Topic: Antminer S3 batch 6 overclocking - page 5. (Read 23027 times)

^^ No worries.
Updated OP with link to adding frequencies to latest firmaware (Aug 2014) and highlighting the deprecation of cgminer 4.6.0

doh!  *grabs more coffee*  sorry.

Thanks for mentioning that the link to 4.6.0 is no longer available (and I will add a note to that), however, just below that link in the OP is the link to cgminer 4.6.1 (including the similar process of installing it).

FYI the link for cgminer is no longer valid.

wget http://ck.kolivas.org/apps/cgminer/antminer/s3/4.6.1-141020/cgminer will get you there

I have days when I notice the HW numbers spike to astronomical figures on the OC'd S3's, but a reboot via SSH always "settles" the unit down. Having said that, it may be one of several things, and I could be wrong but I'll offer a few:
1. You are not powering your S3 properly (I used only 2 connectors, but made sure the cables can adequately juice the unit)
2. If it is an earlier S3, you may need to re-apply the paste (bitmain did a very amateur job!) to enable your unit handle higher temps better.
3. Your unit's freq sweet spot is 250M (some units are less others more ... luck of the draw!).
On a personal level, if it hashes at 505GH/s with acceptable HW errors, I'd be over the moon.

Alright. One final question before I go to bed. I've been messing with overclocking my miner all day, and I've found that running it at 250M delivers expected results. It runs at an average of 505 Ghash/s with low temps and low hardware errors. As soon as I kick it up to 256.25M, my temps stay down and it shows all 'o's for the status, but I get tons of hardware errors and low hash rates. I''m getting half as many hardware errors as accepted shares and about 480 Ghash/s. Why is this?
By the way, thanks for helping out this noob so much. It's much appreciated.

Yep, that is the post.
cgminer 4.6.1 does NOT come with the firmware so you'll need to follow the instructions to install it.
There are tables of power usage at different frequencies floating around, but again the best reference is the S3 thread.

PS. I'll add the info in the link about adding frequecies to the dropdown to the OP at a later time.

Thank you.

https://bitcointalksearch.org/topic/m.8370071

That's the instructions just in case anyone else has the same question. Took me a while to find.

Now for some more probably already answered questions.

Do I have the firmware you mentioned in the beginning of this thread called cgminer 4.6.1?

Has anyone made a table of power usage at different frequencies?

1. Looks like you have the latest firmware (going by the date). When you check the bitmain site for firmware, the latest S3+ firmware is antMiner_S320140826.bin, and as you can decipher from the filename, thats your date! Also, the latest firmware has the dropdown frequency selection, whereas the former one does not.

2. To add the 262.5 frequency to the dropdown, you need to edit (via SSH) a different file ..... can't recall off my head which it is, but search through the S3 support thread, it is mentioned there. And no, you do not have to update your firmware.

Hey so I have a couple questions.

1. How do I know what firmware version I have? Under firmware version in the miner GUI is: Tue Aug 26 10:43:10 CST 2014

2. I have been running at 250M for 24 hours now and my temps are 40 and 43 and I have no "x"s. I'd like to overclock it to 262.5 like you've shown. How do I do this as you told me the other day in a different thread that the only way to overclock on the S3+ firmware is to use the advanced tab under miner configuration, but as I'm sure you know, it only supports a max of 250. Do I have to change firmware?

Thank you

LOL. Now if I can just get my machines to turn into block magnets when they're actually turned on! 

If it is better than before, then you are making some headway. The falling hash-rate, like I always want to think, may just be a more difficult (no proof of this, just my thinking) block we are on so you are better off getting the picture / speed after at least 24hrs.


I am glad that you have taken over that mantle! Slush always seemed to do this to me, though of late, not so much ..... but now I know why, YOU!

Many thanks.  I'll try your suggestion about the big heatsink next.  My present hashrate is 454 but falling very slowly.  Better than before anyway.  As you suggest, best to keep the blue wire hack as the last resort!
It's amazing how Slush always finds a block when I have a machine offline!
 

In my experience (and use), when / if the pool-side speed is within or above the expected, I tend to ignore the x. I actually have an S3 (though overclocked to 262.5) that from time to time gives the a couple of x's on the first chain and on other occassions one x on the second chain. My solution ...? I added a line to the scheduled tasks to reboot after every 8 hrs (that was after observation of the pattern of the x's appearing!).


