Antminer S5 - Underclock - Undervolt - Best J/GH - page 2.

VirosaGITS

legendary

Activity: 1302

Merit: 1068

Quote from: adaseb on December 25, 2015, 01:23:28 PM

Quote from: RichBC on December 24, 2015, 04:19:08 AM

Quote from: adaseb on December 24, 2015, 03:53:44 AM

So how are you are running these undervolted? You have some special PSU that runs at 10Volts? Aren't those noisy and expensive?

I have two systems. First I have a pair of Meanwell 5V PSU's in series, these allow me to adjust between 9V & 12V

Second I have a Server PSU with a Buck Converter on the output, and can set any voltage with that. Neither solution is expensive or more noisy than other PSU. Various pictures & information in this thread.

Rich

Anyone to use a regular computer PSU ?

You can do it but you'll have 12v output so you will need to use buck converter and i don't know if you can make these adjustable. Meanwhile you can get server PSU with an adjustable volt pot so you can get exactly the volt output you need without going through a gadget.

adaseb

legendary

Activity: 3808

Merit: 1723

Quote from: RichBC on December 24, 2015, 04:19:08 AM

Quote from: adaseb on December 24, 2015, 03:53:44 AM

So how are you are running these undervolted? You have some special PSU that runs at 10Volts? Aren't those noisy and expensive?

I have two systems. First I have a pair of Meanwell 5V PSU's in series, these allow me to adjust between 9V & 12V

Second I have a Server PSU with a Buck Converter on the output, and can set any voltage with that. Neither solution is expensive or more noisy than other PSU. Various pictures & information in this thread.

Rich

Anyone to use a regular computer PSU ?

RichBC

hero member

Activity: 588

Merit: 500

Quote from: adaseb on December 24, 2015, 03:53:44 AM

So how are you are running these undervolted? You have some special PSU that runs at 10Volts? Aren't those noisy and expensive?

I have two systems. First I have a pair of Meanwell 5V PSU's in series, these allow me to adjust between 9V & 12V

Second I have a Server PSU with a Buck Converter on the output, and can set any voltage with that. Neither solution is expensive or more noisy than other PSU. Various pictures & information in this thread.

Rich

adaseb

legendary

Activity: 3808

Merit: 1723

So how are you are running these undervolted? You have some special PSU that runs at 10Volts? Aren't those noisy and expensive?

1slickvdc

member

Activity: 67

Merit: 10

Yeah. Just didn't seem like a normal dropout. Lights were both on. Hashing very slowly, at roughly 50% power. Strange.

RichBC

hero member

Activity: 588

Merit: 500

Quote from: 1slickvdc on November 22, 2015, 02:03:21 AM

Quote from: RichBC on November 22, 2015, 02:01:26 AM

The other thing I have found is that each hash board in an S5 gives a different result so as you reduce the volts it is initially just one of the boards that is responsible for the increased HW error rate. It's worth doing some measurements on individual boards to better understand your system.

I've noticed hash rates roughly half of normal at certain combinations, so I must have one board that misbehaves before the other. Too bad there's no tattle-tale on which board.

If one board drops out completely usually but not always the Red led on the board will go out.

Rich

1slickvdc

member

Activity: 67

Merit: 10

Quote from: RichBC on November 22, 2015, 02:01:26 AM

The other thing I have found is that each hash board in an S5 gives a different result so as you reduce the volts it is initially just one of the boards that is responsible for the increased HW error rate. It's worth doing some measurements on individual boards to better understand your system.

I've noticed hash rates roughly half of normal at certain combinations, so I must have one board that misbehaves before the other. Too bad there's no tattle-tale on which board.

RichBC

hero member

Activity: 588

Merit: 500

Quote from: 1slickvdc on November 21, 2015, 02:01:43 PM

A note on my methodology here: I'd mine for no less than 10 minutes at any given freq/volt combination, then reduce voltage (restarting cgminer at each step-down) until GH/s rate reduced significantly, then bring voltage to last working level, and then drop frequency. Lather, rinse, repeat.
Temps are with fixed "30%" fan speed, ~20C ambient.

A nice set of numbers and very consistant with what I have measured. I have found that J/GH continues to improve as you drop voltage / frequency with the absolute best J/Gh being between 150MHz & 175MHz.

The other thing I have found is that each hash board in an S5 gives a different result so as you reduce the volts it is initially just one of the boards that is responsible for the increased HW error rate. It's worth doing some measurements on individual boards to better understand your system.

Rich

1slickvdc

member

Activity: 67

Merit: 10

A note on my methodology here: I'd mine for no less than 10 minutes at any given freq/volt combination, then reduce voltage (restarting cgminer at each step-down) until GH/s rate reduced significantly, then bring voltage to last working level, and then drop frequency. Lather, rinse, repeat.
Temps are with fixed "30%" fan speed, ~20C ambient.

1slickvdc

member

Activity: 67

Merit: 10

Quote from: VirosaGITS on November 21, 2015, 01:45:44 AM

Its very interesting to "poke around" i have 1.1, 1.3, 1.5 and 1.9 and 1.91 mixed in 4 S5 and oddly enough, i have no idea why, but they all cap at 393.75hz on high quality PSU, 375hz on lesser psu (Like gold raidmax)...

Except the 1.9, that one do 400hz with no HW, getting 1.32TH/s while all the other cap at 1.29~TH/s. I don't get it. And i suspect i could push it a bit further but i'll wait a little for the motivation to do further voltage testings. I dont get it because its not the PSU having higher voltage itself. When i swap PSU used by 2 S5, they still get the same hashrate/HWE%.

At least i know that EVGA G2 and GS have consistent top notch clean power, same with the server PSU's i got from Quakefiend.

