L3+/L3++ voltage tuning script

[philipma1957]

what kind of results can one expect? we talking 50watt or 5 ?

Sorry found answer in original thread.

I am going to ping phil to give it a whirl since I know he has a few L3+. Sadly all miner are in a hosting facility except for one lonely s9 that is pos and been in warranty 5 times. But the OP thread it was 600w vw 800w so that is a 20% saving.

Just got a pm to read this

I have multiple L3+ with the undervolt tool

using just the jstefanop  tool  on freq 359 hash rate of 470 I do 667 watts with minimum volts on all 4 boards. using a corsair ax1500i titanium

this is a savings of over 100 watts.

So my first question is  can this scrypt use a lower volt setting then  0xfe?
..

Thanks for checking in. No, unfortunately 0xfe equals 254 and ist the highest undervolting you can set.
If you have 4x 0xfe set, there is nothing more you can do to save on energy (except lowering frequency further), this script doesn't help you then as it is about to find the highest undervolting setting where you miners still work reliably.

You could however get possibly more hashes/W if you increase the frequency to the point where you still can run them with 0xfe on all or most blades.

I will play with this during the week.

[1715444112]

1715444112

[1715444112]

1715444112

[1715444112]

1715444112

[1715444112]

1715444112

[supersonic]

Well that updated version works every time exactly 15mins to the second. Then it suddenly stops.

[CoinMeLadies]

Hi there,

Thank you so much for the tool you've created. It's very helpful.

Today I've run into this error and am wondering if you've encountered it before?

https://imgur.com/a/xIoX9BS

[alexcrys1]

I have L3+(Blissz v1.02) is possible to go lower voltage(save more W)? with freq:300-312?

thx

[psycodad]

Well that updated version works every time exactly 15mins to the second. Then it suddenly stops.

Thank you very much for the reports.
It actually finishes because it thinks the miner is stable and under the max. errror-rate (which I have currently set to 0.2/min, or 1/5min per chain).
The downvolting kicks in as soon as a chain has 0.0 errors in the 10min avg. and the voltage increase kicks in when a chain has more than 0.2 errors per min in the 5min average.

When I look at your end-results, the script assumes the miner as tuned when it does not make any changes within 10min to voltages as it thinks it has found the sweet spot between no errors and too much errors.

I will revisit that part of the code and see what I can improve there, but from what I see it does actually tune your miner, but I agree it should try harder/longer:


| Start 19:20.46 | 0xa0 | 0x8b | 0xcd | 0xa9 |
| Start 19:20.46 | 59 C | 58 C | 54 C | 53 C |
| End   19:35.46 | 0xa0 | 0x92 | 0xcd | 0xb0 |
| End   19:35.46 | 59 C | 58 C | 54 C | 53 C |

| Start 13:39.55 | 0xa3 | 0x6f | 0xa3 | 0x9c |
| Start 13:39.55 | 60 C | 61 C | 58 C | 56 C |
| End   14:01.55 | 0x9f | 0x7d | 0xb1 | 0xa2 |
| End   14:01.55 | 60 C | 60 C | 57 C | 55 C |

| Start  9:40.40 | 0x87 | 0x67 | 0x87 | 0x87 |
| Start  9:40.40 | 60 C | 59 C | 57 C | 54 C |
| End    9:55.40 | 0x8e | 0x6e | 0x87 | 0x87 |
| End    9:55.40 | 59 C | 59 C | 57 C | 54 C |


I was trying to make sure the script doesn't run forever and stops/exits as soon as it can, but I probably went a bit too far wrt this.

For the moment feel free to run the script multiple times, I even recommend to re-run it everytime ambient temperatures change noteably (>=2°C difference). Simply put, the colder the less voltage needed.
Whenever you change frequency you should re-run the script or manually re-tune them anyways.

If you want to run the script multiple times in a row unattended, use a command like below:

Code:

for i in range {1..3}; do /path/l3plus_autotune.py -i x.x.x.x; sleep 60; done

^This will run it three times in a row, just as a workaround until I come up with a smarter solution on finding the right point in time to declare the miner tuned.

I will post, once I have an update.

[psycodad]

I have L3+(Blissz v1.02) is possible to go lower voltage(save more W)? with freq:300-312?

thx

With such low frequency your miners should already be set to minimum voltage/maximum undervolting setting of 0xfe (254). You can't undervolt more than this so no need for tuning.

[philipma1957]

I have L3+(Blissz v1.02) is possible to go lower voltage(save more W)? with freq:300-312?

thx

With such low frequency your miners should already be set to minimum voltage/maximum undervolting setting of 0xfe (254). You can't undervolt more than this so no need for tuning.

My take on the scrypt   is that it is good if you want to try minimum volts setting of 0xfe  and find highest hashrate that works for separate boards.

