[GUIDE] Undervolt antminer s1 [1.19W/GH at the wall]

[klondike_bar]

so i tried undervolting, have 0.73-0.77 measured on all chip sections, but have tried a few frequencies between 193 and 250 with mixed results.

193-200 sems to cause reboots and only one side of chips recognizes, but does not hash at all in the first few minutes, then beeps and restarts
218-225 seems to cause about half the chips to work, for 60-80GH
250 caused errors and almost all chips x out

any suggestion? im using an 8B pencil with reasonable precision.

update: seems to be running pretty well at 212.5MHz and 65 timeout. started up with some errors, but after a reboot its now seeing all chips and producing ~120GH (no kill-a-watt handy though)

just curious - how many people are moving these to smaller/no heatsinks once undervolted? I see Chaositec doing them in a tight stack at ~50GH (~45w?) per board, and I really want to do something similar (mostly to free up some fans to replace other noisy fans on rockminer gear)

[1714968841]

1714968841

[1714968841]

1714968841

[1714968841]

1714968841

[terrapinflyer]

I think I may be missing something here I have not been able to reduce the power draw after completing the pencil mod.

My resistors have been modded to:

Code:

R3 2.96|3.0
R66 2.90|3.02
R38 2.87|2.87
R52 3.05|2.80

.86v

Code:

        option 'freq_value'     '0A81'  #275M
        option 'chip_freq'      '275'        
        option 'timeout'        '45' 

Hashing at 140gh~ with .5% error rate but still pulling 202w from the wall.  

That is reduced power draw. You'd be pulling ~280W at the original voltage for 140GH/s.

Doh!  

What should I do to get it closer to 150w? Keeps lowering the resistance?

I doubt you will get to 150W @ 140GH/s, but ~180W should be possible. Just keep lowering it.

So I have been having some mixed results here too.

I was able to get 3 ants down to .8v by reducing resistor values to 2.5ohms @ 200mhz have been hashing at ~111gh/s-~100gh/s @ 120w-125w at the wall.

Others did not like being that low and would not hash. I had to reset to .84v @ 200mhz but are hashing 100gh/s @ 135w

When I move them all from my workbench to their final homes I will re-measure them all and put the data in a spreadsheet to share with you all.

[klondike_bar]

I haven't yet gotten out a wattmeter, but I have 4 units stable at resistance of 2.05-2.15 kOhm, or about 0.74-0.77V. 193-212 MHz gives 93-112GH per unit

[terrapinflyer]

I haven't yet gotten out a multimeter, but I have 4 units stable at resistance of 2.05-2.15 kOhm, or about 0.74-0.77V. 193-212 MHz gives 93-112GH per unit

Are you seeing lower power draw at that voltage?
Im still struggling to get below 1.26w

Oh for you guys wondering if you can stack these after modding I did this last night.



Two ants down the the size of one hashing 200mhz 2.8kohm .83w ~200gh/s @ 253w . Im working on putting 5 together today, defiantly helps with fitting them in the office for the winter.

I set up the fans in a push pull and they are keeping the chips between 40c and 47c. It is a bit louder then before but that was expected. Im sure that if I used different fans it would be much quieter but I dont really mind.

[helipotte]

Use jumper wires between blades.  This reduces the number of pci-e power connectors needed.  I made small jumpers and ran them from the yellow/black screw terminals.

One pci-e 6-pin will easily handle the load of two blades at these levels.

[terrapinflyer]

Use jumper wires between blades.  This reduces the number of pci-e power connectors needed.  I made small jumpers and ran them from the yellow/black screw terminals.

One pci-e 6-pin will easily handle the load of two blades at these levels.

Nice tip thanks!

[terrapinflyer]

I haven't yet gotten out a wattmeter, but I have 4 units stable at resistance of 2.05-2.15 kOhm, or about 0.74-0.77V. 193-212 MHz gives 93-112GH per unit

Could you share your values for 193mhz please? I can get the two I re-applied graphite on to hash at the 212mhz but get dead chips at 193mhz.

Im also still getting ~130w at the wall on each unit and hashing ~100gh at .76V and 212mhz. Nothing I do seems to get me any closer to 1.19w/gh, so any help would be greatly appreciated.

[klondike_bar]

I haven't yet gotten out a wattmeter, but I have 4 units stable at resistance of 2.05-2.15 kOhm, or about 0.74-0.77V. 193-212 MHz gives 93-112GH per unit

Could you share your values for 193mhz please? I can get the two I re-applied graphite on to hash at the 212mhz but get dead chips at 193mhz.

