Author

Topic: [Antminer S1] - Pencil Voltage Mod guide (Read 5558 times)

legendary
Activity: 2450
Merit: 1002
March 21, 2014, 09:51:43 AM
#29
Nice work GenTarkin!

Really interested in undervolting to the extend the useful life of our S1's down the road. 

1.  Would increasing the resistance at R12 have the desired effect?
2.  How far can we go?
3.  How did you calculate the resistance target?

Ive not tried undervolting but theres a thread where someone is going to do that w/ the R12 resistor =) They are gonna attempt .75v @ 200mhz or so ... iirc'

I didnt calculate resistance really, I just dropped the resistance about 1k at first on the R3 using a pencil and then booting the antminer up and measurig ASIC voltage.
Based on that result, I further experimented w/ more & less resistance accordingly.
sr. member
Activity: 252
Merit: 250
Cryptomancer
Nice work GenTarkin!

Really interested in undervolting to the extend the useful life of our S1's down the road. 

1.  Would increasing the resistance at R12 have the desired effect?
2.  How far can we go?
3.  How did you calculate the resistance target?
legendary
Activity: 2450
Merit: 1002
Also, Im wondering, what exactly draws the current on that particular buck regulator. Just the ASIC chips VDD? because, if so, at 1.1v the (x8) ASICs draw far less than 30A on their VDD, and based on the increase of Amps shown in the datasheet, figuring 1.2v would put the (x8) chips @ 3.9A or so, would only draw just above 30A
legendary
Activity: 2450
Merit: 1002
... the avg voltage has climbed to 1.17v on all chips - think theres a settling period or something..

You're very likely seeing the resistance of the pencil carbon change as it warms and cools. Once you've established the ideal nominal resistance you might want to replace the resistor/pencil for temperature stable resistors. Could also just be carbon flaking off?

As far as overloading the parts is concerned, that little buck regulator is only rated to ~30A, but is capable of supplying much more (core is rated to 40A IIRC). It has built-in thermal protection so the chances of frying it are low - instead of smoke you would instead have thermal shutdowns until the temp returns to 'safe' level.

These shutdowns can be mitigated or even avoided if heat sinks are applied to the back side of the regulator, (which is on the back side of the PCBs obviously). We did this pretty extensively when modding bitfury boards, which use the same part: https://bitcointalksearch.org/topic/m.3152210

Thanks for the informative post! ... been looking for an answer like this regarding that IC
I figured its current handling capabilities were more closely dependent on the temperature it reached.
sr. member
Activity: 434
Merit: 250
... the avg voltage has climbed to 1.17v on all chips - think theres a settling period or something..

You're very likely seeing the resistance of the pencil carbon change as it warms and cools. Once you've established the ideal nominal resistance you might want to replace the resistor/pencil for temperature stable resistors. Could also just be carbon flaking off?

As far as overloading the parts is concerned, that little buck regulator is only rated to ~30A, but is capable of supplying much more (core is rated to 40A IIRC). It has built-in thermal protection so the chances of frying it are low - instead of smoke you would instead have thermal shutdowns until the temp returns to 'safe' level.

These shutdowns can be mitigated or even avoided if heat sinks are applied to the back side of the regulator, (which is on the back side of the PCBs obviously). We did this pretty extensively when modding bitfury boards, which use the same part: https://bitcointalksearch.org/topic/m.3152210
legendary
Activity: 1484
Merit: 1005
gmaxwell was the one who mentioned it to me, if you can track him down you might be able to figure out which components you might need to replace.
legendary
Activity: 2450
Merit: 1002
I wouldn't super be into pencil modding these, a bunch of the components involved in voltage regulation are already overloaded about four fold at stock voltage.

Four fold? I highly doubt that, otherwise shit would be on fire. They are over spec a bit ... but seem to tolerate it fine and components absolute maximum power output, like the main IC for the VRM ..., is nowhere near its maximum.

