Does anyone else have stability issues with linuxcoin

[jake262144]

Christ... each PCB-to-wire connection, each solder point decreases the SNR ratio. If your mobo starts with noisy pcie lines in the first place the SNR can only go down from there.
Should it drop below a threshold value, your reliability goes bye-bye.
Unless that "EMI shield" is properly grounded I'd call it more of an EMI-gathering antenna. If that shield doesn't look as if it's grounded at all try removing it from one of your failing risers.

A poor or overloaded PSU is perfectly capable of generating out of the spec electrical ripple and noise on its output lines. Combine that with a less-than-perfect mobo with a whole damn lot of risers sticking out of it and you've got your recipe for disaster.
Why do you think ripple and noise has been included into ATX spec?

EDIT::On the same subject of SNR, does anybody remember the transition from 40-wire IDE cables to 80-wire? Those additional 40 wires were never used for carrying the signal, they were just grounded. This very trick allowed for over twice higher transfer rates using the same interface.

[rjk]

Technically the extender is just a cable extender and it is dumb ( has no extra circuitry ) so should work as long as it is within the 19 cm or so spec limit.

Its not just length, there is much more to it. Just the fact you have a connector is a huge issue and greatly limits clockspeeds and can introduce all kind of problems. There is a reason soldered GDDR (vram) is so much faster than socketed DDR ram (and a reason you can not socket GDDR).

Quite so, and the ones that Bulanula showed actually have 4 connection points instead of the usual 2 with an extender. First and last points are the motherboard connector, and the video card connector. But in addition, the extenders he uses have 2 more connectors where the ribbon cable meets the PCB on each end.

However, if they fit together tightly, it *should* work fine. Mining uses very little bandwidth, and shouldn't be overly sensitive to latency either. That is why x1 connectors work fine for mining. (I use a few)

[bulanula]

Technically the extender is just a cable extender and it is dumb ( has no extra circuitry ) so should work as long as it is within the 19 cm or so spec limit.

Its not just length, there is much more to it. Just the fact you have a connector is a huge issue and greatly limits clockspeeds and can introduce all kind of problems. There is a reason soldered GDDR (vram) is so much faster than socketed DDR ram (and a reason you can not socket GDDR).

Maybe others using P67 UEFI Asus or similar motherboards can chime in ?

How did dishwara and likuidxd run 8 * 5870 on a Big Bang Marshall mobo ?

Just to say I have also tried Windows 7 64 and same thing happens ( black screen ) although I am not very familiar with Windows and tend to use Linux most of the time.

As I said before. The extenders work FINE with the older mobo but they refuse to work on the new Asus P8P67 WS Revolution with UEFI and P67 chipset so I am 90% sure the extenders are fine and it is a mobo issue or other type of issue.

In Linuxcoin the cards are all detected just fine but the Xserver refuses to load and I get black screen ( with monitor ) and CTRL-ALT-F1 does not work OR any Xserver app refuses to load ( using no monitor and SSH headless setup ).  

I have a MSI Big Bang Marshall and the same thing happens. I will get the Asus P8P67 WS Revolution this week or the next and try and see if the issue is solved but I doubt it.

I even tried standard Debian sid and wheezy and same issue with kernel 3.2.1 and driver 11.12 ( latest available ). Maybe I should start a different thread ?

[jake262144]

What I was trying to say with my previous post is, if you're experiencing stability issues, make sure you have a decent PSU installed, that is isn't being overloaded, and - what should be obvious - that your machine is properly grounded.
A computer is no more than an electronic machine which needs to be grounded to work as expected.
A non-grounded PC (connected just with hot and neutral wires) will boot up and might function well enough for casual use but is still is in violation of the spec. Trying to push such a machine to hard work is asking for trouble. This simple fact of life seems to be often forgotten.
Pay close attention to the foundations guys, ok?

