I'll be completely honest, I'm not really sure what kano's -mine2 algorithm does. I use an older (test build) version of his driver on the production testers because it defaults to spitting out some extra data I like to have for diagnosing.

Glad you got everything working.

You might also tweak the starting frequency. It could be having trouble at the bottom end and needs to be forced over the knee past where the chip stabilizes internally and starts operating close enough to expected parameters that it ramps up on its own. When we test at the factory, the start freq is 300 and target is 400. If we don't see at least 220GH sustained for 24 hours with those settings, it doesn't go out the door.

Yep, exactly. The issue isn't total amount of heat but heat density. It's going to take a particular volume flow of air to clear a certain amount of heat from the unit, so if you increase the volume of the unit you decrease the linear flow rate of the required volume of air. Slower air means slower fans means quieter fans.

Another component of this is reducing the length of the flow path, as the exhaust end will be cooled with air already warmed by every heat source between it and the inlet. The S7 and S9 had trouble with roasting boards due to a 9" flow path. Spondoolies rack gear was even worse, with about 30" flow paths. With the S15 and onward, Bitmain shifted to short wide paths over long narrow paths, more like 6 inches, and with larger heatsinks in the hotter regions to increase the rate of thermal conduction into already hot air. The R4, specifically designed for quiet operation, had an airflow path of about two inches.

Resistive electric space heaters tend to be small because the heat-generating materials (usually basically a length of wire) are able to operate at extremely high temperatures, so the heat density of the machine is allowed to be very high. A fairly quiet fan can still do the job with heating because any air moving through will gain a substantial amount of heat, given that the goal is to pull *some* heat out of the element, not try to keep it operating below a semiconductor-survivable temperature.

Bitcoin miners generating the same amount of heat as a consumer space heater will need enough airflow to keep the chips operating at least below 100C. Ideally, even lower, as with CMOS technology there's a positive temperature coefficient for power use: as temperature increases, the power required to maintain a particular switching speed also increases. So even for a given voltage setpoint and target hashrate, as a miner heats up, it'll draw more power, causing it to heat up further, until an equilibrium is reached or failure occurs. In order to maximize profitability, the chips should be kept as cool as possible to optimize hashes per unit power.

The best way to optimize hashes per unit power is to operate the chip at a minimum stable setpoint - low volt, low speed. But this means low hashrate, and low power, per chip. In order to meet hashrate specs, or heat generation specs, it's required to use a whole mess of chips. This helps retain low density and therefore quiet airflow, but it also means a very high initial cost and a larger device that may not be trivial to unobtrusively integrate into a room.

So the design of the device will be a trade-off between the lowest initial cost (fewest chips) versus the highest operating profit (highest efficiency), coupled with convenience (lowest physical volume) versus comfort (quietest airflow). That's a lot of factors to try and optimize simultaneously.

Hello. I run GekkoScience. If you'd like to talk miners (or electric vehicles), drop me a PM.

Had to redesign hubs too, but we're prepping for a new batch. And should soon be setting up some equipment to help streamline assembly.

The original prototype batch, which was distributed to resellers for testing and publicity photos, were labeled FS7. By production, the S7 had been dropped so all production Compac v8.x are just Compac F. Version changes within the V8.x Compac F line are mostly caused by issues in the semiconductor market requiring design changes as parts become unavailable due to stock shortages. V8.2 saw a change in the LED blink circuit, and V8.3 will see a new main regulator design, which is significant since we designed a pretty kickass power system for the original Compac in 2015 and have been using it ever since. If it ain't broke, don't fix it, until you can't find the parts anymore.

Far as I know, it doesn't matter. USB3 is a nine-pin connection, with a standard 4-pin USB2 bus wired alongside 5 pins of USB3, so when a Gekko stick is plugged in it only utilizes the USB2 lines anyway. USB2-only hubs are probably getting less and less likely in the market since so much even basic stuff is USB3 now, but "basic" is still higher up the technical hierarchy than the basic primitive bus/protocol conversion chips that stick miners utilize. Far as I know, there aren't any USB3-specific adapter chips, which makes sense since USB2 bus bandwidth can already operate about two orders of magnitude faster than a basic serial line would ever need to.

In the past there's been some talk around USB3 hubs tying directly into a Pi for some reason not running sticks stably or optimally, which folks I think solved by just putting a USB2 hub inline? I've really only run USB2 anyway for these guys, since that's what our hubs are built around and even our factory testers use customizations of basically the same platform as the powered hubs we sell.