But as far as capacitors and shit, I have no idea =P
legendary
Activity: 1484
Merit: 1005
I wouldn't super be into pencil modding these, a bunch of the components involved in voltage regulation are already overloaded about four fold at stock voltage.
legendary
Activity: 2450
Merit: 1002
BTW, just an update after a few days, the antminer is still goin strong, avging 213GH @ 418mhz @ .31 %HW ... the avg voltage has climbed to 1.17v on all chips - think theres a settling period or something..
legendary
Activity: 2450
Merit: 1002
no pics?



Not for now, 2 reasons:
1, Im lazy
2, I figure if people cant look at the PCB schematics or the PCB directly and find the components themselves then....its probably too dangerous for them to even attempt this in the first place.
legendary
Activity: 2450
Merit: 1002
Note: Not all pencils are created equal. A generic #2 pencil from Office Depot wouldn't get the voltage over 1.13 volts. A good ole' fat construction pencil from Home Depot worked much better for me.

Remember to always test the first cap (C17, I think), because it will show the highest voltage, in case you over-pencil it.

As far as my experience: I have 6 S1 Antminers, and none of them like the 418 MHz setting. HWE rates were 4.5% - 5.3% with hash rates around 210 GHs even with the pencil mod. At 393 MHz (5f05, 36) I was getting .85% - .90% HWE before the pencil mod, but the mod actually lowered it to .45% - .51% across the board. Hashrate has remained at 200 GHs. I don't have a killawatt, so I have no idea on consumption.

edit:to the OP, what firmware do you run?

I run the newest firmware. Also, I pencil mod w/ a cheap #2 pencil thats like 15+ years old ... took to the voltage up to 1.3v on a module at power on ... while I was experimenting =P
The resistance on R3 dropped to somehing like 3.5k...

What build date are ur antminers? Ive only tried it on the 2013 december antminer.... wonder if it doesnt work as well on 2014 antminers. What voltage were you setting urs to for the 418mhz setting?

Also, cap C4 is the highest voltage cap before it gets to the ASICs ... this is is more representative of the voltage being pushed by the VRM IC, to the ASICs
For easiness sake, I do measure C17 rather than C4 tho =P
member
Activity: 91
Merit: 10
Note: Not all pencils are created equal. A generic #2 pencil from Office Depot wouldn't get the voltage over 1.13 volts. A good ole' fat construction pencil from Home Depot worked much better for me.

Remember to always test the first cap (C17, I think), because it will show the highest voltage, in case you over-pencil it.

As far as my experience: I have 6 S1 Antminers, and none of them like the 418 MHz setting. HWE rates were 4.5% - 5.3% with hash rates around 210 GHs even with the pencil mod. At 393 MHz (5f05, 36) I was getting .85% - .90% HWE before the pencil mod, but the mod actually lowered it to .45% - .51% across the board. Hashrate has remained at 200 GHs. I don't have a killawatt, so I have no idea on consumption.

edit:to the OP, what firmware do you run?
member
Activity: 91
Merit: 10
In a couple of months (like diff x 4) it will be interesting to try to undervolt it and test max freq w/ decent HW error rate.

Yeah, I think to do undervolt, the resistor R3 (right next to R12) can be pencil modded ... this should theoretically drop the voltage.
R12 would not be pencil modded in this case.

careful with R3! The voltage on the first (left most) yellow cap jumps from 1.1v to 2.6v when pencil lead gets on R3. I haven't tried to boot a miner with the first cap being so high. It could lead to disaster.....
ZiG
sr. member
Activity: 406
Merit: 250
Yep..."Time IS Money"...in that business...literally...
legendary
Activity: 2450
Merit: 1002
Super, GenTarkin...!

What is your cooling...Did you upgrade it...?