Unfortunately, measuring electric noise at the PSU won't be an option in most cases but I'd expect serious miners to have an extra PSU in the "trash pile" somewhere. Make sure it's at least as strong as the one currently in use and try replacing the PSUs. If you've tried a crapload of extenders and all the wired ones fail, this might be your chance of restoring just enough magic for the whole rig to work.

[bulanula]

What I was trying to say with my previous post is, if you're experiencing stability issues, make sure you have a decent PSU installed, that is isn't being overloaded, and - what should be obvious - that your machine is properly grounded.
A computer is no more than an electronic machine which needs to be grounded to work as expected.
A non-grounded PC (connected just with hot and neutral wires) will boot up and might function well enough for casual use but is still is in violation of the spec. Trying to push such a machine to hard work is asking for trouble. This simple fact of life seems to be often forgotten.
Pay close attention to the foundations guys, ok?

Well AFAIK I am using a HQ Corsair AX1200 ( best PSU I think but very expensive ) which is grounded and only 4*5870 on that board and it still does not work so it is either software or mobo related not extender or power related ( I think ).

If the P8P67 WS Revolution comes and it does the same BS to me I swear I am going to jump out of the window LOL.

[dishwara]

I mined with 8 cards for only half day i think.
After that something happened which didn't allow to even start pc.
so i took mother board for repair & after that i got many personal problems (seller changed my mobo & gave me another BB, broken one & India is corrupted in everything)

I think may be its a over kill to a motherboard to run all the 8 pcix slots.
motherboards like BB are designed in mind to play games or graphics things which MUST not use 100% GPU all time.
mining may be overkill to them, gradually loosing ability to work, not sure.  

maybe trying with less cards will help.
maybe even a slot on mother board broken. have to check every slot individually to confirm they working correctly.

[jake262144]

Well AFAIK I am using a HQ Corsair AX1200...

A very decent PSU provided it's still in excellent condition. Good stuff... gold certified... the only thing it lacks is sugar frosting and a few specks of chocolate...  

[bulanula]

Well AFAIK I am using a HQ Corsair AX1200...

A very decent PSU provided it's still in excellent condition. Good stuff... gold certified... the only thing it lacks is sugar frosting and a few specks of chocolate...  

All equipment is 100% new bought by myself new June 2011 ( since then I have been trying to make this work ! )

I cannot get the following setup working on the BB Marshall ( and I suspect with the P8P67 WS Revolution as well ) :

2 cards plugged into the mobo directly

2 cards using x16-x16 extenders.

[dishwara]

4 cards work fine with BB marshall.
I am sure abt even 6 cards, coz i managed to mine with 6 cards for days, except i got powercut daily & a lot of heat.

[jake262144]

I cannot get the following setup working on the BB Marshall ( and I suspect with the P8P67 WS Revolution as well ) :

When using only the two extended cards, is the rig stable? What about three cards? Have you tried using those cheap pcie1x-pcie16x powered extenders?
Are your pcie16x-pcie16x risers externally powered?
If they aren't, you may be drawing more juice with those four GPUs than the board is able to supply.
If the rig works fine with less cards but fails at four, make sure you use powered extenders.

Retuening to PSUs, do you have another PSU you could try?
The AX1200 is great but stuff happens. A few failed capacitors may result in elevated ripple levels.
The only problem is, you'd need a local electronics guy with an oscilloscope to be able to say for certain whether or not the PSU is to blame.
Much easier to just use another PSU.

[dishwara]

i used cooler master SPG 1200W , still it refused to power 4 cards.
i can only able to power 3 cards with a single CMSPG1200W, but with just extenders, never used externally powered extenders.

[bulanula]

I cannot get the following setup working on the BB Marshall ( and I suspect with the P8P67 WS Revolution as well ) :

When using only the two extended cards, is the rig stable? What about three cards? Have you tried using those cheap pcie1x-pcie16x powered extenders?
Are your pcie16x-pcie16x risers externally powered?
If they aren't, you may be drawing more juice with those four GPUs than the board is able to supply.
If the rig works fine with less cards but fails at four, make sure you use powered extenders.

