Topic: New Official AMT Thread - page 92. (Read 149472 times)

Gave me an idea... lol... throwing the non-responding card back in...

I wonder how the power internal to the chips is bussed. One can assume that all Vdd pads are tied together at various points, question is, how much do they rely on the main feeds from the pads to be in balance to keep from burning out spots in the internal traces? I would think they can tolerate a local 2x overload at best.

eidt: looking at that that kemet info is great - very good coverage on exactly we are talking about on bypassing

If you have a scope it would be interesting to see what the 'filtered' Vdd feeding each side looks like. Betcha the voltages are different. Ones w/o caps will probably have bigger dips & spike at 2x the clock freq.

And def right on the heatsinks. That is prob #1. My thing on the power comes from the industrial 24x7x365 running mindset. It is sooo easy to do it right the 1st time and be done with it rather than fart around with 'adequate'. Bites ya in the ass every time.

Would be nice to have a schematic of the circuits. The myrid of smaller caps are either across data lines for shaping the data eye or more likely there to knock out very high freq ripple. Is not just the total value of capacitance use but how it is achieved.

A capacitors internal reactance changes with frequency (think crossovers). Large value caps great at lower freq but roll off as it climbs. So is very common to also use several smaller caps to handle the higher freqs. What is the clock freq in the A1? They need to filter 2x that frequency so the switching transitions can be dampened out.

Also is why high freq switching regulators use all them little caps vs a couple big ones.

The capacitors are just there to modulate the voltage,  it is not really critical that you have a little less than normal. 

I think the main problem has been the heatsinks.  The top heatsinks are too small,  the bottom heatsinks have not enough thermal conductivity due to lack of paste.

Arg... I hate when I do it backwards... No wonder I can't ever get my woman pregnant! xD

Well, in that case, since I was backwards... the capacitance is half what it should be. Can't believe I reversed parallel and series.

I know that a screw has a resistance of near zero... lol, which was across both pads, where the resistor should have been. Thus, nullifying the existence of the cap, as that was a direct short to the + terminal side, from the - ground.

Willing to bet that each cap is supposed to feed one of the big Vdd pads. And no such thing as overkill at this. Unless the copper plane are like 20mil thick, without that local bypassing you are going to see at least several 10's of mv unbalance and probably pretty weird spikes/dips/ ringing.

Bzzzt. Other way around. Caps in parallel add together. 1+1=2 ect. In series they divide. Resistors of course do oposite. series adds, parallel is 1/((1/r1)+(1/r2)+ etc.)

Yea, I just looked-up the R+ ones...

Interesting that they have caps in the same shell as a resistor. I just assumed the 6K was Ohms, not Farads. Odd... (KEMET I.D. Symbol) Figures. xD

So they used the CAP, on only one side, but not the resistors... (The resistors would have just bled-out the caps, and added a base-line voltage, where the caps didn't have the full charge.)

May have been overkill... but the thing with caps, is that the farads half when in parallel, they don't add... Thus, cutting one out would have doubled the farads, with half the available amps. Without the resistor, they would just keep filling and providing over-voltage... Might have been the first design, where they thought they could operate fine without those other components, then realized they couldn't, so they added them back. That changes the frequency absorption also. (This I know from making passive and active crossovers for speakers, and noise/notch filters.)

Funny that they have so many smaller caps around them, which could all be one larger cap. I assume for frequency dipping and filtering. However, without the other three connections, there is almost no need to have them on the other side at all. Unless they all share that common plate. (I assume they do. Have not attempted to probe anything yet, until I get my second unit up and running.)

Along these lines, Are they still just paralleling a heap'o fet buck regulators into one common set of power planes? If so, for god sakes why???

The same company that make the one in the 2-chip reference design also makes a 16ch one.  For 8 chips you could give each its own independent Vcore supply with current monitoring for mainly only the cost of 4 more mosfets and 2 more buck inductors... If a chip fails hard the regulator kills power to it and tells ya. Not rocket science here.

Sumptin' ain't right there...

The KO come up as SMD caps from Kemet corp. https://www.pa.msu.edu/ftp/pub/d0/run2b/l1cal/hardware/component_information/kemet_tant_and_ceramic_caps.pdf And they damn well all better be conmected to or big trouble balancing power into the chip! As in probably some Vdd pads pulling different currents leading to hot spots.

The R+ are 10k resistors. Hope folks dinna get them mixed up...

Hopefully each cap feeds/is a bypass for each one of the 4 big Vdd pads on the chip. Wonder if they also feed the 3 other small ones? More incomplete assembly I'd say.

These connections... Labeled "COUT", filled with "KO 337 6K 345" SMD resistors...

Though, on one card there are four...
2x "KO 337 6K 345" TOP-RIGHT and BOTTOM-LEFT...
2x "R+ 107 10K 402" TOP-LEFT and BOTTOM-RIGHT...

On mine, there is only one per chip...
Of 8 chips, 6 have the "KO 337 6K 345" on the BOTTOM-LEFT...
The last two, position 7 and 8, have that same resistor only on the BOTTOM-RIGHT...

That is the card that just stopped responding for no reason... the one with obvious screw-threads dug-in to the solder, on the #7 chip, where the "KO 337 6K 345" should have been, on the BOTTOM-LEFT. (Instead, that is on the BOTTOM-RIGHT, on that chip, and the one above it, #8)

These, outlined in this photo... I have only 1 of four of these, on each chip, on two of my cards actually. I have not tried the other one, due to the lack of a heat-sink on it.

eep... Contact BitMine.ch and see if they will exchange or offer any help? Long shot but can't hurt to try. Would really open them to um, other exposure though.

For now, hate to say but maybe eco-mode it to 20GH/chip??

Wait, gimme the heat-sinks.. I have three boards without heatsinks! xD

P.S. If you send the boards to some guy, I think he will mount them on blank boards made by technobit.

My AMT boards are pretty much next to dead. I have 3 working boards. And I will consider myself fortunate if I even get replacements at this point.

+1 for probably rather satisfying creativity

I am going to make a video of me throwing these boards off a 15 story building maybe it gets good hashing power on the way down lol

Are you referring to the Vcore regulator plane feeds to the A1 chip?? Since the chip has 7 Vdd pads total around its edges that dunna make any sense...

eep!. Just boards by themselves should have survived quite well even if just 'reasonably' packed in a box with some anti-static bubblewrap around each of them... More FedEx Football Follies?

Yeesh! Seems some folks in the pool I use picked up some serious hashing pharm power...
This round has been running only 15 minutes.

Rank    Name    Round Shares    Rate this Round
1    aerobatic    32997979.57216952    107367.61 GH/s
2    Private    21365524.20419547    69518.35 GH/s
3    Private    9269397.08919503    30160.42 GH/s
4    Net151    7655654.88623410    24909.69 GH/s
5    xenominer27    3651731.13999998    11881.87 GH/s
6    Private    2746939.15535708    8937.89 GH/s

107 TH. woof.