Bitmain's Released Antminer S9, World's First 16nm Miner Ready to Order

[ComputerGenie]

...For me, the most cogent and simplest explanation for the burned asics is their wretched autotune software....

That is because you, quite literally, have no clue what you're on about.  

So be so kind to enlighten me with your wisdom and, while doing that, explain why the high failure rate just started with the autotune models.

In Psychology, they call it "selective perception".  

[Biodom]

...For me, the most cogent and simplest explanation for the burned asics is their wretched autotune software....

That is because you, quite literally, have no clue what you're on about.  

So be so kind to enlighten me with your wisdom and, while doing that, explain why the high failure rate just started with the autotune models.

I don't think that it did, really. If anything, autotune was introduced to ameliorate a high failure rate of non-autotuned S9 because, among other things, people probably tried to overclock them as they did with S5 and S7. Naturally, i don't really blame anyone doing this as margins are thin. i don't think that i have the exact numbers, but, anecdotally, non-autotuned S9 had at least 10-13% board failure rate, maybe even more initially.

It would be interesting to know the autotuned numbers. Any large scale miner can calculate this number for his/her farm and at scale it would have a smaller standard error.

[ComputerGenie]

...i don't think that i have the exact numbers, but, anecdotally, non-autotuned S9 had at least 10-13% board failure rate, maybe even more initially...

That's right about in the margins we experienced.

[fanatic26]

So be so kind to enlighten me with your wisdom and, while doing that, explain why the high failure rate just started with the autotune models.

I will be kind enough to enlighten you. You are talking out your ass with no hard numbers to back up the claims you are making. I have personally managed hundreds of these machines and can say with certainty that the autotune software did not increase failure rates. It just made deployments more annoying because in some cases you have to wait hours for it to settle on a speed and start actually hashing so you can verify full functionality.

As far as the testing the autotune function does, it is to find the weakest chip on the chain and not run the system outside of that spec. As these are a string design you cannot target and test individual chips without powering the rest of the chain to pass its data to the controller. It does stability checks to make sure it is running at the optimal voltage for those specific chips on that specific board.. It is not designed to push the absolute maximum out of the board as you seem to erroneously think.

[Micky25]

Thank you guys for shareing your experience. For me the problems started with the autotune models and I can only speak out of personal experience. From my point of view, if the autotune software burns chips, it doesn't do its job good.

[OgNasty]

Thank you guys for shareing your experience. For me the problems started with the autotune models and I can only speak out of personal experience. From my point of view, if the autotune software burns chips, it doesn't do its job good.

Agreed.  There should be an underclock option.

[ComputerGenie]

A literal example of why autofreq doesn't "burn chips" and isn't "pressing the utmost hash out of each chip"...
Chain 1(left out because it's redundant) and Chain 2 are from Batch 4 and Chain 3 is from a "preset" batch:

Code:

read PIC voltage=940 on chain[2]
Chain:2 chipnum=63
...
Asic[ 0]:625 
Asic[ 1]:625 Asic[ 2]:625 Asic[ 3]:625 Asic[ 4]:625 Asic[ 5]:625 Asic[ 6]:625 Asic[ 7]:625 Asic[ 8]:625 
Asic[ 9]:625 Asic[10]:625 Asic[11]:625 Asic[12]:625 Asic[13]:625 Asic[14]:625 Asic[15]:625 Asic[16]:625 
Asic[17]:625 Asic[18]:625 Asic[19]:625 Asic[20]:625 Asic[21]:625 Asic[22]:625 Asic[23]:625 Asic[24]:625 
Asic[25]:625 Asic[26]:625 Asic[27]:625 Asic[28]:625 Asic[29]:625 Asic[30]:625 Asic[31]:625 Asic[32]:625 
Asic[33]:625 Asic[34]:625 Asic[35]:625 Asic[36]:625 Asic[37]:625 Asic[38]:625 Asic[39]:625 Asic[40]:625 
Asic[41]:625 Asic[42]:625 Asic[43]:625 Asic[44]:625 Asic[45]:625 Asic[46]:625 Asic[47]:625 Asic[48]:625 
Asic[49]:625 Asic[50]:625 Asic[51]:625 Asic[52]:625 Asic[53]:625 Asic[54]:625 Asic[55]:625 Asic[56]:625 
Asic[57]:625 Asic[58]:625 Asic[59]:625 Asic[60]:625 Asic[61]:625 Asic[62]:625 
Chain:2 max freq=625
Chain:2 min freq=625

