it, otherwise try to make something out of doing Chia.

SCPrime does sell its own XA-Miner (customized NAS) solution, but it would be limited to just SCPrime. So I decided to setup my own NAS to allow me to do the Chia portion of it. In the end I decided to go with a QNAP TS-873A, which gave me 8 HDD bays, two NVME slots, and a PCIe slot. The QTS operating system allowed me to run VMs on the device, and I can assign the PCIe slot to one of the VMs, and I ended up putting a NVidia 1650 (the highest officially supported by QNAP) onto the system as well. I ran two VMs, one Linux for Chia and SCPrime, and one Windows for NBMiner.

The most expensive part of the system was obviously the hard drives, and I did fill all 8 bays with 16TB drives, setup in a RAID 5, for a total of 101TB of space. I used two 1 TB NVMEs in a RAID 1, which is where I stored the VMs, OS, etc. Since I was running two VMs, I also upgraded the device to 32GB of memory.

For people who know Chia the ‘plotting’ portion is very resource intensive and no way I could have done it on the NAS itself. I considered trying to use existing equipment (not very fast), or building a plotting system (pretty expensive). Instead I decided to use a plotting service, which can build pool plots for ProHashing, and costs $0.30/plot for orders over 100 plots. I decided to fill 90TB of my 101TB with plots, and leave 10 TB for SCPrime, so ordered 900 plots. While this cost $270, that was cheaper than trying to build a plotting rig, so to me, it made sense.

So, how much did this cost….

QNAP TS-873A-8G $1,050
NVIdia GTX 1650 4GT LP $355
PNY 1TB NVME $167 X 2 = $334
Crucial 32GB Memory $152
16TB Drives $289 X 8 = $2,312
900 Plots = $270
Total cost = $4,473

So, how much did this make…

For Chia, I am still downloading my plots (doing 14/day) and currently have 550 of them, and ProHashing is showing $0.77 per day. So I expect once I have them all I might make around $1.20 per day from it.

Running NBMiner off that single 1650 was also adding about $0.25/day.

To my total expectation with this would be about $1.50/day – certainly not going to retire on that, and you are looking at…. let’s just call it “more than a year” for the payback.

I really didn’t expect to be making much from Chia or GPU mining off this, was just more curious about using it for this as a ‘better than nothing’ while SCPrime (hopefully) builds up.

With the SCPrime recommended rates, the 10TB I left for it should bring in about $50/month ($1.67/day) if I manage to rent it all. If I do fill up the 10TB, I would start removing the Chia plots, giving more to SCPrime. In a wonderful world, I would end up with 101 TB of SCPrime data, which would give about $16/day, which would then bring the return on investment down to 280 days, which is not too bad…. But honestly, the chance of getting 101TB of SCPrime data is probably pretty slim!

Going overboard…

As my grand pappy used it say, “if you are going to take the time to do something, might as well go way totally crazy and push the limits”. With that in mind, I wanted to see just how crazy I could go with the GPU mining on the NAS.

As I said, officially QNAP only supports the 1650 at the most, and the built in 250w power supply does not offer an additional PCIe 8 or 6 pin power adapter. So the first thing I did was find a compatible sized 500W power supply, and modified the connectors on it to fit into the NAS, and give me a PCIe power cable.

I first attempted using an AMD 5600xt in there, but that failed horribly. First it was too long to fit in the NAS “normally”, but since the 873A does not have a display on the front I did manage to cut away a bit of the frame and part of the cards fan shroud, so it would fit in there. Second issue was that the QNAP would not recognize the card, so I was unable to assign it to a VM.

My second attempt was a Nvidia 2060 Ti, the Ventus GP OC 6GB with 2 fans. This was able to fit in the case, was recognized by the NAS, and I was able to assign it to a VM and mine off it. This took my GPU mining profits in ProHashing up to $0.75/day – triple the 1650.

But I was not done yet, my third attempt was a 3060 Ti. This took a little more cutting of the NAS frame, but did get it in there, but unfortunately while the NAS could see the card, every time I tried to assign it to a VM I got an error.

Most people would have stopped at this point, but with pappy’s words of wisdom, I marched forward using PCIe risers, and attached both the 2060 and the 1650 to the system – but this setup required a second PSU, really didn’t think I could get both working off the existing PSU. This did actually work, but I found you could only assign one PCI device per VM, so would have had to make another VM for the second card. I also was not happy with the idea of a second PSU.

In the end I went back to just the 2060 Ti ($528), and the upgraded 500W PSU ($142), and since then it has been chugging away just fine. This configuration would cost $4,788 compared to the original $4,473.

Final thoughts…

Overall – I would not recommend this to anybody. It was an interesting project, and I originally hoped that SCPrime would be a success, but that is really going slow. It may still pick up; miners really are eternal optimists!

As I mentioned I tried three other GPUs in there, and I took those, as well as the PCIe risers, and tried making a GPU mining rig - another challenge of mixed AMD/Nvidia cards, as well as a huge range of capability. Since I have the 1650 in that I am mining Kapow, getting a hash rate of about 50Mh/s, and (if I could stop playing with it) would bring about $2.20/day – already more than the NAS, and if I had purchased all the parts (rather than reuse some I already had) it would have cost $2,274.

So if your keeping track, for half about the price my GPU rig would make 45% more than the HDD system. To be fair, the potential (if SCPrime takes off) is much better, as that could make 8 times the amount of the GPUs…. But there is a lot of ‘if’ in that statement.

Again, don’t do it – that’s the important takeaway from this.

[Sarah Manter]

If SCPrime picks up, and you erase plots, will you lose the investment you've made in them, or are you expecting that this won't happen until after you've made your money back?

[TechElucidation]

Hi,

The quick answer is “it doesn’t matter if I recouped the cost or not”. With SCPrime you are supposed to aim to rent 1 TB of storage for $3 to $4. So, if I had to delete 10 plots (~1TB) that cost me $0.30 to purchase, even if I have not made anything off them, in order to rent the same space for $3+, the choice is pretty clear.

The use of Chia is more a fallback in case SCPrime (or I may look at some of the others) does not pick up. I am keeping 10 TB free right now, so even if it starts to work, I would need to have rented quite a bit before needing to worry about deleting any of my plots. If it takes a while to get anything from SCPrime, then hopefully Chia will at least pay for keeping the NAS running and making the offer to rent on SCPrime.

In the base configuration the power supply is a 250w system, so 0.25KWh a day, and in my neck of NH at 19c, that works out to ~5c a day of power. I am currently at 584 plots, and yesterday ProHashing paid out 98c on Chia yesterday – so it is doing the basic goal of keeping the lights on while I wait for SCPrime to pick up.

But as for ROI on all the equipment, that would be well over 10 years. So Chia is not going to be the solution to make anything off it in the long run. Adding in the GPU mining can make a little more, but still not significantly reduce that ROI time overall, especially considering how low end the GPU has to be to work in there.

With SCPrime, if it was to reach an average of 80% capacity, at an average of $3.50/TB, then you are talking about $280/month, which would ROI in ~15 months. Not too terrible, but the issue is getting to the point where you have that average of 80% capacity. This isn’t like buying an ASIC or GPU where your mining at that level right away, and 15 months could sound reasonable, it may take a long time to get to that level before starting that ROI clock.

[Sarah Manter]

Ahh. That makes sense! I hadn't heard of SCPrime. I'll have to read up on that more.

[TechElucidation]

Just as a final update, I have had my full 950 plots (~95TB) on the system for a week now, and on average have earned $1.69 per day over the last week.

I still have 6TB free that I am going to assign to SCPrime, once they figure out what they are doing with "third-party miners" - in other words people who are using their own equipment, rather than what they sell. It is getting confusing, and sloppy, and the project seems to have stalled with various people going in different directions, wild expectations, and end users getting frustrated.

I also moved my 2060 back into the NAS (I had gone back to the 1650 for a bit) and on Ethash that has been doing an average of 31MH/s, and has been very stable running for days without issue. That is pulling in about $0.95/day ontop of the Chia.

But yeah, still stand by the notion that nobody should get into Chia unless they have a good plan for it.

Also still stand by the recommendation that nobody should buy plotting hardware; my 950 plots cost under $300, which is a fraction of the cost of building a plotting rig, and you can get your plots a lot faster (if you can download them fast enough) from online services.

Anyways, just wanted to give an update with all the plots in place.