ANTMINER S3+ Discussion and Support Thread

[MoreBloodWine]

Ok, well no word back yet from he guy hosting my machines but I'll see if when he replies I can get hard numbers on the current PSU's and go from there. I guess all I'll need it just the brand / model.

I just don't want to continue losing hash like this.

could it be just fluctuations: mine keep going up and down, even in the average at the pool.

Possibly and that's what I thought, but it seems like a drastic drop for 2 days and change. At least on the one miner that lost 16.xGh/s

I'm still gonna post the PSU info when I hear back from my host.

[1715518181]

1715518181

[1715518181]

1715518181

[1715518181]

1715518181

[1715518181]

1715518181

[1715518181]

1715518181

[crazyates]

Question: has anyone tried taking these apart, and applying small heatsinks to the DC-DC converters, like people used to do with the S1s? It was my impression that in early models, these were the limiting factor in being able to OC, so keeping them cool might make a difference?

https://i.imgur.com/VN07slQ.jpg

[Biodom]

Question: has anyone tried taking these apart, and applying small heatsinks to the DC-DC converters, like people used to do with the S1s? It was my impression that in early models, these were the limiting factor in being able to OC, so keeping them cool might make a difference?

https://i.imgur.com/VN07slQ.jpg
https://i.imgur.com/VN07slQ.jpg

https://i.imgur.com/RkuPkLy.jpg

https://i.imgur.com/hjOa049.jpg

apparently, some people did (plus on smaller black chips) with no discernible positive effects (i don't have links).
Also, the fact that batch 5-6 were much more stable and OCable also shows that it was just something that was fixed by Bitmain (without heat sinks).

[crazyates]

Another question: I just set up and configured one of my S3s, and it was hashing fine, until I rebooted it. Now, it never starts hashing. The miner status page never shows anything, it never pulls more than 45W, I'm not getting any shares, but 'top' tells me cgminer is using 2-30% CPU every couple of minutes, which I'm assuming is it checking into the pool and prepping more work (that never gets used...)

EDIT: Got it all figured out.

[jpchrist]

Actually, it does matter - I have an 850 Watt PSU that can't handle OC'ing a single S3 (barely handles stock speeds).  It loses hash compared to now that I run the 850 one one blade using two separate cables from the PSU (i.e. not a single cable from the PSU that has two 6-pin PCI connectors) and one 600 on the other blade.

Yep, a single 650 will likely not be able to handle OC'ing one S3.  IMHO.

Was that 850W a single 12V rail PSU?

My wimpy 850 was single rail, rated at 58A.  The voltage stability and level on cheaper power supplies can have a lot of variance and can even be just under 12V on a multimeter.  Don't trust ratings alone.  ;-)

What brand?

It was a SolidGear Neutron 850 - it's what I had lying around and thought/hoped it would work.  Not so much.   

I had a solid gear psu that fried an S1. Avoid them at all costs!

[Biodom]

Another question: I just set up and configured one of my S3s, and it was hashing fine, until I rebooted it.

Is this normal?

rebooted how: cgminer, power cycling or poweoff -d <number> command in ssh?
In my opinion S3 does not "like" to be power cycled, especially if you don't give it 10-20 sec before turning it back "ON"
try rebooting the cgminer in <system>. It might work.

[twib2]

I had a solid gear psu that fried an S1. Avoid them at all costs!

For me, they're just anemic, but I knew that going in.  But yep, I'm probably cycling all of my cheapos up as soon as I can.

[crazyates]

Another question: I just set up and configured one of my S3s, and it was hashing fine, until I rebooted it. Now, it never starts hashing. The miner status page never shows anything, it never pulls more than 45W, I'm not getting any shares, but 'top' tells me cgminer is using 2-30% CPU every couple of minutes, which I'm assuming is it checking into the pool and prepping more work (that never gets used...).