read PIC voltage=940 on chain[3]
Chain:3 chipnum=63
...
Asic[ 0]:568 
Asic[ 1]:606 Asic[ 2]:568 Asic[ 3]:593 Asic[ 4]:606 Asic[ 5]:575 Asic[ 6]:593 Asic[ 7]:516 Asic[ 8]:612 
Asic[ 9]:593 Asic[10]:556 Asic[11]:612 Asic[12]:593 Asic[13]:593 Asic[14]:533 Asic[15]:600 Asic[16]:606 
Asic[17]:550 Asic[18]:600 Asic[19]:606 Asic[20]:533 Asic[21]:606 Asic[22]:606 Asic[23]:504 Asic[24]:606 
Asic[25]:606 Asic[26]:500 Asic[27]:606 Asic[28]:606 Asic[29]:587 Asic[30]:606 Asic[31]:606 Asic[32]:600 
Asic[33]:606 Asic[34]:606 Asic[35]:612 Asic[36]:606 Asic[37]:612 Asic[38]:612 Asic[39]:606 Asic[40]:612 
Asic[41]:612 Asic[42]:587 Asic[43]:612 Asic[44]:612 Asic[45]:606 Asic[46]:612 Asic[47]:606 Asic[48]:606 
Asic[49]:606 Asic[50]:612 Asic[51]:606 Asic[52]:612 Asic[53]:612 Asic[54]:606 Asic[55]:575 Asic[56]:612 
Asic[57]:593 Asic[58]:612 Asic[59]:612 Asic[60]:606 Asic[61]:606 Asic[62]:612 
Chain:3 max freq=612
Chain:3 min freq=500

Code:

Chain#	ASIC#	Frequency(avg)	GH/S(ideal)	Temp(Chip2)
				
1	63	625.00		4,488.75	87
2	63	625.00		4,488.75	84
3	63	594.57		4,270.21	73

[Dibblah]

Indeed. It also doesn't seem to do _anything_ per-run - it's just validating the speeds stored in the PIC allow the card to hash sensibly.

It is possible to force it to recalibrate itsself, but apparently that's a support call to BitMain where they add your specific MAC to 'the list' and it gets to run their utility, which presumably does some sort of encrypted communication to the mothership.

Interesting that the non-autotuned boards all show a static clock speed. I have one of these in my autotuning miner and was wondering the cause - it would seem that this might just have been old stock boards.

[Sierra8561]

You are free to run whatever client you wish or support a different chain. Nobody is forcing you to use Core. However, the best and most competent developers contribute  and maintain code to the core client. You are free to open a Git and submit proposals if know of a beneficial feature that is not currently available.

If you do not support segwit, run versions prior to 0.13.1, that is if you run a node.
...

Yes, it is wholly true that if you want to run a Core node, you are free to not support segwit by running an older, less optimized version of Core that is worse for the network as a whole.
Meanwhile in the decentralized world where Bitcoin is supposed to be run by a consensus, if the majority want x in a wallet, then Core devs should put x in the wallet. In the decentralized world where Bitcoin is supposed to be run by a consensus, one shouldn't have to make the choice between a client that is, literally, slower and one that signals something that the majority obviously don't want.

"You are free to open a Git and submit proposals if know of a beneficial feature that is not currently available." Satoshi Nakamoto, himself, could write a BIP that 99% of the community wants and if the devs personally dislike the change, then it's not going into Core.

Why not run a bitcoin unlimited node?

[Micky25]

Thank you guys for shareing your experience. For me the problems started with the autotune models and I can only speak out of personal experience. From my point of view, if the autotune software burns chips, it doesn't do its job good.

Agreed.  There should be an underclock option.

Well, apparently we seem to get this option by doing the HagssFIN hack - if we dare    Did somebody?

There use to be such high demand for manual control ability with autotune S9 models, but now, has someone tried my method with S9?
I've done this successfully to my two R4s.

This is also S9 related, so I thought to share this here as well.
If there is someone with a desire for non-autotune settings, be brave and please report if you get same results as I had  

Originally posted in the R4 thread https://bitcointalk.org/index.php?topic=1597639.msg18267781#msg18267781

Well, this is an interesting find and I thought I'll share it with you.

