Topic: Been reading about a lot of dying sp20's - page 2. (Read 3477 times)

I have 35 SP20 and all running around 1350-1400 mhash. Not a single one has died and has been running flawless since January.
The key thing with SP20 is to run them downclocked and COLD. Mine is running in ambient temp of 16 degrees celcius all are running with 2800-3000 watt ibm server psu and breakout board. Rebooting them every 16-18 days.
If I could I would buy more..

running 4 underclocked and undervolted (0.61V, 1TH/s) and Fan speed at 2. As of today all 4 are doing fine.
 Updated the firmware to 2.7.1

yeah I never was able to go over 1650 on any of my 13, but they suck up a ton of power at that speed.

Running 6 here all flawlessly for months.
3 bought direct and 3 bought from here all running at around 1.3TH/s

As for them not running at 1.7TH/s when i first got them and just because I wound everything up to 11 i got 1.6, 1.68 and 1.77.

They are far happier at 1.3TH/s and so are my power bills.

I'd like more but we'll more chance of pigs flying..

I have wondered this before given the reports (mine included) of a SP20 having a messed up controller requiring a SD card after a power outage.  When the power goes out the fans stop.  When the fans stop the still hot ASICs/heatsinks dissipate their heat upward-- and that is where the controller board is at.

I've been tempted to rig up a fan on a relay and UPS to start up when the power goes out just to provide some airflow through/across the miners for 20 minutes or so after an outage.  Alternatively (and simpler), maybe just run them upside down so the controller is underneath the ASICs.  I'm not sure if that would cause other problems though.

So far 235 sp20's mentioned.  lots of good info. thanks to all

13     13 - 0     Philipma1957
50     50 - 0     shacky
27     22 - 5     Pt0x   -----------    but the 5 do recover with sd cards
  8      6 - 2      Xian01 ----------   a controller and a pcie issue fixed by owner
  3       1 - 2     p3yot33at3r ----- dead controllers
  2       1 - 1     r00t$ --------------    needs an sdcard and 2 bad asics
  9       9 - 0     Prelude ------------ all good has run them at 1.1th
  5       5 - 0     sjc1490 ------------ all good  run at 1.1 to 1.3th
 17     17 - 0    Tupsu --------------- all good
 17     17 - 0    Finsky --------------- all good some ip issues ran them hot 1300 to 1500gh
 58     56 - 2    MCHouston --------- 2 with dead asic issues.
 17     14 - 3    sidehack ------------- 3 with shipping damage
 1         1-  0    alh ------------------- 1 good with a slightly hot asic chip
 4         4 - 0    GenTarkin ----------- 4 good run 1250 to 1400
 4         4 - 0    tlhIlwI---------------- all good but the eBay purchase runs hot
 1         1 - 0    CrazyGuy ------------ 1 chip runs hot but works
 3         2 - 1    Biodom --------------- 1 all good 1 runs hot 1 came doa easy fix
12       10 - 2   Quakefiend ---------- 2 dead controllers
 1         1 - 0    edonkey -------------- all good runs a bit hot

so of the 252 looks like 234 still run  and 18 needed parts from sp-tech or run a bit less then then should


thanks for info so far.

I know it is hard to not complain about the rma or company not doing the right repairs.  

Thanks for not pushing too hard in that direction.

I am hoping for  a few more replies today.

Thanks again Phil

 
Totals

the running rate with little or no issues is over 92.85%  

the remaining 7.15%   run somewhat after repairs.

I will leave thread open longer for more replies.

Thanks for not pushing the anger on the imperfections.

Best to all phil

If you were to project this to all gear (a big stretch ) 

the network  of 350ph has lost about 17ph in the last 7 months to gear dying.

I will leave the thread open until monday and then do one on s-5's

I know this is only an approximate number,but one can see why network growth lags since last Dec.
  If your gear dies and you were doing marginal profits a buyer of gear may not replace it.
We all know a builder of gear that self mines will find it easy to replace it compared to a buyer of gear.
  This has really shown me something   using a 5% death rate of gear gives us 17.5ph of gear shut down . using a 10% death rate 35ph of gear has shut down.

Just had a random thought. If it interferes too much with your data gathering feel free to delete.

I am wondering if the controller on the SP20 were to run warm for some reason (e.g. max speed, poor intake air temp, whatever), could that shorten the life of the installed flash on the controller? I am reluctant to tinker with my SP20, but I think it's basically a Beaglebone Black (BB) with an extra FPGA to connect it to the hashing boards. I think the BB has a builtin in EMCC(?) flash that it normally runs from. Lets say it was running for weeks at an elevated temperature, could some of the flash cells lose their state and hence end up with a corrupted flash (i.e it's bricked)? The external SD card is almost outside, and likely wouldn't get as hot, besides it's relatively easy to just swap in a new SD card if that gets corrupted.

