Pages:
Author

Topic: [GUIDE] Undervolt antminer s1 [1.19W/GH at the wall] - page 19. (Read 69365 times)

newbie
Activity: 42
Merit: 0
Before mod my antminer was consuming 348W on the wall on 180GH/s but now it takes only 260W on 165GH/s.

Assuming $0.20/kWh
Without undervolting: Daily $6.04 mined minus $1.67 electricity = $4.37 daily profit
After undervolting: Daily $5.53 mined minus $1.20 electricity = $4.33 daily profit

Right now we are at the point where undervolting starts making sense for those with expensive electricity.
I am one of those. Just want to know the risks before I start doing it.

I read about undervolting CPUs that the only risk is less stability, i.e. it may reboot now and then. I guess for Antminer that a symptom of too low voltage is high HW error. Right?
hero member
Activity: 728
Merit: 500
Thanks couldn't see how to post the picture directly in the reply.


When you post, over the window you type, there some small icons.
On the left ones , you will find the one for posting pictures urls.
member
Activity: 119
Merit: 10
Thanks couldn't see how to post the picture directly in the reply.
hero member
Activity: 728
Merit: 500
Added link to my pictures of the 10k pot addition to S1's.


http://imgur.com/a/8SUO8





Now looks better.
member
Activity: 106
Merit: 10
🤖UBEX.COM 🤖
Added link to my pictures of the 10k pot addition to S1's.

http://imgur.com/a/8SUO8

Nice job!
*looks for a pencil*
member
Activity: 119
Merit: 10
Added link to my pictures of the 10k pot addition to S1's.


http://imgur.com/a/8SUO8
legendary
Activity: 2114
Merit: 1005
ASIC Wannabe
if you are going to trouble yourself to soldering a solid resistor then it would make more sense to solder A POTentiometer and have the flexibility of adjustable resistor settings.



That would be ideal for fine tuning but for being novice at soldering, wiring in a pot adds more difficulty than I am comfortable with. Tacking on a resistor to the sides of the existing resistor seems a little more doable for my skill level.  And it's only $5 vs about $55 for the pots.

But, now that I think about it more, pencil would work just fine. I wanted to add the resistor so I wouldn't have to worry about the graphite potentially flaking off. But even if it did, having them hooked to a kill a watt would show a rise in electricity use.  Then I would just have to check each voltage to find the culprit and just add more pencil.

go with a pencil mod - i did it for my bitfury cards (exact same regulator with 2 voltage-controlling resisters) and it works great. It isnt worth the effort and risks of desoldering pcb components and swapping in new resistors when a few rubs with a pencil works fine. tips:

1) get an artist pencil such as a 3B instead of a 2. It is a bit softer and leaves a tiny bit more soft graphite for a rub - a little bit easier to tweak the resistance this way
2) start small. Over time, the pencil mod will 'age' and slightly decrease the resistance even frther than when it was first applied, as the graphite is baked by the hot pcb. leave a little bit of headroom when you do a mod, and expect the resulting voltage to drift a little bit over a 4-12 hour period. (drifts up if overvolting and down if undervolting)
member
Activity: 106
Merit: 10
🤖UBEX.COM 🤖
if you are going to trouble yourself to soldering a solid resistor then it would make more sense to solder A POTentiometer and have the flexibility of adjustable resistor settings.



That would be ideal for fine tuning but for being novice at soldering, wiring in a pot adds more difficulty than I am comfortable with. Tacking on a resistor to the sides of the existing resistor seems a little more doable for my skill level.  And it's only $5 vs about $55 for the pots.

