Bitcoin Forum
May 04, 2024, 12:48:04 PM *
News: Latest Bitcoin Core release: 27.0 [Torrent]
 
   Home   Help Search Login Register More  
Pages: [1]
  Print  
Author Topic: Stability Issues (270X, h61 brc pro, 2 PSU,..)  (Read 6786 times)
dani (OP)
Hero Member
*****
Offline Offline

Activity: 525
Merit: 500


..yeah


View Profile
February 08, 2014, 05:16:32 PM
Last edit: February 09, 2014, 10:45:38 AM by dani
 #1

Hi folks,

I found many threads and subreddits where people came across my kind of errors, but none of the fixes applied so far to me. Thats where I come to ask the pros  Grin

Hardware:

Litecoin BAMT 1.5 (right now running with vertminer instead of cgminer, didnt change my problems..)
Asrock H61 BTC PRO
G1610 Celeron dual core
corsair 4GB cms3
6x 270X Powercolor PCS+
2x PSU Coolermaster B700
AData 16GB USB
LAN connection

GPU connections:

I have 1 GPU 16x-16x risered (non powered) and 4 GPUs with 1x-16x powered risers (although "not needed as advertised"  Angry )

Power connections:

5 cards:
PSU1: mobo (24pin, 8pin, 2x 4pin molex), gpu 1+2, + all powered risers.
PSU2: gpu 2+3+4

Made a ground connection to be safe. The 2nd PSU poweres on via a pin shortage (green and black, PSU_ON and GROUND). Works perfectly so far.

awkward:
When not using powered risers (but connecting the two molex pins on the mobo) all the cards with 1x risers won't be recognized by the OS/BIOS. When turning the machine on the one working card will spin its fan, stop the fan, spin it up again and boot. It will hash. But the other fans just sit spinning from the beginning, not bein recognized.
When connecting the powered risers to the PSU they will behave like card 1, get recognized and start hashing.


My problem:

Not running stable. SGminer or Cgminer (both) freeze within hours, sometimes minutes. The OS is still responsive, but cgminer freezes, I can log in and watch the screen etc. When lookin in htop ("taskmanager") one of two CPU cores is running 100%. I cannot kill the process (even sudo kill -9 cgminer won't work) and cannot reboot. I need to hardreset at this point (psu). Best I made was a 10hours run.

What I tried:

disconnection card after card.
I had a double riser (riser into riser) and eliminated that error, can't tell the results yet.
connected each card with PCI-E AND riser to their PSU, no difference so far. I found on another rig that when powering each cards riser with its PSU it won't run (restart when starting to hash).
I disconnected the onboard molex - no changes.
All the hardware is brand new. Don't have any temperature issues - it's cool (mostly <75°C no matter how high I overclock, usually running <70 °C).
Cgminer won't report any errors - it just freezes. (on my 2nd rig I sometimes get a error with 511°C and 0 RPM on one card, but that's another problem.. for another time)

hashrate

My hashrate (scrypt, cgminer) has been lower than other people report. I tried many settings where people reported 470khash+ (up to 500). The best I ended up on was 435 (not stable). Here my last settings as I remember:

Quote
{
"pools" : [
        {
                "url" : "stratum+tcp://eu.multipool.us:3352",
                "user" : "ddcoin.1",
                "pass" : "x"
        },
        {
                "url" : "stratum+tcp://stratum.dogehouse.org:3333",
                "user" : "coinedd.1",
                "pass" : "x"
        }
],

"scrypt" : true,
"worksize" : "256",
"intensity" : "18",
"gpu-fan" : "60",
"thread-concurrency" : "16000",
"gpu-engine" : "1000",
"gpu-memclock" : "1425",
"temp-cutoff" : "90,90,90,90",
"temp-overheat" : "85,85,85,85",
"gpu-powertune" : "2",

"failover-only" : true,
"gpu-threads" : "1",
"temp-hysteresis" : "1",
"kernel-path" : "/usr/local/bin"
}

I should also note that my vertmining hashrate is lower than others get (185 vs 220). I can't find what is wrong, but I know other people are having the same issues. Gather, unlucky miners!  Cheesy

On a serious note: this has cost me too much money on stupid risers, fans, cable extensions, psus.. and a lot, a lot of time. Please give a miner some advice! This is not the first rig I've built, but the most fucked-up one, yet. Huh

This thread will soon look like a log because I'll try anything. This way others can see what I tried and what the effects were. Maybe someone can find a clue to why he has this error - and maybe, maybe I'll find the error myself.

Hai
1714826884
Hero Member
*
Offline Offline

Posts: 1714826884

View Profile Personal Message (Offline)

Ignore
1714826884
Reply with quote  #2

1714826884
Report to moderator
1714826884
Hero Member
*
Offline Offline

Posts: 1714826884

View Profile Personal Message (Offline)

Ignore
1714826884
Reply with quote  #2

1714826884
Report to moderator
"This isn't the kind of software where we can leave so many unresolved bugs that we need a tracker for them." -- Satoshi
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction.
Gator-hex
Hero Member
*****
Offline Offline

Activity: 490
Merit: 500


View Profile
February 08, 2014, 06:12:21 PM
Last edit: February 08, 2014, 06:24:41 PM by Gator-hex
 #2

Quote
6x 270X Powercolor PCS+
75+75+75 = 225W
6x = 1350W

is probably too much for

Quote
2x PSU Coolermaster B700
660W +12v single rail
2x = 1320W

+ 100W for you CPU/MB/HDD etc. (150W if AMD)

= about 130W short.

Make sure you never share a PSU on a GPU when using multiple PSU.

It could also just be your intensity/concurrence is just too high. Google how to tune scrypt.

Every card can be different sometimes the silicon is just not the same and cannot handle the same overclock and you have to tune each individually.

dani (OP)
Hero Member
*****
Offline Offline

Activity: 525
Merit: 500


..yeah


View Profile
February 08, 2014, 06:34:42 PM
Last edit: February 08, 2014, 10:26:15 PM by dani
 #3

Hi Gator-hex

thanks for your answer. I know that these cards run around 150-180 (max!), so I assume 180*6 = 1080. I'm only running 5 right now (I'm short on pci-e adapter cables). So 5x180 = 900. I don't have a kill-a-watt around, I'll updates once I have one around.

I think 100W is a little high, but regardless I'm off the limit. Last setup was 3 cards (2 on PSU1, 1 on PSU2). so 360W + 100W board = 460 load on PSU1. Should have been cool, still crashed.

I used settings other confirmed stable, I tried lower TC, Intensity and clocks. Though you might have a point on the OC, maybe some (or all) of the cards just don't want to be pushed. I'll have it underclock this night and we'll see.

UPDATE:

since I've reduced my double-riser-connection to only 1 riser it's running stable (around 4-5 hours now), even on higher settings. So far so good  Smiley
Added the 4th card, had a crash within 10 minutes. Saw the riser had a lot of tension, reduced that.. if this fails, I'll change riser cables. Maybe this one is bad

Hai
dani (OP)
Hero Member
*****
Offline Offline

Activity: 525
Merit: 500


..yeah


View Profile
February 09, 2014, 10:02:19 AM
Last edit: February 10, 2014, 11:54:39 AM by dani
 #4

went back to 3 cards, unstable again.  Trying now to power 3 cards off the same PSU.

UPDATE1:
3rd card says "SICK" in cgminer. Changing riser cables now.

UPDATE2:
Didn't help, still sick. Plugged the sick card back into the 2nd PSU, working again.

UPDATE3:
Didn't help, freezes again.
I noticed card 3 always starts with 1 HW. Not getting anymore, but this is strange. Also cgminer freezes around the same time, when each card had around ~1000 accepted shares.
Switched the pci-e slot to the one next to it.

UPDATE4:
working so far. 1 HW error is gone now.

UPDATE5:
now the 2nd GPU got "SICK". Screw this, what's wrong here?!
running stock now.

UPDATE6:

at least I found out something: If using poweres risers (the once that have an extra yellow wire, like this: http://www.cryptocables.com/wp-content/uploads/2013/04/16.jpg ) won't work unpowered while a 1x - 16x (non powered) with the wire jumper (as seen here http://www.gobitgo.com/articles/1001/How-To-Correctly-Use-and-Install-PCI-E-Riser-Cables/) will work right away. Using the powered riser with this wire jumper won't work. I think about cutting that yellow extra wire and using the wire-jumper-mod to get those risers running as well - because one fucking risers is around 17-18€ around here. Fucking ebay..

UPDATE7:

I guess I found it!
So I found that when using my kind of powered riser without the power connection, it cannot be powered by the mainboard, because all the 12V lanes are cut and soldered to the molex connector. By using a normal 1x-16x pci-e riser (without wire jumper) it will hash (with the 2 molex connectors to the mainboard). It hashed since yesterday afternoon, through the night and looks rock solid. Fuck me! I'll get myself more 1x16x risers (fuck the chinese and especially ebay!)  Cry

Edit: still stable, almost 24hours now.

Hai
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!