But that if you clock at freq of 359 or lower  like I do most boards simply  do the 0xfe  with close to no errors.

Still if you have 2 board that do   384 and one that does 359  this is a good  addition to the voltage control software.

[supersonic]

And how do i run script on multiple L3's at the same time ? :>

[psycodad]

And how do i run script on multiple L3's at the same time ? :>

Simply open another terminal and start it against another miner (or use screen or tmux). Even on an old and outdated PC you should easily be able to tune a dozen or so miners at once.

Just make sure you do not have 2 script instances running against the same miner at the same time (it should not have any desastrous consequences but both scripts will probably fail with an error when they try to read out or set voltage at the same time).

[supersonic]

this time i got this erroe on miner that i already run script before:

Undefined errors occured fetching voltage settings from miner:
sh: /config/sv: Permission denied

Aborting.

[psycodad]

this time i got this erroe on miner that i already run script before:

Undefined errors occured fetching voltage settings from miner:
sh: /config/sv: Permission denied

Aborting.

I just saw this very same error on one of my miners a few hours ago, but I couldn't find the cause for this behaviour so far.

I had not much time to dig deeper yet, but what happened here was that /config/sv (the slightly varied binary of jstefanops work we install on first run of the script) was truncated from  the 9788 bytes it should have to something around 1000 bytes and has changed permissions (hence the Permission denied).

The simplest way to resolve this is by removing the /config/sv file on the antminer via ssh and letting the script re-install it again. Though my miner where I saw this had to rebooted afterwards as the PIC version suddenly was 0x00 (probably meaning the PIC MCU didn't reply anymore). After a cold boot things were all fine and dandy again and I successfully run the script on that miner.

Could you check for me two things with the miner where you saw that:

1. Was the miner rebooted recently and/or did you have to reboot it to fully recover from this?
2. Before you remove the (assumedly) truncated /config/sv, can you check what is in there with i.e. 'hexdump -C /config/sv' (this might give us a hint what process is responsible for the corrupt binary)

[Ameador1]

this time i got this erroe on miner that i already run script before:

Undefined errors occured fetching voltage settings from miner:
sh: /config/sv: Permission denied

Aborting.

I just saw this very same error on one of my miners a few hours ago, but I couldn't find the cause for this behaviour so far.

I had not much time to dig deeper yet, but what happened here was that /config/sv (the slightly varied binary of jstefanops work we install on first run of the script) was truncated from  the 9788 bytes it should have to something around 1000 bytes and has changed permissions (hence the Permission denied).

The simplest way to resolve this is by removing the /config/sv file on the antminer via ssh and letting the script re-install it again. Though my miner where I saw this had to rebooted afterwards as the PIC version suddenly was 0x00 (probably meaning the PIC MCU didn't reply anymore). After a cold boot things were all fine and dandy again and I successfully run the script on that miner.

Could you check for me two things with the miner where you saw that:

1. Was the miner rebooted recently and/or did you have to reboot it to fully recover from this?
2. Before you remove the (assumedly) truncated /config/sv, can you check what is in there with i.e. 'hexdump -C /config/sv' (this might give us a hint what process is responsible for the corrupt binary)

Is your script checking to see if sv is already on the miner before installing? Maybe it is trying to re-install it again at some point and failing, leaving it in corrupted state. If check for existence prior to install maybe it will stop this. I did not read your source - just a quick thought.

[psycodad]

..

Is your script checking to see if sv is already on the miner before installing? Maybe it is trying to re-install it again at some point and failing, leaving it in corrupted state. If check for existence prior to install maybe it will stop this. I did not read your source - just a quick thought.

Yes, the install function is only called if the error string from the miner trying to call sv on ends with " not found":

https://github.com/psycodad-bct/bitmain-tools/blob/master/scripts/l3plus_autotune.py#L158

On top of that a flag is set, that makes sure that it only tries once to install the binary (most failures installing it exit the script anyway) per script run:

https://github.com/psycodad-bct/bitmain-tools/blob/master/scripts/l3plus_autotune.py#L199

[PontusLSE]

Fantastic tool!

But I do feel that something about it is a bit off, everytime I run the script i feel  like it keeps setting the voltage lower and lower.

I ran the script 7 times in a row without doing any changes what so ever in between and this is the results (I converted the hex into decimal too to make it easier so see for us that are not used to hex values )

Is this normal or am I doing something wrong?

Chain 1   
b0   176
be   190
c5   197
cc   204
da   218
e1   225
e6   230
Chain 2   
95   149
ad   173
b4   180
bb   187
bb   187
c2   194
c8   200
Chain 3   
84   132
7f   127
7f   127
7f   127
8d   141
8d   141
97   151
Chain 4   
9c   156
b5   181
bc   188
c3   195
c1   193
c1   193
c8   200

[psycodad]

Fantastic tool!