But, now that I think about it more, pencil would work just fine. I wanted to add the resistor so I wouldn't have to worry about the graphite potentially flaking off. But even if it did, having them hooked to a kill a watt would show a rise in electricity use.  Then I would just have to check each voltage to find the culprit and just add more pencil.
member
Activity: 119
Merit: 10
If you are going to pull off the resistors you had better have one big ass magnifying glass.
I've just completed this on one of my S1's to play with (able to put 0.6V to 1.2V to VDD of Bitmains chip now).
I took some pictures and I'll try to get them on later today.

sr. member
Activity: 406
Merit: 250
if you are going to trouble yourself to soldering a solid resistor then it would make more sense to solder A POTentiometer and have the flexibility of adjustable resistor settings.

member
Activity: 106
Merit: 10
🤖UBEX.COM 🤖
Thanks for the guide, OP. I plan on sucking every last ounce out of profitability out of my S1s. This will definitely come in handy in a couple months.

Instead of pencil, I plan on a more permanent solution by soldering a 9.1K ohm resistor in parallel. I'm very new to soldering, so I won't dare to try to replace the flat resistor. I plan on just tinning the ends of the 9.1k resistor, hold it to the side of the flat resistor, then hitting the flat resistor with the iron to get it to tack. Then, repeat for the other side. Is this correct? Any suggestions would be appreciated.

Also, is a 9.1K ohm resistor correct?

r2 = (r1 * rtotal)/(r1 - rtotal)
r2 = (4.47K * 3K)/(4.47K - 3K)
r2 = 9.12K ohms

Thanks
newbie
Activity: 42
Merit: 0
What are the risks of undervolting?
I don't know the physics of it, so please excuse my ignorance  Cheesy

1. If I undervolt but do not underclock, what will happen? Will HW errors simply increase, or do I risk the antminer getting damaged?

2. If I both undervolt and underclock, but it turns out I undervolted too much, the same logic applies as in #1?

3. I do not have a volt meter (and I dont know how to use one). Can I simply try with a little bit? The risk is applying too much graphite, not too little, right?
legendary
Activity: 2114
Merit: 1005
ASIC Wannabe
Has nobody overclocked or overvolted an S2 yet?

640 chips (an S1 has 64 chips), 16 chips per TPS53355 regulator  (S1 has 8 per)
S1 was forcing around 30A through the regulator at stock speeds, and up to 40A with overclocking and some overvolting (based on some of the 210GH+ reports in this thread)
S2 is putting about 22-24A on the regulator at stock. Presumably with a bit of voltage modding and overclocking the regulators could handle up to about 1.4TH/2kW

only issues:
1) power. To seriously overclock, adding a 600W+ PSU will be necessary
2) cooling. More fans and/or individual chip heatsinks are necessary to properly exhaust any extra heat since the units read 50C+ temperatures stock
2b) PCI entenders. Looks like the boards are all mounted using PCI (PCIe 16x?) slots. if you used some risers (with a nice pair of 14AWG power cables to protect the risers) you could get the boards up out of the case and arrange them with airflow directly aimed at the heatsinks for better cooling.

I can't wait for my batch 2 unit to arrive - if noone overclocks by then I might have to be the first. (donations will be accepted in case i kill my miner :p)
newbie
Activity: 11
Merit: 0
I have undervolted my Antminer to 1,0V (R3 = 3.9kOhm) and I am really happy with results. Before mod my antminer was consuming 348W on the wall on 180GH/s but now it takes only 260W on 165GH/s.

So 8,3% performance decrease means 25.3% power input decrease. Nice Cool

I am running about hour now and no HW errors at all. I also tried 180GH/s with lowered voltage, but a lot of HW errors resulted in 170GH/s output performance too.

Mod was really easy. 10 minutes of work and everything worked on the first try.

Thanks for the guide.

SVK
sr. member
Activity: 378
Merit: 250
Many thanks for a guide.
I see that yous are modifying S1 left and right  Grin

I have one spare blade with half of the chips not working and was wondering if there is any way I can use this half broken board ?

It's the board which didn't have controller board attached to it. Is there any way to connect that board to RPi or even to existing and working S1 ?

Thank you very much
 

there may be a way to connect it to an upcoming S2 unit if they simply modified the voltage/clocks on the current blade design.

