ReviewToday I want to share my thoughts on the GekkoScience Terminus R909 pod miner, after owning it for almost a year.
HardwareThe device is roughly the same size as my previously reviewed
Apollo BTC, and I paid 599.00€ for it one year ago. Bitshopper now has some sort of 20% promotion where they
dropped the price to 479.00€, which is a great deal in my opinion, especially with current Bitcoin prices.
It is easy to see why I'll compare this pod-style miner to the Apollo here and there, since they're in a similar price bracket and form factor. They are marketed as home miners, so we should be able to run them at home, which puts several constraints on such devices. The most obvious is an
acceptable noise level as well as
decent efficiency, since home electricity can be quite expensive especially in Europe.
This is why the importance of the chosen cooling solution cannot be emphasized enough for such a product. Installing a more quiet fan is sometimes not possible, such as in the Apollo BTC, since it is just too inefficient / putting out too much heat for its size, so there are no real quiet alternatives to the supplied Delta fan. Some home mining solutions like GekkoScience's Compac F, just don't come with any active cooling, so users end up with somewhat scary looking setups with loose fans positioned in the vicinity of their mining setup, especially considering that stick miners often don't sense overheating and just continue running until the solder melts off and becomes a fire hazard.
All that is just to say how much
I appreciate the cooling solution of the R909. It is a simple 80mm PWM 12V fan, so even though the preinstalled Arctic F8 is not the best, it gets the job done and can be replaced by a much quieter Noctua model. I did run the miner with the stock fan for a month, and it was very noticeable hum in the background, so I replaced it with a
Noctua NF-A8 PWM with 2200RPM and it's been running happily like that ever since. While it's not technically silent, it's totally living room-friendly in this state and cools the chips more than well enough; more on that later in the performance section.
It also seems that GekkoScience have done a great job with thermal interface material between the
6x BM1397 ASIC chips and the beefy heatsinks on either side of the PCB, as shown above.
SoftwareHere's where it gets a bit more nuanced. While I had pretty much nothing but praise regarding the device's hardware,
the software side of things may or may not be suitable for everyone.
Compared to the
Futurebit Apollo BTC, this is not a plug-and-play solution, but it's also not meant to be, nor marketed as such. It's just something to be aware of before buying. The
R909 is extremely well suited to tinkerers and tuners and people who want to get their hands dirty with configurations and dialing in even individual chips, if they choose to do so.
In order to make it easier to get up & running quickly, I'll shortly release a full step-by-step installation guide based on Raspbian. Any old Model 2B+ (or newer) will do, as not much computing power is required at all; in fact, I'd recommend not to use something too powerful to keep power consumption as low as possible.
However, this means more hardware to buy, more individual 'electronic components' laying around and a bit more work for the user.
In order to operate an R909, you will need to install
kano's version of
CGMiner with the driver for GekkoScience products and configure a system service that automatically restarts it upon boot or any other reasons for failure.
While it is not required to install Bitcoin Core and synchronize the blockchain for pooled mainnet mining, such as with
kano's own pool, this unit is very well suited (even somewhat overpowered) for solo-mining Testnet BTC.
I recommend following
nullama's solo mining guide. Do make sure to follow the recommendations and fixes I pointed out
here, as they were not edited into his original post.
PerformanceI can confirm that
the numbers this miner is advertised with, are correct.
The miner is guaranteed to run at 1.5TH from under 100 DC watts at stock voltage and do it nearly silently. Better performance isn't guaranteed, but is probably attainable.
I'm measuring power at the wall, including PSU losses and the power consumption of a Raspberry Pi Model 2B+. In my opinion, this is what ultimately counts, even if the miner alone may pull 5-10W less from the 12V rail.
Throughout all my testing, I powered the whole system from a high-quality
Seasonic Focus Plus 550 Gold ATX power supply unit; however if you need to buy something new anyway, I would recommend getting a smaller unit such that it operates at 50% of its rated capacity, or whatever the manufacturer states to be the point of highest efficiency. In my case, pulling between 87-126W (16%-23%) corresponds to the unit's point of lowest efficiency - about 90%. Therefore, pulling 100W from the wall means the miner only actually consumes 90W.