Ok, I put it asside, and moved onto setting up my next one. Idk what I'm doing wrong, but this one's doing the same thing. I've narrowed it down to when I change the pool info, but I'm using the exact same pool info as all my other Ants (Ctrl C+V right out of a spreadsheet), and cgminer isn't loading once I change the pool info. I've tried just one pool, then the other.

Another question: I just set up and configured one of my S3s, and it was hashing fine, until I rebooted it.

Is this normal?

rebooted how: cgminer, power cycling or poweoff -d <number> command in ssh?
In my opinion S3 does not "like" to be power cycled, especially if you don't give it 10-20 sec before turning it back "ON"
try rebooting the cgminer in <system>. It might work.

Yep I usually reboot them through the web page. When that didn't work, I did power cycle it for ~10-15 seconds. See above, as I think it's just a config issues...

[crazyates]

I've narrowed it down to when I change the pool info, but I'm using the exact same pool info as all my other Ants (Ctrl C+V right out of a spreadsheet), and cgminer isn't loading once I change the pool info. I've tried just one pool, then the other.

Weird, the Ants don't seem to like my 192.168.0.1 gateway as a dns server, which is what every other computer is set to. Set the DNS to 8.8.8.8, and now it's mining just fine.

[MoreBloodWine]

Ok, so were 3 days 5 hrs in on the 243.75's and I've noticed some drops in hash. I know machines have variance of +/- 10% sometimes but it seems like a steady but slow decline. What's weird is the time passed being only 2 days and change, compared to say a month or something.

Anyone else get this with OC'd machines ?

Power supplies.  They're probably close enough to the edge of the kind of power the S3's like (beefier than they draw) that just staying hot for a long time is enough to suck hashing power.

Do you have 2 PSU's on each OC'd unit, or only one?  And are you using two separate 6-pin PCI power connectors for each blade (not daisy-chain style) or only one?

My suggestion is to either cool the PSUs and see if your rate goes up, or use more/better PSUs.  Poor power doesn't always show up in x's or -'s now that I've been playing a bit with my good S3 and just got 60 GH/s back by adding another PSU.

Worth a try!

My friend hosting them said and sent pics that each one is on a singular supply of 650W or greater with each miner using all four power ports.

Should I find out the specific PSU not that I think it matters with each one being 650w or greater.

Edit:

--- And are you using two separate 6-pin PCI power connectors for each blade (not daisy-chain style) or only one?

I actually just shot him an email about this but do you mean four cables using four ports on both the miner and PSU ?

I also dont get the beefier thing.

So I just heard back from my host.

He said hat each miner is in deed powered by all four ports on both the PSU / miner using four individual power cables. I have since asked him to please find out the specific make / model of each of these PSU's so the rail voltage info etc. can be looked up to see if maybe, just maybe, it is the PSU's which is questionable a this point since it still could just be system variance / an OC thing.

[philipma1957]

They will ROI within 2.5 months as long as you ordered enough of them to beat the curve.

Someone needs to work on their math skills, if one ant is unprofitable then 1000 of them will be as well as long as the cost of running them /ant stays the same.

well it depends.

1 s-3 and 1 evga 1300 for your psu won't roi.

3 s-3's and 1 evga 1300 for psu might roi .

you did say as long as cost to run them does not change but  maybe the poster was thinking like I just wrote

I am running 3 S3 (237.5 mhz) on a single EVGA1300 using PCI-e splitters on two machines (so each has 4 connectors).
What was your wattage on EVGA1300 per S3 when running at 250mhz. I wonder if I have enough power to bump all three to 250mhz.

 390 watts on a seasonic platinum          I do 231.25+231.25+237.50 and pull 1110 watts on the evga 1300    .  you could try 243.25 243.75  I think that would have enough juice.

@ twib2 I sent a pm to you psu info

[MoreBloodWine]

They will ROI within 2.5 months as long as you ordered enough of them to beat the curve.

Someone needs to work on their math skills, if one ant is unprofitable then 1000 of them will be as well as long as the cost of running them /ant stays the same.

well it depends.