Stock cooling, for now Im in a semi cool environment, so no need to uprade it. When summer comes, may have to undo the mod =(
I wish I woulda thought bout figuring this out ... months ago .. =)
legendary
Activity: 1386
Merit: 1004
In a couple of months (like diff x 4) it will be interesting to try to undervolt it and test max freq w/ decent HW error rate.

Yeah, I think to do undervolt, the resistor R3 (right next to R12) can be pencil modded ... this should theoretically drop the voltage.
R12 would not be pencil modded in this case.

This sounds a little more interesting.  If the power consumption can drop 25% but the hash rate maybe 10% that would give them a longer life in the future. 
ZiG
sr. member
Activity: 406
Merit: 250
Super, GenTarkin...!

What is your cooling...Did you upgrade it...?
legendary
Activity: 2450
Merit: 1002
A quick update, 16hrs later, shes still running strong.
418mhz, .8% HW, 212.3GH MHSav

Using an 80% effecient PSU ... power draw from wall - 510w

What are your OC settings, hex and timeout for 418?

418mhz = 6105
timeout I have set to 33
legendary
Activity: 2450
Merit: 1002
In a couple of months (like diff x 4) it will be interesting to try to undervolt it and test max freq w/ decent HW error rate.

Yeah, I think to do undervolt, the resistor R3 (right next to R12) can be pencil modded ... this should theoretically drop the voltage.
R12 would not be pencil modded in this case.
sr. member
Activity: 504
Merit: 250
A quick update, 16hrs later, shes still running strong.
418mhz, .8% HW, 212.3GH MHSav

Using an 80% effecient PSU ... power draw from wall - 510w

This is impressive. Some of us ignore the powerdrawn and only want GH/s
I already have 1 selected good performer unit that does 205GH to pool with low HW at 400mhz. All my other units have been under 200Gh at pool.

I will start same test with my special unit  Roll Eyes
member
Activity: 91
Merit: 10
March 16, 2014, 02:49:56 PM
#9
A quick update, 16hrs later, shes still running strong.
418mhz, .8% HW, 212.3GH MHSav

Using an 80% effecient PSU ... power draw from wall - 510w

What are your OC settings, hex and timeout for 418?
legendary
Activity: 1946
Merit: 1035
March 16, 2014, 02:49:36 PM
#8
In a couple of months (like diff x 4) it will be interesting to try to undervolt it and test max freq w/ decent HW error rate.
legendary
Activity: 2450
Merit: 1002
March 16, 2014, 02:47:07 PM
#7
A quick update, 16hrs later, shes still running strong.
418mhz, .8% HW, 212.3GH MHSav

Using an 80% effecient PSU ... power draw from wall - 510w
ZiG
sr. member
Activity: 406
Merit: 250
March 16, 2014, 02:32:46 PM
#6
Good job, buddy...Thank you for posting...

How are your specs/results now...Everything OK...?



legendary
Activity: 2128
Merit: 1005
ASIC Wannabe
March 16, 2014, 10:21:32 AM
#5
It sounds like a lot of additional power for another 10GH. It's my opinion that we passed the point of diminishing returns at 190.

I think that pushing it as far as safe is good right now, but in the next few months it may become practical to bring down the power draw a bit.

most of mine run at 387.5, with a few at 375 or 400
legendary
Activity: 1666
Merit: 1185
dogiecoin.com
March 16, 2014, 07:06:43 AM
#4
It sounds like a lot of additional power for another 10GH. It's my opinion that we passed the point of diminishing returns at 190.
hero member
Activity: 692
Merit: 500
March 16, 2014, 03:53:17 AM
#3
Photos ?
legendary
Activity: 2450
Merit: 1002
March 16, 2014, 01:13:01 AM
#2
UPDATE: ... decided to start upping the clockspeed....
3hrs @ 418mhz, HW 1%, MHSav 212GH...


