A lot of folks talk bad about the S17, but why I wondered? Before I bought my first unit I did the research and came to the conclusion that it got a really bad reputation based off some poor engineering decisions and poor manufacturing early on. There are 3 versions of this, the S17, S17 Pro, and S17+, each with its own set of issues, but is there a diamond in the rough?
HISTORY
Bitmain first released the S17 in April 2019. There was a 50TH/s and 53TH/s version that ran around 2300W. This first run had a tremendous failure rate in the field, reported from 20-30%. Bitmain at first denied there were problems but early on in 2020 they admitted to having design issues and high failure rates. Two of the biggest issues stemmed from their power supply design and heat sink design. The power supply seems to not be able to pump out the required current to keep the hash boards running properly and the heat sink, well, they were just plain falling off and shorting out the unit (ouch.) Another couple issues that seems to be a common issue across all the 17 series is the units dropping a hash board randomly and bad I/O cables. The dropping hash board may point to power supply issues, but it has gone unresolved. This is fixed by a simple reboot most of the times. Same with the I/O cables. Doesn’t seem hard to get that right, but alas they didn’t.
Bitmain followed this up with the S17 Pro, also released in April 2019. This had a 50TH/s, 53TH/s, and 59TH/s version. The big difference between this and the S17 is that the S17 Pro has a Turbo mode in the settings, allowing it higher hashing power. This model was produced longer than the S17 and the later S17 Pro 59TH didn’t have as many heatsink and power supply failures. This does however have the “dropping hash board” issue that is generally fixed upon rebooting the unit.
Bitmain’s last release was the disastrous S17+, introduced in December, 2019. I’m still not sure what the root of the issues were, manufacturing, firmware, cables, but these were plagued with issues from the start. It was also a power hog at 3000W for roughly 72TH/. Why does that number matter? Well the S17 and S17 Pro all ran under 2600W max, which luckily allowed circuits to be designed such that two units could fit on one 30A circuit. Take that same setup and you can’t cram two S17+ units on the same circuit without overloading it (beyond 80%.) IMHO I wouldn’t have released a version that just went over that number, I would have kept it under 70TH/s so it could have been a plug in replacement (electrically speaking) for the previous generation. There, off my soap box. If you’re going over that, go big, which the next generation S19’s did.
COMPARISON
So there’s our quick history lesson, so let’s bring it in to the main question. Which one is the best of the “average” and what does that look like. After doing my research I settled on the S17 Pro 59TH. I have had a lot of success with this model, albeit I have seen the missing hash board problem here and there. Knowing these were some of the last units manufactured I had more confidence that their processes had been worked out. Additionally I looked at the return based off cost and hashing power.
I started by looking at a more recent SHA-256 miner, the S19 Pro 110TH. If we look at the price per TH we get (based off today’s used prices) $11,000 / 110TH = $100/TH (I love clean math.) Keeping apples to apples we get the following for these other SHA-256 miners:
S17 Pro 59TH – $3000 / 59TH = $50.84
S17 50TH – $2500 / 50TH = $50.00
S9 SE 16TH – $550 / 16TH = $34.38
S9 13.5TH – $450 / 13.5TH = $33.33
As far as I understand it, a TH is a TH when it comes to SHA 256. That’s why so many people still use the S9, the same rate applies whether you are at 13.5TH or 110TH. So basically when it comes to straight hashing power (not profitability), 8 S9’s at $450 apiece ($3600 total) equals one S19 110TH at $11,000 I believe. I know, factor in power, but I’m just speaking straight hashing power at this point.
LOAD BALANCING
Now that we have our cost per TH, I looked at what would maximize my profits based off available panel space. I have 60A available, split into three 20A circuits. Keeping that configuration, and picking the most robust of the flavors, I could get the following:
Per 20A Circuit (max 3840W)
- 3 x S9 13.5TH (1200W) = 3600W (40.5TH total @ $1350 total cost = $33.33 per/TH)
- S17 Pro 59TH (2400W) + S9 13.5TH (1200W) = 3600W (72.5TH total @ $3450 total cost = $47.59 per/TH)
- S19 Pro 110TH (3250W) + nada = 3250W (110TH total @ $11,000 total cost = $100 per/TH)
If I change my wiring and turn the three 20A into two 30A circuits I get the following:
Per 30A Circuit (max 5760W):
- 4 x S9 13.5TH (1200W) = 4800W (54TH total @ $1800 total = $33.33 per/TH)
- 2 x S17 Pro 59TH (2400W) = 4800W (118TH total @ $6000 total = $50.85 per/TH)
- S17 Pro 59TH (2400W) + 2 x S9 13.5TH (1200W) = 4800W (86TH total @ $3900 total = $45.35 per/TH)
- S19 Pro 110TH (3250W) + 2 x S9 13.5TH (1200W) = 5650W (137TH total @ $11,900 total = $86.86 per/TH)
SUMMARY
All were viable options, clearly the most hashing I could get involved getting an S19 into the mix and the cheapest was keeping with the S9. So many numbers, so much to consider. I decided that my best, and cheapest route, was to stick with my 20A circuits (didn’t feel like rewiring the basement) and go with the S17 Pro 59TH. Why? The big reason was cost per TH balanced with getting my fair share of hashing power. The S17 Pro fit nicely in the middle and puts my ROI on the unit at more than half of what it would be for an S19 Pro 110TH.
I’ve since installed HiveOS onto some of the S17 Pro’s with mixed results. It’s great to have more control over the frequency and voltage, but some of the core issues that weren’t firmware related still rear their head (I’m looking at you ghost hash board.) That aside, I’ve had some of these running around a year and still hashing great.