Test pads. 1V8 for measuring the 1.8V logic voltage, 0V9 for measuring the 0.9V PLL voltage, and RESET resets the ASIC. Most people will never need those but they're handy for troubleshooting.
|
|
|
That's what I was thinking too, why I was recommending he test using only one board at a time.
|
|
|
Well, I definitely ignored this thread for about two weeks. Whoops. Followed up to most of y'all.
|
|
|
So I mentioned earlier that the software guy is working on a test stick around a 14nm ASIC. I just found out the chip supplier won't be selling more chips, so it looks like the few I can build with sample chips will be the only ones to ever exist.
The software guy is gonna keep working on a driver because the hardware is pretty well working already.
The payment arrangement with him was a percentage of sales. But it looks an awful lot like that's not gonna happen, so I plan to auction off the few functional sticks I can build with sample chips in order to get him paid and cover some of the other dev costs.
I'll probably put up some other failed prototypes (A3218 and BM1385 based) and maybe some pod tests and such.
Not sure yet when that'll happen.
On a "good news" front, I did just send a rough hardware layout to Novak, who will be working on the controls for the new 16nm stick in progress. The control setup should port over pretty readily to pods and full-scale miners with the same chip. Hopefully this time the supplier doesn't fall through. So, our hopes for having a 100W or less miner with 16nm chips is still alive.
|
|
|
Did you try powering the controller and only one or two boards at a time, to make sure it's not the PSU overloading?
|
|
|
Unless the IO board has a nice high-current 3.3V buck regulator on it, and the FPGA on there functions as an IO expander.
|
|
|
Did you lose some time in there? At 330MHz I'd expect to see a combined average around 54GH and a per-stick WU about 250.
|
|
|
Depends on the software guy. He's actually already integrated BM1384 Compac into the Icarus driver, shouldn't be an issue being able to compile new stuff alongside that.
I am also super excited. Hopefully stuff works.
|
|
|
The hardware discussed in this thread will cover your Molex needs. Something like a PicoPSU would power the motherboard. I can put you together a kit price estimate if you like.
|
|
|
Not sure I follow that logic. For a switching regulator, power out equals power in times efficiency (usually at least 80%) and for a linear regulator current out equals current in. There should never be a condition where a voltage-reducing regulator intakes more current than it outputs (except for edge conditions like where output current is dwarfed by regulator supply current requirements or where the vout/vin ratio on a switcher is higher than conversion efficiency).
|
|
|
From a designer standpoint, it's a tradeoff between initial cost and operating cost. More chips at a lower frequency will get you a higher initial cost per hash (chips being the largest single expense on a miner), but at a lower power draw per hash so reduced power cost over time and, in the end, a longer viable life because of this. Fewer chips at a higher frequency will reduce the cost to build it so in theory the miner is cheaper, which for greedy entities means a higher profit margin when you sell it for the same price. The increased power draw per hash also reduces the viable lifetime, which for greedy entities means faster obsolescence and increased sales of the next generation.
Pretty sure this is was the driving consideration behind the switch from 54-chip to 45-chip boards in the S7, which maximizes profits for the manufacturer at the consumer's expense.
I like going with a middle ground, stack a decent number of chips at less than peak clock and let the user overclock (possible with overvolting and improved cooling) if he wants to to increase initial hashrate vs investment cost, and still have the ability to undervolt and underclock to run chips colder to extend the viable life. That's about the best way I know of to maximize the consumer's returns.
|
|
|
From what I understand, some risers take in 12V only and have an internal 3.3V regulator that steps down the 12V for the socket. These do not need my board. Risers which take in 12V from a Molex or SATA and step the 5V down to 3.3V, like MarkAz uses in his builds, could benefit from it. There probably is no actual 5V going into your GPU, just 12V and 3.3V
I don't know enough about ribbon risers to know what voltages (and at what allowed currents) they pull from the socket.
|
|
|
There is no relation between a pod miner and any S1/S3/S5 hardware.
Anything mounting to S1/S3/S5 chassis will not be a pod.
|
|
|
I'm working on a little microcontroller PWM driver today which will come in handy for three different projects. Maybe if I have some free time next week I'll see about coding a duty cycle detection onto it and make a max-fan-speed inline module.
If you wanted to circumvent the auto fan entirely, you could probably use a standard fan speed controller and just make sure the tach line is plugged in so your miner controller still sees the fan as spinning.
|
|
|
A new Compac may be forthcoming depending on chip sources and the software guy currently working on a test stick.
I got some other sample chips and just got datasheets for them yesterday, and have already started scheming with software guys about what we want to do for a 1-chip stick and probably a 10-chip pod (estimate in the 80W peak power range) whose concepts will transfer pretty readily to a full S1 refit board. I won't set a tentative ETA because there are entirely too many variables, but I would really really like to be selling stick miners and maybe have some pods by Christmas.
|
|
|
You can buy the miner and have it shipped here, and I will host it. I charge monthly, don't usually ask for extended contracts. Anyone hosting with me is free to have his miner shipped anywhere at any time.
|
|
|
And everyone else helping out. To be fair, I'm just the project hub and nothing would get done if there weren't three or four other people pitching in on different parts of it.
|
|
|
I cannot announce anything official, but let it be known I am working on both ~100W pod and new sticks with 14/16nm chips.
|
|
|
Probably the only way around it is to intercept the PWM signal to the fan and route it thought a circuit to impose a maximum duty cycle. If you use the rising edge to trigger a pre-timed one-shot and AND that with the original pulsetrain, the lower duty cycle will win out and it's all done in hardware but that won't fix 100% duty cycle. Using a microcontroller to sample duty cycle and output a pulsetrain of the lesser of preset max duty cycle and controller-mandated duty cycle would give you a maximum.
Hm, I wonder if that would be worth making anyway.
|
|
|
Everyone who's asked about stuff I'll PM with prices today.
I just got word the PCBs should be shipping out October 10th, so factoring in transit time and assembly I should start shipping orders week of the 17th.
|
|
|
|