Topic: S3+ (BM1382) Overclocking with voltage setting - page 10. (Read 28804 times)

I have just updated the 262.5 overclock over a full day web UI to compare equally with the August 2014 overclock (last image in OP). The HW error rate remains at 0.0004 over the day, a total of ONLY 46 HW errors! Now, there is something to shout home about.

Your guess is as good as mine .... best to drop them a line but I'd defer on the description (encompassing the prior updates). Also, you can try posting on the cgminer thread, I know ckolivas and kano do release binaries compiled for antminer rigs independent of bitmain.

I guess it was before then.  Do their updates include other than items listed in the description on their site?

Did you look? I had'nt before but just did and found this: https://bitmaintech.com/files/download/antMiner_S320141126.bin

Really?  I just checked and didn't see it.  I wonder if this includes cgminer 4.7.1

Aside:  Congrats to NASA and the successful launch of the Orion Rocket!

Yep, the voltage setting does help with reducing the HW error % if set to its respective freq (as per my last edit to the OP @ 262.5). Having said that, if you are running at a freq of 250, there's no harm at all in notching it up to 262.5 since it is the same voltage setting (and thus same wattage draw only run faster!), and with 750w PSU's, you're covered mate!

Basically, the overclocking should have been tied to a voltage setting, but somehow I fear bitmain missed this bit, seeing they only introduced the voltage setting with the newer firmware. On that note, I have also learnt that there is an even newer firmware, relased in Nov 2014 but have not tried it yet.

I'll take the peril of your guidance over the peril of going it alone!  And I suppose you are right.  I have all of my S3+'s (batch 8 and above) now running at 250 on the 4.6.1-1 cgminer and only .0001% HW and ~40-45 degrees.  They all have their own 750w corsair as well.  I tried updating the cgminer to 4.7.1 haven't found a working package yet.

I agree, lower temps help (and I keep my rigs on the balcony .... and we are in winter now thus the 26 degrees in temp!). But more importantly, re-applying the heat paste (or adding heat pads to the front heatsink / chips) does improve the heat dissipation no end, and I'd rather stress that than scare people off with scare stories.

I have also been running this setting for the last few days, and have not for now, (touch wood!), had any x's or dashes. And if you think the HW error % is good, I've had a lot better than that as it tends to settle the longer it runs! But as usual, though I am confident of the valididity of my post, OC at your own risk.

nice info ! very low HW for such high OC n my guess would be the freezing temps you have. tbh, not advised to go that high if one cannot get enough cold temps as the chip temp will soar high hence more hw, prolly some x's or dashses - , fan spinning at higher rpm = not efficient, higher rate of failures, higher power consumption but wth it's at 1 own risk.

op, i'm not trying to spoil the thread but just my honest opinions nevertheless it's a good research n experience op is sharing with the community.

be warned oc at your own risk

peace.

Follow me at your own peril!
Seriously, there is nothing un-stable about it so long as you have enough power to meet your overclocked wattage draw AND adequate heat dissipation. If the HW error % gets to single digits, then I'd be worried, otherwise these things are built like tanks! And by the way, I'll add my poolside hash-rate for comparison in time too, just to underline my assertions.

I was talking about overclocking beyond the stock asic-freq config.  I thought I read somewhere it was unstable.  *shrug* maybe not, it would have been you saying it if so.  I don't follow anyone else in regard to antminers.

Not sure what yo mean there ....


It is a go for either, so long as you have the latest firmware installed.
I have to add that if you are trying it on a unit that you have not retouched with heat paste / heat pads, try the lower frequencies (thus lower voltages)  first as bitmain's heat paste application is good for the stock frequencies (they've significantly improved on later batches though). The unit that I show here is one I re-did with heat pads on the chips / front heatsink + re-applied heat paste on the rear heatsink.

Fantastic work.
Is this a go for S3 or just S3+

Though I would caution against it, as I think they do as well in that post.  Your call though.

I like that you think so ....


Re-read the post. I put a link to a forum post that explains how to add the extra frequencies.

How are you able get your freq beyond 250?  Are you puttying into the unit and manually updating the asic-freq file?

You, sir, are amazing. 

It has been suggested that the voltage setting under the Miner Configuration -> Advanced Settings tab that is in the new S3+ firmware is a left over from the S4 firmware and is of no effect to the S3+, however this is not the case. While trying to figure out the effect the setting had, I changed one of my slightly OC'ed S3's voltage setting to 0850 and left it there for as it seemed to slightly improve the speed it was hashing at. This unit (like the rest of my S3's) was powered by a 550W server PSU, and having left the unit to run for weeks, the PSU started restarting every so often, which none of the others ever did. I finally realised that the PSU was dead and the only difference in settings was the voltage, which confirmed to me that the setting surely has an effect on the S3+, thus this post.

I looked up the datasheet for the BM1382 chip that is used in the S3+ and found 2 sections that helped me work out what caused the fauilre of my PSU (and also led me to properly use the voltage setting to properly overclock an S3+).

1. The typical hash rate and power table



2. The input timing for the chip



The initial table has voltage settings and hash rates at each voltage, while the second table has the hash rates and frequencies to achieve them. Putting these together, I came up with the following.



I have added an extra row for each voltage setting of 0.75v and 0.80v as I found that with the two units I overclocked they had a higher HW error % at the frequencies in the datasheet. When I upped the frequency a notch, the HW error % reduced to acceptable levels. It goes without saying that you may need to add the added frequencies as they are not in the shipped firmware, here's alink to how you can add them: https://bitcointalksearch.org/topic/m.8370071

Also worthy of note is that setting a chip voltage to 0.85 results in a (datasheet / theoretical) wattage draw for the S3 (32 chips) of 544 watts. Remember the server PSU that I was using on the over-volted S3 that I mentioned earlier was rated at 550 watts ..... no wonder it gave up its ghost! So ensure you have an adequate PSU powering your S3 before you begin. Addidtionally, it may be worth noting that a fully populated 6 pin PCI-e connector will carry a maximum of 24 Amps @ 12V = 288 Watts (though I normally assume they can only carry 22.5 Amps @ 12V = 270 Watts), so you may need to power your S3 with all 4 pins rather than just 2.

Finally, here's an image of one of the OC'ed units. The HW error % is a bit high on this, but the unit has hashed at lower rates for longer runs, and also higher hash rates! EDIT: The lower part of the image is the poolside registered rate(s) after OC'ing.



EDIT: It pains me to have to admit that the same voltage OC'ing will apply to the bitmain variants that have the BM1382 chips, e.g the S2 and S4; the reason it pains me is I do not have one!

EDIT 2: I thought I'd also post this to compare with my earlier OC'ing of my batch 6 S3 in this post: https://bitcointalksearch.org/topic/antminer-s3-batch-6-overclocking-750220 (also in my signature).

When I ran the test in August 2014 at freq 262.5 with the old firmware, I achieved an average hash-speed over a day of 529 Gh/s with 200 errors (0.00187113284%). Having set the voltage to 0750 in the newer firmware, I have a hash-speed 528.53 and a HW error rate of only 46 HW errors (0.0004%).

So again, not only does the voltage setting aid in OC'ing, it certainly does help with running efficiently!



EDIT: 6th Dec 2014

I had to post this. An update for the 262.5 freq @ 0.75 volts (0750 setting) after running for 2 days. There are only 96 92 HW errors and the % is still the same! The consistency you get from setting the voltage correctly is astoundingly repeatable / maintainable given what we've come to expect of ASIC rigs.