I'd just like to understand why the 1.9, i assume, have higher core voltage than all the others. It is my only guess as to why it is stable at higher OC.

My 1.3, using that DPS-800 PSU (which runs at 12.65V loaded) can run 400-425MHz stable and less than .1% HWE most of the time. Runs around 1.3-1.4TH at that speed. Gets seriously hot and needs to run the fan balls out even with a window nearby (outside ambient temp <10C). Just put a NEX750G on it and brought it back down to 350MHz, nice and cool at 50C, although running a .025% HWE, which seems high at stock speeds.

I'm getting some numbers on my 1.91, although I don't know if any of y'all tried undervolting at stock speeds, and speeds near that. I'm starting at 350MHz and working my way down with voltage and core frequencies.

VirosaGITS

legendary

Activity: 1302

Merit: 1068

Quote from: 1slickvdc on November 20, 2015, 11:02:25 PM

My "brand new" S5 v1.91 is here, including a nice 80+ Gold PSU for my 1.3 to bring those consumption numbers down. I'll poke around with this and get some numbers on my 1.91.

Its very interesting to "poke around" i have 1.1, 1.3, 1.5 and 1.9 and 1.91 mixed in 4 S5 and oddly enough, i have no idea why, but they all cap at 393.75hz on high quality PSU, 375hz on lesser psu (Like gold raidmax)...

Except the 1.9, that one do 400hz with no HW, getting 1.32TH/s while all the other cap at 1.29~TH/s. I don't get it. And i suspect i could push it a bit further but i'll wait a little for the motivation to do further voltage testings. I dont get it because its not the PSU having higher voltage itself. When i swap PSU used by 2 S5, they still get the same hashrate/HWE%.

At least i know that EVGA G2 and GS have consistent top notch clean power, same with the server PSU's i got from Quakefiend.

I'd just like to understand why the 1.9, i assume, have higher core voltage than all the others. It is my only guess as to why it is stable at higher OC.

1slickvdc

member

Activity: 67

Merit: 10

My "brand new" S5 v1.91 is here, including a nice 80+ Gold PSU for my 1.3 to bring those consumption numbers down. I'll poke around with this and get some numbers on my 1.91.

1slickvdc

member

Activity: 67

Merit: 10

Might be the images are too large or too high of a resolution. Damn phone shoots like 16MP with a 5MB file.

VirosaGITS

legendary

Activity: 1302

Merit: 1068

Quote from: 1slickvdc on November 17, 2015, 04:09:25 PM

I've been running my V1.3 at 400MHz with .013% HWE with that monster PSU.

I tried those IMG tags and that image proxy/invalid image thing is what comes up

Ah i see what you mean.

I use this format normally it work;

Code:

[img]https://i.imgur.com/l45Vt4I.jpg[/img]

Something might be wrong with the picture itself. Since this work;

Code:

[img]https://i.imgur.com/3UjmfRn.jpg[/img]

1slickvdc

member

Activity: 67

Merit: 10

I've been running my V1.3 at 400MHz with .013% HWE with that monster PSU.

I tried those IMG tags and that image proxy/invalid image thing is what comes up

VirosaGITS

legendary

Activity: 1302

Merit: 1068

I got my hand on a 1.9, was surprised since it take over 393.75 freq effectively, running 400hz at 1.32+ TH/s with virtually no HWE, all the others cap at 393.75hz at 1.29-1.3~.

Quote from: 1slickvdc on November 17, 2015, 12:20:30 PM

SMD's aren't that bad. We'll see if this board is a 1.9 or 1.91 and see what happens from there. Maybe we'll have another project on this thread.

Also forgot to post this up... High res shot of the 1.3 hash board

http://wavemadison.net/mine/s5_1.3_hash.jpg

dammit, how does everyone embed these images?

Code:

[img][/img]

1slickvdc

member

Activity: 67

Merit: 10

SMD's aren't that bad. We'll see if this board is a 1.9 or 1.91 and see what happens from there. Maybe we'll have another project on this thread.

Also forgot to post this up... High res shot of the 1.3 hash board

http://wavemadison.net/mine/s5_1.3_hash.jpg

dammit, how does everyone embed these images?

RichBC

hero member

Activity: 588

Merit: 500

Yes could be modified. There are 3 crystals with a few discrete components to add and then lift the level shifting diode on those stages. Then remove the first 14 LDO Regulators and fit a zero ohms link to connect the IO voltage directly to the stage above.

So if you can solder small surface mount components then relatively simple.

Rich

1slickvdc

member

Activity: 67

Merit: 10

My new (hopefully) v1.91 is on its way, or should be. Something dawned on me. The difference between the v1.9 and 1.91 was very small IIRC, could a 1.9 be modified by someone with the right equipment and skills to work like a 1.91?

RichBC

hero member

Activity: 588

Merit: 500

Quote from: adaseb on November 17, 2015, 06:50:44 AM

Didn't have time to read this entire thread. So basically there is no way to mod the S5 by reducing the resistance like on the S1/S3 ?

Only way is to supply 10V from a power supply?

Can't you put a resistor between the PCIe connection to drop the voltage down to 10V from the PSU's 12V or is there some kind of protection built into PSUs ?

Simple answer, no you cannot do simple resistance change as is S1/S3.

Reason is that there are no DC-DC converters in the S5, the ASIC's are connected in a string so the only way to reduce the voltage is by reducing the 12V supply. Resistor on the input also not practical as the volts it drops will vary with current, and will be dissipating significant power which will negate a lot of the saving. You would need to use one of the options for voltage reduction covered in this thread.

Rich

Topic: Antminer S5 - Underclock - Undervolt - Best J/GH - page 2. (Read 31123 times)