My point exactly above (with regard to the x disappearing after reboot). Bummer that you had to re-do the paste, but since the S3 units to some extent are unique, that seems the only way you can get a solution for any individual unit / batch. Still, if all fails for now (and I were in your position), my last throw of the dice would be the blue wire hack, aka both fans at full pelt!

PS. If you get the chance to re-do your rig and detach the board from the big heatsink, could you clean the area in the middle behind the heat sensor chip? As in remove all and any paste that may be there (that will mean (I hope) the sensor will detect higher temps and thus run the fans faster).
EDIT: - aka in lieu of the blue wire hack!

Yes, it is one of the early units and when received, thermal paste was splashed all over the asic connectors. After cleaning, I first tried Artic Silver, which I use when o/c CPUs. This gave me the lowest temps I have seen with this S3, but after scare stories on the S3 o/c thread about Arctic Silver being electrically conductive, I removed it and applied Prolimatech, which increased temps by 1 degC compared to Arctic Silver.  I wasn't impressed by that, so bought some Zalman paste to try, which comes in a little bottle with a brush - easiest to apply by a mile, but quite expensive.  This gave the same temps as the Prolimatech.  TBH, I wish I hadn't removed the Arctic Silver now, although it's horrible to apply as it tends to trail as you lift off the applicator, so massive care is needed. 
Anyway, to try to get rid of the 'x', I redid all of the paste this morning, using the Zalman, as it's the quickest.  Unfortunately, the first time I ran at 225 again, the 'x' showed up in chain 1.  Had to reboot to get rid of it and reset to 218.  Just out of interest, I then tried it at 231.25.  After about 30 minutes, I got an 'x' again, but in chain 2 this time!  The chain 1 'x' has not re-appeared after 1.5 hours at 231.25.  Is there any risk, do you think, running with an 'x', because it seems to run very well at 231.25 apart from that?  Oh, BTW, I'm also running the S3 on its own Enermax PSU now, although as you suspected, I don't think that's the problem.

It may indeed be the powering but that PSU should be very confortable running 2 S3's at stock (even overclocked within reason!).
That you get an x when you clock to 225 is rather telling though. If this is one of the very early S3's, did you get a chance to re-apply the heat paste?
I mention this because with one of my upgraded S3's (from S1), I used heat-pads rather than paste on the chips, and intitally some of the pads moved slightly from the chips when I was adding the heatsink; that unit kept "throwing up" and x now and again, and after updating to 4.6.1 it became a "main-stay". I decided to check the rig, unscrewed the heatsink and discovered what I mentioned above, aka the slipped heat-pads. I put them in place and added a bit of paste to the back and the x is now history (well, its only been a few days, but it has not appeared again ..... yet).
So it may be the case that if you have an early S3 (which were notorious for too much paste or non at all!) AND the unit was hashing OK before applying 4.6.1 then it is worth investigating that, otherwise I'd say 4.6.1 does not suit that unit.

This S3 is one of two powered by an EVGA1000G. It's current average is 440 (I rebooted it 2.5 hours ago as it was drifting downwards quite badly), running at stock speed.  HW is 2.  Ambient is 27 degC - I close the windows at dusk, before that it was 24.  Temps currently 45 (was 43 with windows open).  Increasing frequency to just 225 immediately causes an 'x' in one of the asics.  I will try running the S3 on its own PSU tomorrow to test the power aspect.  Cheers.

Are you powering your rig properly and is your HW rate "normal"? I'd not worry too much about the 5s rate, as it can shoot all the way up to (in my case) 650, the key one is the avg rate. 438 seems low, but it "should" pick up in time (mine usually take 45mins to an hour to reach their peak avg).
PS. Also, what is your ambient? Do your temps look OK?

My initial impressions with the new firmware and 4.6.1 is that the hash rate is very erratic, e.g. set on standard frequency of 218.75, it's varying between about 327 and 500, average so far 438, but gradually reducing.  I'll give it some time to settle.

Yes, you were right. 
From the S3 support thread, you use: /usr/bin/cgminer-api
Along with a lot of other stuff, this reports that I do have version 4.6.1.
Cheers.