Many thanks for response Klondike Smiley
I'm incapable to try anything like this on my own so thought it might be useful to ask on here. So many gurus on here Smiley

Well I will keep board just in case Smiley

Yeah, i cant think of anyway that you could connect it to an S1. AFAIK neither:
a) cgminer settings are capable of detecting more than 2 boards
b) 2 cards cannot share a dataline for double hashrate. you would either see a lot of duplicate throwaways or HW errors. I do not think it is a 'shared comms' type of connection where multiple boards in parallel would be capable of sharing instructions or hashing 'n-times' faster

I was looking at possibility to solder USB on it or something else and then run it through RPi.
legendary
Activity: 2114
Merit: 1005
ASIC Wannabe
Many thanks for a guide.
I see that yous are modifying S1 left and right  Grin

I have one spare blade with half of the chips not working and was wondering if there is any way I can use this half broken board ?

It's the board which didn't have controller board attached to it. Is there any way to connect that board to RPi or even to existing and working S1 ?

Thank you very much
 

there may be a way to connect it to an upcoming S2 unit if they simply modified the voltage/clocks on the current blade design.

Many thanks for response Klondike Smiley
I'm incapable to try anything like this on my own so thought it might be useful to ask on here. So many gurus on here Smiley

Well I will keep board just in case Smiley

Yeah, i cant think of anyway that you could connect it to an S1. AFAIK neither:
a) cgminer settings are capable of detecting more than 2 boards
b) 2 cards cannot share a dataline for double hashrate. you would either see a lot of duplicate throwaways or HW errors. I do not think it is a 'shared comms' type of connection where multiple boards in parallel would be capable of sharing instructions or hashing 'n-times' faster
SVK
sr. member
Activity: 378
Merit: 250
Many thanks for a guide.
I see that yous are modifying S1 left and right  Grin

I have one spare blade with half of the chips not working and was wondering if there is any way I can use this half broken board ?

It's the board which didn't have controller board attached to it. Is there any way to connect that board to RPi or even to existing and working S1 ?

Thank you very much
 

there may be a way to connect it to an upcoming S2 unit if they simply modified the voltage/clocks on the current blade design.

Many thanks for response Klondike Smiley
I'm incapable to try anything like this on my own so thought it might be useful to ask on here. So many gurus on here Smiley

Well I will keep board just in case Smiley
legendary
Activity: 2114
Merit: 1005
ASIC Wannabe
Many thanks for a guide.
I see that yous are modifying S1 left and right  Grin

I have one spare blade with half of the chips not working and was wondering if there is any way I can use this half broken board ?

It's the board which didn't have controller board attached to it. Is there any way to connect that board to RPi or even to existing and working S1 ?

Thank you very much
 

there may be a way to connect it to an upcoming S2 unit if they simply modified the voltage/clocks on the current blade design.
SVK
sr. member
Activity: 378
Merit: 250
Many thanks for a guide.
I see that yous are modifying S1 left and right  Grin

I have one spare blade with half of the chips not working and was wondering if there is any way I can use this half broken board ?

It's the board which didn't have controller board attached to it. Is there any way to connect that board to RPi or even to existing and working S1 ?

Thank you very much
 
legendary
Activity: 2114
Merit: 1005
ASIC Wannabe

awsome mod.

I applied it and set the clock at 325.   

165 Gh pulling 277watts at the wall

normally 190gh pulls 380watts

Good to know.  This mod makes the S1s theoretically cheaper per Ghash than the S2 without sacrificing much power consumption.

According to Jim above's volt mod experience S2 are still much better for power consumption for not much more BTC which i foresee batch 3 to be priced around $3000.

1000gh / 165 = 6.061 x 277 wall watts = 1678 watts

6.061 x S1's = 5.41 BTC

Now add the cost of all the power supplies needed for 6 x S1's and the S2 batch 3 will clearly be the smarter option by far.

only keep in mind that for the next 2-3 month window, having 1.2TH/2.4kW is better than 1TH/1kW.
Pages:
Jump to: