S3+ (BM1382) Overclocking with voltage setting

[HerbPean]

Second B1 Unit (one that wasn't working good) has been going down after one hour. Will try 250 freq now.

EDIT: Both now are going down.  Will lower the freq further to 237.5
EDIT: Trying 225 with 0.72V. 237.5 wasn't working

[canford]

I have concluded that with the stock heatsinks and paste, that increasing the voltage on my six S3+ units does not work for me.  I tried everything, and got some clear short term performance improvements.  However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.  Finally reflashed to the 10/24 firmware and went with frequencies of: 225, 225, 237, 237, 237, 243.  Increasing frequencies above those six points results in dropping hashrates, with or without corresponding voltage changes.  I assume the successful overclocking by others is either different chips or better cooling.  It was frustrating to see clear jumps above 500 GH/s, but then rates that dropped below stock of 453 GH/s.

[HerbPean]

I have concluded that with the stock heatsinks and paste, that increasing the voltage on my six S3+ units does not work for me.  I tried everything, and got some clear short term performance improvements.  However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.  Finally reflashed to the 10/24 firmware and went with frequencies of: 225, 225, 237, 237, 237, 243.  Increasing frequencies above those six points results in dropping hashrates, with or without corresponding voltage changes.  I assume the successful overclocking by others is either different chips or better cooling.  It was frustrating to see clear jumps above 500 GH/s, but then rates that dropped below stock of 453 GH/s.

I came to the same conclusion as well. I got the one unit that was running steady at 504 GH/s before and now running 529 gh/s with a voltage of 0.75v. Which is impressive since it's taking less power and i have been able to push it a bit more.

For the unit that I had trouble with. Playing with the voltage doesn't help much in the long term. One of the two is running 231.25 with a voltage of 0.725 and it's getting 450 gh/s. It's not bad since the voltage is lower. it's more efficient but not performing as like it should at this frequency.

The other one (From Batch 1) is a pain in the ***. Whatever I tried is just go down and down on the long term. getting 444 gh/s at 231.25 with 0.725v. Still saving power but under performing as well.

B?     0d 5h 11m 32s   451.51   450.39   0   0.0159%   231.25
B1   0d 10h 33m 42s   530.09   529.58   0   0.0004%   45 44   262.5
B1   0d 6h 39m 32s   458.03   444.47   0   0.0002%   231.25

The bottom line is Quite happy that i'm saving power. If you have a unit that you can OC to 250 with the default setting, you can make a huge power saving and even pushing it harder and getting awesome result.

If anyone has found a trick with the units that is somehow under performing with the stock setting, share it with us so we can make them more profitable.

[pekatete]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I assume the successful overclocking by others is either different chips or better cooling.  It was frustrating to see clear jumps above 500 GH/s, but then rates that dropped below stock of 453 GH/s.

They are all BM1382's, though the earlier S3's (apparently) have different voltage regulators (and who knows what else!). Cooling wise, you may have a point, though debatable, with the earlier S3's, but not relating to S3+'s. Again, I have to say you did not run them for long enough .... at the very least I'd run it for 4hrs. And the clear jumps you refer to, is that with the 5s or the avg hash-rate? For the 5s I've seen as high as 700Gh/s and as low as 12Gh/s, so it does not really matter, rather the avg rate.

EDIT: Finally, it is not a case of increasing / decreasing the voltage at which the chips run, it is about setting the appropriate voltage for the frequency you are running at. At the very least, you'll notice a decrease in the HW errors as the S3's (and S3+'s) seem to ship with an non-datasheet freq - voltage setting. For other units, you'll also see an improvement in hash-rate but more pertinently, if you are overclocking, then you'll be able to do it at its appropriate voltage and will be in a better position to decide when you've stretched your unit far enough. In my case, I have one unit (an S1 to S3 upgrade) that I am limited by the PSU and datasheet voltage settings as I feel it can be pushed further!

[pekatete]

The bottom line is Quite happy that i'm saving power. If you have a unit that you can OC to 250 with the default setting, you can make a huge power saving and even pushing it harder and getting awesome result.

If anyone has found a trick with the units that is somehow under performing with the stock setting, share it with us so we can make them more profitable.

It is false to say that you are making a power saving, rather, that you can achieve the same rate with a PSU that can efficiently supply the  wattage at the datasheet voltage (plus a bit more for the fans and lost power). Thus if you are running with a voltage setting of 0750 and freq of 250, you can run the rig using a PSU with a rated wattage of ~400 watts and achieve at least 504 Gh/s, or a voltage setting of 0720 with a freq of 225, a PSU rated at ~350watts to achieve 453Gh/s. Of course, for running them that tight, they'll also have to be at least bronze rated too. If, however, you are running a PSU rated at 650watts, you'll not see the drop at the wall as you expect.