1 s-3 and 1 evga 1300 for your psu won't roi.

3 s-3's and 1 evga 1300 for psu might roi .

you did say as long as cost to run them does not change but  maybe the poster was thinking like I just wrote

I am running 3 S3 (237.5 mhz) on a single EVGA1300 using PCI-e splitters on two machines (so each has 4 connectors).
What was your wattage on EVGA1300 per S3 when running at 250mhz. I wonder if I have enough power to bump all three to 250mhz.

  390 watts on a seasonic platinum          I do 231.25+231.25+237.50 and pull 1110 watts on the evga 1300    .  you could try 243.25  I think that would have enough juice.

Don't you mean 243.75 ?

That's what you told me and what I'm using on 3 of 4 miners.

[Pieh0]

I've narrowed it down to when I change the pool info, but I'm using the exact same pool info as all my other Ants (Ctrl C+V right out of a spreadsheet), and cgminer isn't loading once I change the pool info. I've tried just one pool, then the other.

Weird, the Ants don't seem to like my 192.168.0.1 gateway as a dns server, which is what every other computer is set to. Set the DNS to 8.8.8.8, and now it's mining just fine.

Maybe because your router isn't a dns server/forwarder, most routers just send the isp's dns server ip's over DHCP.

[philipma1957]

They will ROI within 2.5 months as long as you ordered enough of them to beat the curve.

Someone needs to work on their math skills, if one ant is unprofitable then 1000 of them will be as well as long as the cost of running them /ant stays the same.

well it depends.

1 s-3 and 1 evga 1300 for your psu won't roi.

3 s-3's and 1 evga 1300 for psu might roi .

you did say as long as cost to run them does not change but  maybe the poster was thinking like I just wrote

I am running 3 S3 (237.5 mhz) on a single EVGA1300 using PCI-e splitters on two machines (so each has 4 connectors).
What was your wattage on EVGA1300 per S3 when running at 250mhz. I wonder if I have enough power to bump all three to 250mhz.

 390 watts on a seasonic platinum          I do 231.25+231.25+237.50 and pull 1110 watts on the evga 1300    .  you could try 243.25  I think that would have enough juice.

Don't you mean 243.75 ?

That's what you told me and what I'm using on 3 of 4 miners.

 yeah i did a  typo  

 243.75  is correct number. 


my 1300 watt evga was able to do 3 s-1's   and at freq 400 and 200gh the 3 pulled 1215 watts for 1 month  24/7 .  I then backed them off to 387.5 and 190gh pull about 1150 watts ran better using 1150 watts then 1215 watts

[visdude]

You only need to read a few pages back to discover that miners (including myself) have successfully been OC'ing S3s with mere 500W PSUs with only 2 PCIe power inputs.

Yes, I've been following this thread since the beginning.  ("discover" - really?  Nice.)

But I'm talking about his results and the fact that his situation with his 650 matches my experience with a wimpy 850.  The QUALITY of a PSU is more important than the rating above a certain level (and 500 sounds about right for a really good high-efficiency PSU).

I'm not as concerned with the 2 PCIe inputs as I am with good PSUs.  But if you have a weak one with separate weak or marginal rails, then 2 inputs per blade could very well help.  My thought is that it is the "sag" the poorer powersupplies experience with a strong maintained load that is the problem.

Back in the day with AM Cubes, we had the same persnicketiness with the quality of solid, stable PSUs.  I had a couple of server 750s that would each run 2 OC'd Cubes no problem at a time when folks sometimes couldn't get their Cubes to run with a single 750 (or even 850).

This is not my first rodeo, but thanks for your kind input.  I stand by what I wrote, and I think he would be well-served to try better (or more) PSUs - it worked for me, and his stated symptoms match what I was seeing.

Yes, "really".  I was responding to your post at face value (I honestly thought that you didn't have a clue) so I attempted to let you be aware of a fact.  That's all.  If "discover" touched a nerve in you, that's too bad.  If it makes you feel any better, I hereby retract "discover" and substitute it with "find out".