The unit is still alive after total of 6hrs hashing w/ increased voltage =)
legendary
Activity: 2450
Merit: 1002
March 16, 2014, 01:06:16 AM
#1
For those interested:
Ive succeeded in pencil voltmodding my S1 ... after a days worth of digging through the PCB spec sheet from antminer and not knowing hardly anything about electronics... I dared to try.... =)
So far, its not blown up and Ive modded each of the 8 "modules" on the S1 successfully.
This has been tested only on the December version of the units, after studying differences between Dec/Jan units...not sure if this works the same on Jan ones.... but should.

RESULTS:
Stock:
1.11v on chips measured on top portion of the modules
400mhz @ 1.8% HW

Modded:
1.15-1.16v measured same place, consistency is hard to get w/ a pencil so.... there is some variance between modules
400mhz @ .002% HW

Temps rose about 5c and fan spins a few hundred RPM faster to compensate.
Power also increased by about 25-30w measured at wall

Further OC results are not in yet, thats next on the list.
For now I am keeping a close eye on voltages / temperatures VRMs & ASICs - all seem within good spec
Ultimately hoping it wont die over the next few hours...

ITEMS NEEDED:
voltmeter(that can measure resistance) & pencil

METHOD:
Before proceeding get a ballpark measurement of the chips on the "modules" by doing this:
Make sure voltmeter is set to VDC measurement!

Put negative voltmeter probe on the "GND" contact on the first module
Put positive probe on the bottom side of the orange square component thats in the middle of the "module"
Do this for all modules and note.
Stock they should be right around 1.10-1.11v

NEXT..THE MOD:

Power off the antminer completely!
When looking at a "blade" ... the left most "module" will have a resistor labeled R12.
Measure its resistance from end to end w/ voltmeter, it should read around 4.4-4.45k

Now, take pencil and scribble on top of the resistor from end to end... probably around 10-15 light strokes

Measure the resistor once again ... our target is 4.25-4.3k

If its lower than that, take ur finger and using finger tip skin, swipe it downwards across the resitor.. lightly

Measure resistor again .... if took too much off (back to 4.4 or in the 4.35k area)

Then apply more pencil....

Repeat this process till u meet target of 4.25-4.3k

Once this resistor reads correctly move over to same resistor on the next "module"

Once an entire blade all reads similar target resistance .. Turn the antminer on.
During its bootup process measure resultant voltage by doing this:

Make sure voltmeter is set to VDC measurement!

Put negative voltmeter probe on the "GND" contact on the first module
Put positive probe on the bottom side of the orange square component thats in the middle of the "module"

Do this for all "modules" on the modified "blade"
The voltages so all be close to 1.15-1.17v
Anything higher than 1.2v .. IMMEDIATELY power off the antminer.

If all is well, then you can proceed to the other "blade" on the antminer.



POST METHOD:
Monitor the temps of the antminer, I use finger test method of VRM components ... note the temps of individual components on top of each module .. if after the mod they are scorching hot to the touch and immeditely can feel them "burn" ur finger... well, thats too hot for my comfort ... so may want to undo the mod.
Also, note the difference in feel of temp of the chips before / after mod ...
The hottest VRM / ASIC is gonna be the furthest from the fan ...
Optionally can run the voltage a bit lower on these modules furthest from the fan too... =)


Results should be similar, significantly lowered HW error at same clock speeds before mod. Therefore allowing you to eventually clock it higher ... once you have confirmed the unit is stable and running within temp spec ... for at least a day or 2 I say.

Its all risk so, its up to you of course =)

* HAVE FUN! BE CAREFUL!

** I cant be held responsible if you blow up ur shit!

*** POST RESULTS!

TIP: the little VRM heatsink packs I see being sold on ebay for antminers would keep the VRM's very comfy for a mod like this! I would recommend this for attempting anything thats 1.2v+
The pencil mod can easily take ur antminer into the 1.35v range w/ a solid amount of graphite applied... but u would need SUPERDUPER cooling on that puppy! HIGHLY NOT RECOMMENDED!
Jump to: