Topic: Antminer S5 high error rate at stock speed (Read 3919 times)

Sounds about right. When one of my laptops drops off the internet, I can submit shares quite a bit later and still come through as valid.

I am using GPRS mobile connection with 300-800ms delays without any trouble. Almost no stale shares. Using high diff to keep GPRS 32kbits connection alive.

Latency won't have an impact on hardware errors, but will do so on stale shares.

What's the latency to the pool? Ping comes back high 100's may want to switch.

I had two of these psu's http://www.fractal-design.com/home/product/power-supplies/newton/newton-r3-1000w-white


ran s-1's on them.  had a problem like the op. and after 35 days they failed.  the psu's claim to do 1000 watt.  they were doing 800 watts on two over clocked s-1's .

 turns out that was a hair too much and slowly but surely the two psu's failed.  if he went from .0002 to .0075 to .0138 in 4 or 5 days. it bears watching

0.013% is very low error rate. (granted, its higher than the 0.0008-0.003% usually seen in the S5)

as suggested by phillipma, if the error rate doesnt increase over time theres really no concern. 0.013% is pretty insignificant, its a difference of pennies/month at the most

check your psu very carefully .  this one

 http://www.newegg.com/Product/Product.aspx?Item=N82E16817438027&cm_re=evga_supernova_750-_-17-438-027-_-Product

has separate rails  if you plug it in wrong you could be straining it.

this one has 1 rail

http://www.newegg.com/Product/Product.aspx?Item=N82E16817438017&cm_re=evga_supernova_750-_-17-438-017-_-Product

but the rail rates for 744.8 watts.  if you run the s-5 at stock it takes about 610 watts.  which is 80% of this psu capacity.  

I think your error increase is due to this psu.  it is just about maxed for 24/7/365 use.  

humor me and set freq at 337.5  or 343.75m    ..

run for a few days at 343.75   if errors stay flat lined then it is the better setting for the psu and the miner.

 a review for your psu

http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story&reid=380

if you look at cold test 600 watts it is at 12.06 volts
if you look at cold test 748 watts it is at 12.03 volts

if you look at hot test 600 watts it is at 12.05 volts  his hot test is a few hours and there is some sag.
and his hot test at 748 watts       is at 12.02 volts

my guess is you are very close to max on the psu.   and after days of running your psu is a bit too tired.

 it sagged from 12.06 to 12.05 in 2 or 3 hour hot test at 600 watts
it sagged from 12.03 to 12.02 in a 2 or 3 hours hot test at 748 watts
 
rollback on the freq  343  or 337  and test for hw's

or keep it at freq and see if the hw's stop increasing. (they will if the psu is just past it limits for 24/7/365)
I have seen this happen and take 20 or 30 or 40 days to burn the psu.

I'm using an EVGA SuperNova 750W G2 so I don't think it's the psu.
I'm at 0.0133%/0.0138% error rate currently with 49C/51C temperatures.

well watch for the error rate and see if it continues to drift upward.

The numbers are good.  BTW what is the exact make and model of your psu.  maybe it is dropping the dc under 12 volts with continued power on.

 LOWER  quality or over taxed psu's can slowly drop volts under 24/7/365 use.

I wasn't worried about HW errors as in lost revenue, I was worried that the hardware itself might have a problem and/or break after the 90 day warranty.

yep since it is 99.99 correct.  so for every 100 usd it earns  that .01 costs 1 penny.

So many people complaining about hardware errors. How about we word it like this? A S5 running at 1300gh and 0.01% errors equals about 0.00004195 btc per month or about a 1 cent loss.

just one more note - HW errors can be highly influenced by supply voltage. Use voltmeter and measure voltage supplied to the miner directly on miner's PCB PCI-e connectors. Should be 12V, if less --> more HW errors.

Your miner is working perfectly. 700HW during 10hours is absolutely perfect. It is hard to reach such perfect work on C1 or S3.

It is dis/advantage of S5 design - if any of ASIC goes over treshold of HW errors it makes power consumption variation in chain and excessive voltage on one ASIC drops down whole chain, so advantage is that all ASIC must be working perfectly, disadvantage is that they operate at much higher voltage compared to C1 style design, so chip efficiency is worser. Overall efficiency might be better due to missing DC/DC losses.

which is what I told him.  he just needs to see if he gets a continued drift higher.  0.0075% is nothing.



he is failing to see that his correct number is 99.99%   rounding off the last 2 numbers.

Just because mine is  99.9999%  and his was 99.9998% he is nervous.

if it drifts from 99.9998 to 99.9925 to 99.9 then to 99 he can worry.

FTFY

There is absolutely no problem with these numbers, its not indicative of a problem. For some context, a year ago we would have been happy with <1% error rates.

That's what I mean philipma1957, I have lower temps than your miner and way more errors.
You have 30 errors in ~48 hours, I have 900 in ~12 hours.
I just don't want the miner to crap out completely after the 90 day warranty period (had it for about a week).

you are fine.

you are 99.9925 % good and 0.0075 % bad


granted 99.9998 % good and 0.0002% bad is a better set of numbers.

you keep an eye out to see if it drops to 99.9000 %  good  0.1000 % bad


if it keeps shifting worse something is wrong.

my miner is doing better then yours

In the first picture it says HW 718 errors and 0.0075% for Diff1# and 0.0075% for DiffA# so those are the error percentages correct?
From what I've read/seen on these forums people run the S5 overclocked and have fewer errors.
Also, it didn't have as many errors the first 3-4 days, less than 0.0002%.