[kendog77]

I have been running B11 units at 262.5 / 0750 voltage for the past three days, and they have hashing at ~530 GH/s with very low (.0001-.0002%) hardware errors. B11 units appear to be very stable at this speed and voltage, and use ~415 Watts at the wall (.78W/GH).

Has anyone tried pushing the S3+ a bit further? I'm tempted to up the speed to 268 or 275 and see what happens... 

[pekatete]

I have been running B11 units at 262.5 / 0750 voltage for the past three days, and they have hashing at ~530 GH/s with very low (.0001-.0002%) hardware errors. B11 units appear to be very stable at this speed and voltage, and use ~415 Watts at the wall (.78W/GH).

Has anyone tried pushing the S3+ a bit further? I'm tempted to up the speed to 268 or 275 and see what happens... 

Up it, no worries so long as you have it adequately powered. You may need to play a bit with the voltage setting for freq 268, i.e 0750, 0755 or 0800 since it is in-between the datasheet settings guided by the HW errors you may get.

If you look at the OP, I had a rig at freq 281 though the HW is a lot higher than I got running it at freq 268.75 so I reverted to that. Saying that, like most of my S3's (and +'s) , I run it off a 550W PSU ..... so thre may be something in that.

[HerbPean]

The bottom line is Quite happy that i'm saving power. If you have a unit that you can OC to 250 with the default setting, you can make a huge power saving and even pushing it harder and getting awesome result.

If anyone has found a trick with the units that is somehow under performing with the stock setting, share it with us so we can make them more profitable.

It is false to say that you are making a power saving, rather, that you can achieve the same rate with a PSU that can efficiently supply the  wattage at the datasheet voltage (plus a bit more for the fans and lost power). Thus if you are running with a voltage setting of 0750 and freq of 250, you can run the rig using a PSU with a rated wattage of ~400 watts and achieve at least 504 Gh/s, or a voltage setting of 0720 with a freq of 225, a PSU rated at ~350watts to achieve 453Gh/s. Of course, for running them that tight, they'll also have to be at least bronze rated too. If, however, you are running a PSU rated at 650watts, you'll not see the drop at the wall as you expect.

I see, i run two on a Corsair 1000 Gold and one Corsair 750 Bronze. I was sure the lower voltage would drain less power. My bad i guess.

[HerbPean]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I let them run for over 10 hours, One of mine went down to 400 GH/s and the other to 420GH. So on the long run, mines are not performing well.

My good machine was still doing fine but loosing speed to0 ... about 515gh in almost 24 hours.

[pekatete]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I let them run for over 10 hours, One of mine went down to 400 GH/s and the other to 420GH. So on the long run, mines are not performing well.

My good machine was still doing fine but loosing speed to0 ... about 515gh in almost 24 hours.

10 hrs is a good enough indicator I'd think. On the positive side, if these are the same units that could not go over 450 in the first place, but can now hash for a while over 500, then you probably have to revert to restarting them often like you were previously, at least now they'll be running faster. On that note, you may find that simply restarting cgminer will "clear things up".

[HerbPean]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I let them run for over 10 hours, One of mine went down to 400 GH/s and the other to 420GH. So on the long run, mines are not performing well.

My good machine was still doing fine but loosing speed to0 ... about 515gh in almost 24 hours.

10 hrs is a good enough indicator I'd think. On the positive side, if these are the same units that could not go over 450 in the first place, but can now hash for a while over 500, then you probably have to revert to restarting them often like you were previously, at least now they'll be running faster. On that note, you may find that simply restarting cgminer will "clear things up".

Yeah, I wish i had some kind of auto-reboot triggered on a timer.

How do you retstart cgminer quickly without rebooting and changing the freq ?

[pekatete]

Yeah, I wish i had some kind of auto-reboot triggered on a timer.

How do you retstart cgminer quickly without rebooting and changing the freq ?

You can do it via a script that you run via scheduled tasks. I had one before I stumbled accross the voltage settings but removed it when my affected rigs stabilised after the changes. You can put a shout out in the S3 support thread or the cgminer thread and someone will respond with a solution.

[kendog77]

262.5/0750 appears to be the sweet spot (530 GH/s at 415W).

Errors go up and efficiency goes way down above that setting.

[canford]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I let them run for over 10 hours, One of mine went down to 400 GH/s and the other to 420GH. So on the long run, mines are not performing well.

My good machine was still doing fine but loosing speed to0 ... about 515gh in almost 24 hours.