R606 has basically been discontinued. Reliability was never that good, and the mechanical parts were way too labor-intensive to manufacture and assemble. If/when there's another pod, it'll use the same basic dimensions but a different mechanical design (including heatsinks) so it goes together a lot easier. It's really demoralizing to put a lot of time into assembling and testing a device only to have it come back for warranty repair.

5kWh/month means a draw of about 7 watts to the chip. I think people still mine on S9-era gear, so let's look at S7 for the "best of the obsolete" category. Down around 7 watts, the BM1385 ASIC used will get you about 30GH/s, which is definitely not zero hashes but puts your odds of hitting a block in the neighborhood of, I think, 1 in 6 billion.

Which is probably still better than a regular lottery?

Of course, that assumes the device you're plugging into is internet-connected and can run the software to configure a pool and pipe work into the miner without a significant power hit. That's probably doable given how much crap is "smart" these days.

If PD plus data was not terribly difficult, and hubs with real power throughput were more common, and cables didn't generally suck, I'd absolutely be thinking about a one-cable pod miner using USB-C. It's probably been two years since I first mentioned wanting to do that. But all the hubs I can find are more built for charging something and also attaching like extra screens and 10g NICs and stuff. Miners have a somewhat unique power/data profile and I don't really want to build a USB3/4 hub just to make it work when, like, power bricks are already a thing.

But a one-cable pod miner would be pretty sweet.

Cool so we'll switch to a less mechanically rigid connector and tie it into a more complex frontend that digitially negotiates a high voltage low current power and pushes that into a less efficient main regulator while leveraging a data bus designed for multi-gigabit transfer converted to a sub-megabit primitive protocol.

I mean, maybe if I didn't already have a bunch of other stuff to do I'd think about it, but I already have a bunch of other stuff to do so I would kinda rather keep doing what's worked so far. Especially since the price tag of that 3-port fancy hub you linked is about equal to what some people mentioned was an over-price for the badass-for-power seven-port regular hubs also available.

Not to say nobody else should be doing this, just that Gekko probably isn't doing this.

I can almost guarantee it isn't, just that the listing pictures and stuff are super old.

Wow that's an old listing. We stopped using the black plastic housing material after the first batch in 2018.

Does it deliver 12A per port for that 60W, or does it deliver 20V 3A?

Every time this idea has been mentioned in the last couple years, it comes down to that USB-C spec doesn't allow for both Power Delivery and data transfer at the same time. I didn't follow up on any of your links but unless that changes, it's a no-go.

Temperature is measured at very few points on the board, so any one individual chip could be smoking red hot and it wouldn't show up on the reading.
And while one chip can run without a heatsink, it'll probably fail pretty quickly. And depending on how it fails, it can be fairly meaningless or it can cripple the whole board.

Your point stands though that the backside heatsinks aren't nearly as significant as the chipside heatsinks for effective cooling, and in fact several are missing right up against the fan, by design.

Actually if they overheat, parts start to fall off.

There's no real "this temperature is fine, this is bad" standard metric for these guys. But if your somewhat desensitized fingers can stand them, they're not too hot. Basically, they're probably safe until you start to worry that they're not. Or parts start to fall off. That's definitely hotter than safe.

They're bench-calibrated with power On at the factory. I tweak them with power On. Fuzzy, what's your specific reason for not doing that?

It's about the same process for all my sticks going back to the original Compac. You fire up cgminer set to the clock speed you want it to run at, and if it runs smoothly (basically, not a lot of errors and the hashrate should be close to GH=0.672*MHz) you can dial back the voltage a bit by turning the knob anticlockwise. Repeat until it starts to run poorly, and now you know where the threshold is so turn it back up a little (knob clockwise).
If it runs poorly from the start, turn voltage up (knob clockwise) until the numbers smooth out.

The most efficient mining will occur at the lowest voltage setting at which the numbers are stable.

If you're looking to maximize hashrate at a particular power draw, tune the stick to a particular hashrate and check the power use. If it's under your target, tune again to a higher speed.

Eyeboot asked last year if I'd be interested in designing a new router style device better than the R1. It's an interesting idea, but I've had a lot of other stuff going on. This year should be good for new projects. Who knows, might get around to a better mining router later on, one where the miner part is upgradeable.