MeasurementsThe following numbers have been recorded without any fine tuning, i.e. I did not touch the core voltage at all. Slightly reducing the voltage would lead to even higher efficiency, however it needs to be dialed in carefully in order to keep all chips hashing stable. If you reduce it too much, they'll stop working at all or intermittently, until you increase it again.
This is what's attainable by simply changing the following parameter in the cgminer config file.
"gekko-r909-freq" : "450"
| | Clock speed | | | Data | | | Power at the wall | | | Hashrate | | | Efficiency | | |
| |
| | |
| | |
| | |
| | |
| | |
| | Stock (450.00MHz x6) | | | 107.35kWh in 1235:22h | | | 86.9W | | | ~1.74TH/s | | | 20GH/s/W (50W/TH) | | |
| | OC (500.00MHz x6) | | | 16.39kWh in 168:33h | | | 97.2W | | | ~1.95TH/s | | | 20GH/s/W (50W/TH) | | |
| | OC (550.00MHz x6) | | | 26.17kWh in 238:52h | | | 109.6W | | | ~2.14TH/s | | | 19.53GH/s/W (51.2W/TH) | | |
| | OC (600.0MHz x6) | | | 7.65kWh in 60:56h | | | 125.5W | | | ~2.33TH/s | | | 18.56GH/s/W (53.9W/TH) | | |
For comparison, the Apollo pulls about 200W while churning out between 2TH/s and 2.5TH/s, leading to an efficiency of 12.5GH/s/W at best, so it's about half as efficient as the R909. This corresponds to 80W/TH.
At stock speed, my unit exceeded the
guaranteed performance of '1.5TH from under 100 DC Watts' easily, as it pulled under 90W at 1.74TH/s. To me, achieving 1.8TH/s at 70W DC, seems very attainable with minimal voltage tuning, too.
Sidehack also talks about a 2.1TH/s machine pulling just 88W. I needed 550MHz to achieve this hashrate, and pulled 109.6W from the wall, i.e. ~98W DC. If we subtract 5W for the Raspberry Pi, there's a small difference of 5W. This could simply be due to
silicon lottery. Still a great result for a home miner!
Here's a non-exhaustive overview
[lower is better] of some popular industrial miners' efficiency over the years / generations. It's important to take into account that it is an extremely difficult process to get your hands on bare ASIC chips in the first place and reverse-engineer them in order to get them running in your own product. Home miners have always been 'lagging behind' in this regard, since none of the companies building them, develop or manufacture their own chips.
- ~100W/TH Bitmain S9
- ~80W/TH Futurebit Apollo BTC
- ~50W/TH Bitmain S17
- ~50W/TH Termius R909
- ~35W/TH Bitmain S19 / Whatsminer M30S+
- ~30W/TH: Canaan Avalon A1346 / Whatsminer M50 / M53 / M56
- ~25W/TH: Canaan Avalon A1366 / A1446 / Bitmain S19k Pro
- ~20W/TH: Canaan Avalon A1466 / Bitmain S19XP / Whatsminer M60 / M63 / M66 / Bitmain T21
- <20W/TH: WhatsMiner M60S / M63S / M66S / Bitmain S21
Final VerdictWho should buy it:
- If you want to spend about 500€.
- If you want to consume about 100W or less power.
- If you want something really, really quiet (think: one-bedroom apartment, living room mining or similar).
- If you want something reliable.
If you have a higher budget for purchase & power bills and no noise requirements, you could look into bigger gear, such as industrial miners with latest-gen chips that yield higher efficiency. I have recently heard about Futurebit's Apollo Version 2 which seems pretty efficient, but I have not tried it myself. It consumes more power, will be louder, and more expensive to buy.
Other than that, I see no alternatives to the Terminus R909.