But I do feel that something about it is a bit off, everytime I run the script i feel  like it keeps setting the voltage lower and lower.

I ran the script 7 times in a row without doing any changes what so ever in between and this is the results (I converted the hex into decimal too to make it easier so see for us that are not used to hex values )

Is this normal or am I doing something wrong?

...

Thanks.

The current version on github has still a few minor but annoying bugs, one of them being that it declares a miner that can be downvolted more as "tuned" and exits. I have a version that has these fixed as well as the tuning process improved that I am currently testing on my miners. I will push that version to github in the next two days and will post here then.

Regarding the script running multiple times: That's no problem, the script turns the voltage down as long as there is less than 1 error/min on the respective chain, once there is more errors it starts to increase the voltage again until a point is found where it sill produces errors but less than 1/min and chain.
You can run it as often as you like, with every run it should tune your miner closer to it's ideal voltage at current conditions (frequency and ambient).

[PontusLSE]

Fantastic tool!

But I do feel that something about it is a bit off, everytime I run the script i feel  like it keeps setting the voltage lower and lower.

I ran the script 7 times in a row without doing any changes what so ever in between and this is the results (I converted the hex into decimal too to make it easier so see for us that are not used to hex values )

Is this normal or am I doing something wrong?

...

Thanks.

The current version on github has still a few minor but annoying bugs, one of them being that it declares a miner that can be downvolted more as "tuned" and exits. I have a version that has these fixed as well as the tuning process improved that I am currently testing on my miners. I will push that version to github in the next two days and will post here then.

Regarding the script running multiple times: That's no problem, the script turns the voltage down as long as there is less than 1 error/min on the respective chain, once there is more errors it starts to increase the voltage again until a point is found where it sill produces errors but less than 1/min and chain.
You can run it as often as you like, with every run it should tune your miner closer to it's ideal voltage at current conditions (frequency and ambient).

Ok, thanks. Looking forward to the new version.  A nice feature would be to keep the script running in an infinite loop to always adjust the voltage. The ambient temperature differs quite a bit for my miners and therefore the optimal voltage is different from time to time.

[bigdude]

Fantastic tool!

But I do feel that something about it is a bit off, everytime I run the script i feel  like it keeps setting the voltage lower and lower.

I ran the script 7 times in a row without doing any changes what so ever in between and this is the results (I converted the hex into decimal too to make it easier so see for us that are not used to hex values )

Is this normal or am I doing something wrong?

...

Thanks.

The current version on github has still a few minor but annoying bugs, one of them being that it declares a miner that can be downvolted more as "tuned" and exits. I have a version that has these fixed as well as the tuning process improved that I am currently testing on my miners. I will push that version to github in the next two days and will post here then.

Regarding the script running multiple times: That's no problem, the script turns the voltage down as long as there is less than 1 error/min on the respective chain, once there is more errors it starts to increase the voltage again until a point is found where it sill produces errors but less than 1/min and chain.
You can run it as often as you like, with every run it should tune your miner closer to it's ideal voltage at current conditions (frequency and ambient).

Ok, thanks. Looking forward to the new version.  A nice feature would be to keep the script running in an infinite loop to always adjust the voltage. The ambient temperature differs quite a bit for my miners and therefore the optimal voltage is different from time to time.

So finally is this work for now?
it seem very interesting as l3+ now super low profit at all for me.
but i still mining with it and will try this tweak tomorrow

[psycodad]

I have pushed a new version to my github and encourage all that are using the script to update. There are a few small bugs fixed and the tuning process should be more accurate and even faster in finding the right values for each of your chains.

A nice feature would be to keep the script running in an infinite loop to always adjust the voltage. The ambient temperature differs quite a bit for my miners and therefore the optimal voltage is different from time to time.

I see your point and considered it but the changes in the script would be quite big with the chance of introducing many new and unneeded bugs. You can already easily achieve the same by a simple shell loop like:

Code:

while true; do ./l3plus_autotune.py -i 10.10.10.33; sleep 900; done

The above line starts the tuning script and waits until it exits, sleeps then 900s (15min) and restarts the script again in a never ending loop (or until you hit CTRL-C).

Happy tuning.

BTW: If people with wattmeters on their miners would post their measurements (before and after tuning and with the respective frequency and voltage-setting) I could try to interpolate watt savings and print a very rough estimate on the savings in Watt for each run of the script.

[patchybeard]

I love what this can do - Any interest in creating one for the A3 miner?

[psycodad]

I love what this can do - Any interest in creating one for the A3 miner?

I have only L3+ so somebody else would have to do this.
But it should be very easy to port over to A3 or other Antminers that offer the same possibilities (voltage control).