Is it true that you can't reflash the native EMCC from the SD card? That seems like an unfortunate design situation. Would it "help" to periodically reflash the EMCC while it's still sane, in order to "refresh" the cells?

Of course this whole line of thinking might be completely off. If Spondoolies looked at this, they might also have some good ideas, and learn something from our experience. I doubt they would "fix" anything, but might have ideas to offer. They might also learn something for their generation of hardware (i.e. make sure you have an easily replaced flash).

Continuing to keep my fingers crossed and my SP20 "cool".

I've got one SP20e that's been mining for 211 days. It's been very reliable.

Loop 3 does run hot, so the hash rate fluctuates a bit. I've always run it in an air-conditioned environment, so it's fed with cool air. That seems to help with the overall hash rate. It's performance is very tightly tied to the ambient temperature. More so than other miners that I've had (like the S3's).

There have been a few times where the unit went off line, stopped hashing, and was unreachable (even via ssh). Power cycling has always brought it back. Fortunately I have it at a data center with a PDU, so remote power cycling is easy.

good to hear you can supply a few parts.

I will tally this up again later today.

Have had 12 units total, 2 controllers died, no other issues.

BTW, to anyone reading this thread that has dead hashing boards, I have 3 good ones left.  PM me if you need.

I had three.
One is still running (since mid December). I run it at 1.2Th with low fan. One loop has one chip at 85C, but another at 115C. That high temp chip seems to get progressively hotter over the months (it was 105-110 initially, then went to 120, so I had to decrease power to that loop). A good machine, nevertheless.
One machine I run a little, then sold-it was good as well.
One SP20 had one nonfunctional loop from the get-go. Spondoolies was very nice to fully compensate. I sold it here, and it turned out that one connector did not properly "bite" the cable, so it was a relatively easy fix by someone who was ready to void the warranty.

I would score it as two/three with no problems and one with a fixable problem

I've got one left. Other than one die that runs pretty hot (~120c) and shuts off every couple weeks, it's still running strong.

going to delete as off topic.   but in a while when I try to do the s-5 I am sure we will hear that problem over and over again.

once I get a decent amount of numbers  on the sp20's   (so  far we have 235 in)

I will do the s-5's next.

So far 235 sp20's mentioned.  lots of good info. thanks to all

13     13 - 0     Philipma1957
50     50 - 0     shacky
27     22 - 5     Pt0x   -----------    but the 5 do recover with sd cards
  8      6 - 2      Xian01 ----------   a controller and a pcie issue fixed by owner
  3       1 - 2     p3yot33at3r ----- dead controllers
  2       1 - 1     r00t$ --------------    needs an sdcard and 2 bad asics
  9       9 - 0     Prelude ------------ all good has run them at 1.1th
  5       5 - 0     sjc1490 ------------ all good  run at 1.1 to 1.3th
 17     17 - 0    Tupsu --------------- all good
 17     17 - 0    Finsky --------------- all good some ip issues ran them hot 1300 to 1500gh
 58     56 - 2    MCHouston --------- 2 with dead asic issues.
 17     14 - 3    sidehack ------------- 3 with shipping damage
 1         1-  0    alh ------------------- 1 good with a slightly hot asic chip
 4         4 - 0    GenTarkin ----------- 4 good run 1250 to 1400
 4         4 - 0    tlhIlwI---------------- all good but the eBay purchase runs hot
 1         1 - 0    CrazyGuy ------------ 1 chip runs hot but works
 3         2 - 1    Biodom --------------- 1 all good 1 runs hot 1 came doa easy fix
12       10 - 2   Quakefiend ---------- 2 dead controllers
 1         1 - 0    edonkey -------------- all good runs a bit hot

so of the 252 looks like 234 still run  and 18 needed parts from sp-tech or run a bit less then then should


thanks for info so far.

I know it is hard to not complain about the rma or company not doing the right repairs.  

Thanks for not pushing too hard in that direction.

I am hoping for  a few more replies today.

Thanks again Phil

 
Totals

the running rate with little or no issues is over 92.85%  

the remaining 7.15%   run somewhat after repairs.

I will leave thread open longer for more replies.

Thanks for not pushing the anger on the imperfections.

Best to all phil

If you were to project this to all gear (a big stretch ) 

the network  of 350ph has lost about 17ph in the last 7 months to gear dying.

I will leave the thread open until monday and then do one on s-5's

I know this is only an approximate number,but one can see why network growth lags since last Dec.
  If your gear dies and you were doing marginal profits a buyer of gear may not replace it.
We all know a builder of gear that self mines will find it easy to replace it compared to a buyer of gear.

I have 4 SP20s.  Two of them have metal shields on the serial connection and two just have black plastic connectors there.  In my testing the ones with the silver serial connections run slower but undervolt better, whereas the ones with the black connectors run much faster but don't undervolt as well.  I mention this because I suspect those connectors might correlate to production batches.

Anyway, of my SP20s:

Two of them (both with black connectors) run perfectly and have been at around 1.40T-1.50T (0.72V) their entire life with an occasional bump to 1.60T (0.75V) when WestHash was paying ridiculous amounts.

One of them (with metal connector) struggles to reach 1.37T and runs really hot at the ASICs (7 of 8 hitting 120C with 29C intake).  The exhaust temp on this one never exceeds 60C and feels relatively cool compared to the others, so I suspect a problem with the thermal compound between the ASIC and heatsinks (the heat isn't getting transferred into the air)-- possibly the compound is drying out.  I never set this one past 0.72V and am more comfortable with it at 0.69V max .  I'm tempted to pull the heatsinks and change the compound, but am a bit concerned about cracking the ASICs if the compound has already hardened too much.  It's bad enough that a couple of the ASICs are thermal limiting to around 0.60V.

The last one (also with the metal connector) bricked after a power outage.  Recovery SD didn't fix it, but it will run with a boot SD inserted (and that is how it now runs).  It has been run at 1.35T-1.45T (0.72V) with the occasional bump to 1.55T (0.75V) during the WestHash runs.

I run all of mine with the fans set to 100%.  They are in a location where noise isn't a problem, so I prefer the extra airflow over quiet operation.

Edit: All of mine except for the one with thermal compound issues were new purchases back in January (I know their history).  The one with thermal issues was an eBay purchase, so I have no idea the history (probably ran hard before I got it).

I will tally numbers in a little bit.

So far looks to be better then 90% good.

The ones that that I own and failed to boot, did not started after several tries with the permanent boot sd card inserted. I can tell you that I tried to boot a least 5 times per miner with different time intervals between tries. But once I got them to boot at least once, they started to work again as expected.

Every time I had a power outage, I could loose as much of 2 or 3 days of mining until I used the SD cards. Now they boot immediately after the power comes back or in the second try.

Now I keep some SD cards ready all the time waiting for the next board with dead flash.

4x SP20e been mining on them since jan / feb without issue.

In recent months all 4 of them run towards 120C at chip temps on loop 2 & 4, VRM's around 100C ... they are tanking the heat like warriors.
I did up the fan to keep them as cool as possible. Throughout the winter, fan was at 10% and clocked much higher than they are now. I think they were running close to 120C in winter as well w/o issue. They declock / devolt as needed to stay under 120C

Run them anywhere between 1250-1400GH

My single SP20 was purchased used in February of 2015. For the most part it's been run underclocked in the range of 1000-1200 GH/s, with a few hour burst run when it was colder. The max has been about 1400 GH, since I only have a 1000W power supply, and only let it go to about 950W from the supply.

I try not to reboot it often, and sometimes when it's a power-cycle type of reboot, it seems to go through 1 or more additional internal reboots to get itself all squared up. I use my router to assign a static IP address. It's connected to my WiFi network via a Netgear WiFi-->Ethernet bridge that's powered off the SP20 USB port. I haven't tried to make the SP20 run a WiFi dongle in the USB port as others have.

I's been essentially trouble-free, with one hot ASIC, that's usually about 20C hotter than the others. That ASIC is on a loop that is usually 30+ Watts over the limit I specify. The other 3 loops run as specified, and seem cooler.

I cross my fingers whenever I have to reboot it.

Sacrificing complexity (and in my opinion, heat density) would sure be nice, as that tends to decrease costs and increase reliability. 200A four-phase VRMs per 120C-rated chip and an FPGA-implemented custom comm protocol into a custom controller board isn't doing the trick.

I've had 17 running on my hosting shelves for several months now with occasional issues. Three of one guy's six miners were damaged out of the box, including cracked inductors on the VRMs, cut ribbon cables and a smashed LED. One of the damaged units has a pair of ASICs that have never worked since being here, and the unit will not hash at all if that loop isn't manually disabled.