The degradation in hashrate was undeniable within 1 hour, but I ran them another 12 hours to be sure about what I was seeing.  Essentially, I was not able to get any useful increase in hashrate by adjusting the voltage.  I saw some great first hour numbers, but dropping back to just adjusting frequency on the stock 10/24 firmware was the better alternative.  If others are seeing better results, I guess it is because they have different chips or better cooling.

[pekatete]

However, nothing lasted more than an hour at ambient temperatures of 17 degrees C.

1 hour is nowhere near enought to come to such a conclusion, unless ofcourse the rate is way off, in which case, evn 10 mins would do.

I let them run for over 10 hours, One of mine went down to 400 GH/s and the other to 420GH. So on the long run, mines are not performing well.

My good machine was still doing fine but loosing speed to0 ... about 515gh in almost 24 hours.

The degradation in hashrate was undeniable within 1 hour, but I ran them another 12 hours to be sure about what I was seeing.  Essentially, I was not able to get any useful increase in hashrate by adjusting the voltage.  I saw some great first hour numbers, but dropping back to just adjusting frequency on the stock 10/24 firmware was the better alternative.  If others are seeing better results, I guess it is because they have different chips or better cooling.

I'll have to put it down to all units in the S3 clan being built different ... yet the same!

1. On all units the that I have changed voltage settings, those that showed a good / increase in hash-rate in the 1st hour have always been able to out-perform their previous default setting hash-rate sustainably, albeit, some have an increased HW rate (which I tend to turn down a notch).
2. The voltage setting should, without any qualms, improve efficiency in lower HW rates but more so if you overclock the unit. If you are running at stock freqs, the story may be different .... In effect prior to the voltage setting, we've been overclocking / running rigs at voltages that are not optimal for the higher frequencies. That is why you have to get a better HW rate if you set the correct voltage when you overclock, the speed comes with how far you overclock!

[pekatete]

Yeah, I wish i had some kind of auto-reboot triggered on a timer.

How do you retstart cgminer quickly without rebooting and changing the freq ?

You can do it via a script that you run via scheduled tasks. I had one before I stumbled accross the voltage settings but removed it when my affected rigs stabilised after the changes. You can put a shout out in the S3 support thread or the cgminer thread and someone will respond with a solution.

I dug this up after being asked on the S3 Discussion thread .... again, I have not tried this but it simply reboots rather than restarts cgminer (I now recall I had an issue with restarting cgminer via api, but you can still restart it via init.d script).

Code:

mingh=505
var=$(`cgminer-api lcd|grep -o '\[GHSavg\] => .*$'|cut -d' ' -f3|cut -d'.' -f1`)
if [ $var -lt mingh ]
then
/sbin/reboot
fi

[IITravel01]

Would be nice to see a chart/graph with various frequency settings from 200M up to 262.5M on the S3+ with settings from 0725 to 0770 voltage settings in 005 increments to find the most efficient settings for each frequency in between.  Anyone want to spent the time and equipment to do 11 x 10 = 110 tests???  Would probably become the basic go to OC guide for the Antminer S3.

[luckypyrate]

Would be nice to see a chart with various frequency settings from 200M up to 262.5M on the S3+ with settings from 0725 to 0770 voltage settings in 005 increments to find the most efficient settings for each frequency in between.  Anyone want to spent the time and equipment to do 11 x 10 = 110 tests???  Would probably become the basic go to OC guide for the Antminer S3.

...you mean like the one on pagE 1 of this thread?

[IITravel01]

Would be nice to see a chart with various frequency settings from 200M up to 262.5M on the S3+ with settings from 0725 to 0770 voltage settings in 005 increments to find the most efficient settings for each frequency in between.  Anyone want to spent the time and equipment to do 11 x 10 = 110 tests???  Would probably become the basic go to OC guide for the Antminer S3.

...you mean like the one on pagE 1 of this thread?

Not quite.

[pekatete]

Would be nice to see a chart/graph with various frequency settings from 200M up to 262.5M on the S3+ with settings from 0725 to 0770 voltage settings in 005 increments to find the most efficient settings for each frequency in between.  Anyone want to spent the time and equipment to do 11 x 10 = 110 tests???  Would probably become the basic go to OC guide for the Antminer S3.

To be honest, I've (and many more out there!) have run a good number of that range - ofcourse tending to the OC side - but not for long enough (IMO) to be authoritative enough to form the definitive guide, also undocumented!
Saying that, the full range may be overkill and lack any useful result over and above what we already know, thus I'd suggest voltage increments of .25v over a 6hr period minimum. If anyone has 10 or more S3+'s, that would be trivial, then again, they'd have to "sacrifice" their earnings. If you only have a coupl of S3+'s, it won't require a herculian effort, but trivial it won't be.
Thus IITravel01, start a fund to get this off the ground and I am sure you won't be shy of takers!