It's not "first rodeo" for the majority of us in here either.  Therefore, the common knowledge that the quality of a PSU affects its performance is nothing new in here (or any decent tech forum for that matter) and is not exclusive to you.

To reiterate, contrary to your post (verbatim) that I originally responded to, it's a fact that the S3s have been successfully overclocked using 500W PSUs utilizing only two PCIe power inputs for each unit.  I refuse to allow disinformation to propagate.

[visdude]

Ok, so were 3 days 5 hrs in on the 243.75's and I've noticed some drops in hash. I know machines have variance of +/- 10% sometimes but it seems like a steady but slow decline. What's weird is the time passed being only 2 days and change, compared to say a month or something.

Anyone else get this with OC'd machines ?

First image is the 237.5 machine that seems to be "leaking" as well.

http://www.eojmarket.com/NonForumStuff/images/bitcoin_stuff/823S30001.jpg (- 6.x Gh/s)

http://www.eojmarket.com/NonForumStuff/images/bitcoin_stuff/823S30002.jpg (+1.x Gh/s)

http://www.eojmarket.com/NonForumStuff/images/bitcoin_stuff/823S30003.jpg (-16.x Gh/s )

http://www.eojmarket.com/NonForumStuff/images/bitcoin_stuff/823S30004.jpg (-5.x Gh/s)

Total loss around 27 Gh/s, 26 if you subtract the +1 on machine two.

The S3 exhibits inconsistent and confusing hashing patterns; at least in my case anyway.

At 243.75M, my units could be around 478GH/s within the first hour and slowly improve to reach a little over 491GH/s in the next 4 hours or so.  But then again, in a different session, they could also be already at 491GH/s within the first hour and slowly deteriorate to around 488GH/s after an extensive period of time (a day or so) but only to eventually creep back up to between 490-491GH/s and stay there for a while.  My pool side averages are much higher (generally a tad over 500GH/s) but just as erratic.

I miss the consistency of the S1s.  They were always at 179.xxGH/s from week to week (up to a couple of months) continuously.

[nexus99]

Or you could have units that never get past 420 GH/s.

[philipma1957]

Or you could have units that never get past 420 GH/s.

I have a batch one pair that do 430 on 212.5.

  I was not able to do better then those numbers.  my batch 4 and batch 6 pairs are better.

[twib2]

Yes, "really".  I was responding to your post at face value (I honestly thought that you didn't have a clue) so I attempted to let you be aware of a fact.  That's all.  If "discover" touched a nerve in you, that's too bad.  If it makes you feel any better, I hereby retract "discover" and substitute it with "find out".

It's not "first rodeo" for the majority of us in here either.  Therefore, the common knowledge that the quality of a PSU affects its performance is nothing new in here (or any decent tech forum for that matter) and is not exclusive to you.

To reiterate, contrary to your post (verbatim) that I originally responded to, it's a fact that the S3s have been successfully overclocked using 500W PSUs utilizing only two PCIe power inputs for each unit.  I refuse to allow disinformation to propagate.

Since we're all retract-y here, I'll choose one that actually might make a difference: I hereby retract my "a" and replace it with "your" when I was talking about the 650.

Note that I never disagreed with your assertion and even referred to it later.  Nevertheless, I still think my response will prove to be helpful, which was my only intent.  If you disagree with the crux of the advice (better or more power might be worth investigating), that's fine, but if you didn't then indulging in pedantry for the sake of a side-point just distracts.

[twib2]

Or you could have units that never get past 420 GH/s.

I had the same thing until I added more power or used better PSUs, even going from using a piss-poor 850.  If you have other PUSs handy, it might be worth it to test that and rule it out to be sure.  Other things could be network problems, pool problems, using the same miner, sub-optimal miner difficulty settings, or even the already-noted S3 mining rate instability.  I'm sure you've already ruled out most of those, just throwing them out there anyway...