I had a problem booting my Antminer R4 8.0 Th/s, batch 2, autotune-model.

So I went looking in here:
Bitmain.com: Three Ways to Restore Factory Settings (R4/S9/T9)
https://enforum.bitmain.com/bbs/topics/3957

I used the last option in the list, IP Reporter button restore.

Usage: Please power off the miner, then hold down the IP Reporter and don’t release it. At the same time, please power on the miner.
Releasing the IP Reporter after 5 seconds, the machine will automatically restore factory settings.

My miner restored with firmware Aug. 9 2016 and autotune settings were gone.
I am now suddenly able to control frequency and custom fan settings.  

This was not my original problem, but I'm happy with it.
My original problem was that the miner jammed somehow in the booting process and didn't even show up in my router ip list.

[ComputerGenie]

Why not run a bitcoin unlimited node?

Yes, because the answer to stopping the centralization of Bitcoin around the Core dev's beliefs is to centralize Bitcoin around the BU dev's beliefs.... 

[ComputerGenie]

Oh, great douchebag of Bitcoinia... Please tell us which blessed software graces your full-node ?

A version of Core 0.14 that took 2 weeks to edit out the segwit garbage.

[ComputerGenie]

A version of Core 0.14 that took 2 weeks to edit out the segwit garbage.

Oh my ! LOL !

Yeah, I know it's funny; it should have taken me way less than 2 weeks to recode a wallet that took a team of devs 2 months to code, right?

[Bigsky]

I hope someone more knowledgeable than myself may be able to give me an Idea
of what may be the issue with my batch2 S9, it is currently running somewhere
very cold, keeping the eskimo's warm, and I dont have remote access to it I just
moniter it poolside

It ran pretty well flawlessly for around 6 months at 12.8th, it then started having issues of
dropping back to 8.6th for a week or longer then back to 12.8 then to around 11th for
four days then back to 8.6 for a few days them 12.8 then 8.6 where it has been for
a week or so now. Reboot/restart worked the first time, but last time did not seem to
make any difference.

Does this seem like a blade issue or control board issue, or might there be another
cause for this behaviour

Appreciate any help

[ComputerGenie]

...Appreciate any help

Is the inside clean? Are both fans running? Is the PSU fully working?

[Bigsky]

I have asked the host to have a good look at it give it a blow out, which I
assume has been done, it is on the other side of the world from where i am.

Host thinks one blade was gone, until it came back, then says he thinks the
blade is on the way out, just thought this issue might have been encountered
already, its a bit strange

[ComputerGenie]

I have asked the host to have a good look at it give it a blow out, which I
assume has been done, it is on the other side of the world from where i am.

Host thinks one blade was gone, until it came back, then says he thinks the
blade is on the way out, just thought this issue might have been encountered
already, its a bit strange

And therein lies the problem with paying for a piece of equipment and surrendering total control of it to someone else. 

[Bigsky]

Yep, I agree, but it is either have a miner hosted or not mine at all.

Our electricity rates are over US18c a KWH, and summer temps are
over 40c.  I had to stop running S1s and S3s for a few months a year
back when I mined with them at home and my wife thought the noise
was bad then.

I bought the batch 2 before any reliability issues where reported with
the S9 units and had it delivered straight to the host..anyway, it has
nearly paid for itself, as I paid in fiat and just wondering what I am
going to do with it, whether keeping it running all over the place like
it currently does is doing it harm, obviously not worth sending it
back to China with the postage cost/time cost, maybe there is another
host that may be able to repair it and host it.

Thoughts welcomed

cheers

[NotFuzzyWarm]

Either have the host access it or remote in and drop its speed a bit. Bet it is happy again.

I have a b1, 2x b3's, and 2x b5's. Initially the B1 and 3's ran at OEM speed but a while back I had to drop their speed by 50MHz or so because 1 board was starting to drop out from time to time. The b5's are still running full speed with zero problems after all this time...

edit: Oh, if in North America, use Bitmain Warranty in CO. Way better and easier/cheaper to ship to.

[Gren]

I just got my latest batch of s9 in... 2 dead hashbaords in 24 hours, wew that was fast !

What kindof turnanround are you guys are seeing for mailing in the miner for warranty work?