Im also still getting ~130w at the wall on each unit and hashing ~100gh at .76V and 212mhz. Nothing I do seems to get me any closer to 1.19w/gh, so any help would be greatly appreciated.

tbh i have had the same troubles with frequencies below 200MHz - they just dont handle well and i see lots of dead chips. i dont think its a voltage issue, but rather something related to the controller. at 200MHz timeout values as high as 60-65 are required, and timeout of 70+ results in no hashing at all (for any frequency - not sure if this is replicable using the cp2102 and a newer cgminer)

tbh, 200-212.5mhz is more stable for me, and I run them at about 2.05-2.20 kohm (0.74-0.77V)

[CHAOSiTEC]

I haven't yet gotten out a wattmeter, but I have 4 units stable at resistance of 2.05-2.15 kOhm, or about 0.74-0.77V. 193-212 MHz gives 93-112GH per unit

Could you share your values for 193mhz please? I can get the two I re-applied graphite on to hash at the 212mhz but get dead chips at 193mhz.

Im also still getting ~130w at the wall on each unit and hashing ~100gh at .76V and 212mhz. Nothing I do seems to get me any closer to 1.19w/gh, so any help would be greatly appreciated.

Code:

desired setpoint frequency? 193
BS: 1 M: 30 N: 0 OD: 2
Hex value: 4f02
 Resulting actual clock: 193.750000
 Error from desired clock: 0.388598 %
 Expected hashrate output: 99.200000 GH/s
 Best estimate timeout: 70

option 'freq_value'   '4f02'
option 'chip_freq'    '193.75'
option 'timeout'      '70'

also, take a look here:
https://bitcointalk.org/index.php?topic=857967.0

[CHAOSiTEC]

well, got my ant blades down below 1 W/GH


254 GHS vs 241 W :-D (5x ant s1 blades) = 0,949 W/GH

[sidehack]

The best machine I've worked up so far ran 140GH (275MHz) at 152W DC (including fan and controller) with 1%HW errors. Most of the ones I've checked were about 165W DC, with a median around 0.84V for stable operation. Doing resistor replacements, not pencils.

[CHAOSiTEC]

The best machine I've worked up so far ran 140GH (275MHz) at 152W DC (including fan and controller) with 1%HW errors. Most of the ones I've checked were about 165W DC, with a median around 0.84V for stable operation. Doing resistor replacements, not pencils.

im running with 0.75[0-8] Vcore on each blade, 200 MHz, 0.0027 - 0.157 % HW errors, connected to rm850 taking 241 W at the wall

[sidehack]

I'll crank mine down that far in a few months. I bet with the heatsinks on they'd run fanless if you sat 'em vertical on a wire rack for convection.
Are you tuning your banks individually, or just using the same voltage on all of them? I'm adjusting each bank as I go until the errors meet a threshold, then moving to the next bank. The entire machine, then, is running pretty much as efficient as it can for a given clock speed.

[CHAOSiTEC]

I'll crank mine down that far in a few months. I bet with the heatsinks on they'd run fanless if you sat 'em vertical on a wire rack for convection.
Are you tuning your banks individually, or just using the same voltage on all of them? I'm adjusting each bank as I go until the errors meet a threshold, then moving to the next bank. The entire machine, then, is running pretty much as efficient as it can for a given clock speed.

these are just blades (5 of them stacked with spacers), no heatsinks, but i have 2 fans in push/pull on them, i start out with setting them each at 0.75, then adjust until im satisfied with speed and error rate

[sidehack]

I haven't messed with condensing units yet. They're all at the shop, so there's no lack of space to stack 'em. Right now I have 12 stacked up pulling from a single DPS2K PSU, one 16AWG pair per blade. One of the units is clocked at 225, and its fan is spinning so slowly you can just about count the blades. Pretty cool they'll run without heatsinks, but those settings are <1W/GH and not terribly chip-dense on the boards so it makes sense.

[wpgdeez]

I've got to get off my lazy ass and mod my 20 S1's I've got sitting idle!

[CHAOSiTEC]

current info: running time - 15 minutes
GHS: 473.69
Watt: 469 at wall

when i start it up several of the blades produce a lot of errors, but it equals out over time...

i will post stats after 6 hours again

[klondike_bar]

whats your resistance/voltage/freq/timeout settings like?

Ive found that 2.05-2.15 kohm is well suited to 206.25MHz/60 timeout, but it can vary a bit between boards.

[CHAOSiTEC]

whats your resistance/voltage/freq/timeout settings like?

Ive found that 2.05-2.15 kohm is well suited to 206.25MHz/60 timeout, but it can vary a bit between boards.

im using 1k85 (aprox) because each blade is optimized by them selves to reduce HW running at 200 MHz/60

[klondike_bar]

whats your resistance/voltage/freq/timeout settings like?

Ive found that 2.05-2.15 kohm is well suited to 206.25MHz/60 timeout, but it can vary a bit between boards.

im using 1k85 (aprox) because each blade is optimized by them selves to reduce HW running at 200 MHz/60

thats actually really helpful to know - I thought you were operating closer to 2k4 for some reason. that compares nicely against my finding that 2k1 is well suited to 212.5Mhz/60 and 2k is suited to 206.25Mhz/60

maybe i should push mine even lower then.