Retuening to PSUs, do you have another PSU you could try?
The AX1200 is great but stuff happens. A few failed capacitors may result in elevated ripple levels.
The only problem is, you'd need a local electronics guy with an oscilloscope to be able to say for certain whether or not the PSU is to blame.
Much easier to just use another PSU.

The rig does not work even with just one 5870 connected with an extender. I just get black screen at Linuxcoin.

If that same 5870 is used but in the slot it works fine.

If 4*5870s are used directly to the mobo it works perfect.

If 3*5870 are direct on the mobo and just one 5870 uses an extender Linuxcoin fails to start.

Basically whenever I use just one extender on the BB Marshall it makes the driver somehow crash so none of the other cards also work.

Tried with Linux, Debian, Windows etc. BAMT and same problem everywhere. I got multiple AX1200s and BB Marshalls so it really is strange.

BIOS works fine all the time with extenders. Open source radeon driver works 100% as well and all the cards are detected in lspci and aticonfig --list-adapters.

Xserver does not work and I get black screen all the time even with just one card and one extender and even with SSH any X application does not launch at all.

My personal belief is that there is some extra latency and the aticonfig / fglrx driver module does not properly initialize when using an extender but I am not 100% sure.

I am pretty sure it is not a power problem because even powered extenders do not work. It is either a software ( fglrx issue / bug ) or a mobo issue I think.

I can post detailed dmesg and other logs if needed.

Thank you for trying to help !

[jake262144]

The solution is:

Code:

use powered extenders || use some other powered extenders || try another PSU || defenestrate

I sincerely hope you won't get to the defenestration part

[dishwara]

It took more than 2 days for me to figure out that slot 3 & slot 7 NEED 16x extenders, while other slots work fine with 1x extenders.
I used extenders for all cards & no gpu is directly connected to mobo.
Try extenders for all.

[bulanula]

It took more than 2 days for me to figure out that slot 3 & slot 7 NEED 16x extenders, while other slots work fine with 1x extenders.
I used extenders for all cards & no gpu is directly connected to mobo.
Try extenders for all.

These are the extenders I have :

http://www.delock.com/produkte/gruppen/zubehoer/Delock_Riser_card_PCI_Express_x16_with_flexible_cable_left_insertion_89130.html

Using just one 5870 and this extender and Linuxcoin still fails and even with plain Debian the driver fglrx does not properly function. 

[dishwara]

check the slot switches in BB.
it has hardware switch to disable pcie slot 1,2,3,4 on top left.

EDIT:
also make sure OC GENIE & MUTLI BIOS switches DIDN'T pressed.

[bulanula]

check the slot switches in BB.
it has hardware switch to disable pcie slot 1,2,3,4 on top left.

Yeah. I had them all set to on already.

[jake262144]

These are the extenders I have ...

Can't you get a couple cheap ones from Cablesaurus? Should the cheapest pcie1x-pcie16x (powered!) ones do the job, you're gonna laugh your ass off.
That might be the cheapest solution I can come up with.
Each day that rig of yours is inoperable you're losing profit.
I'm going out of my way to help you out here but remotely, only that much can be done.

[dishwara]

It took more than 2 days for me to figure out that slot 3 & slot 7 NEED 16x extenders, while other slots work fine with 1x extenders.
I used extenders for all cards & no gpu is directly connected to mobo.
Try extenders for all.

These are the extenders I have :

http://www.delock.com/produkte/gruppen/zubehoer/Delock_Riser_card_PCI_Express_x16_with_flexible_cable_left_insertion_89130.html

Using just one 5870 and this extender and Linuxcoin still fails and even with plain Debian the driver fglrx does not properly function.  

If even with one extender with card fail means , it must be faulty extender or not properly getting connection.
u sure, the fans on card works when u use extenders?
coz u using a left side extender, so instead of standing straight the card will be rotated/tilted 90 degrees & fan faces down & may be not rotating.
i used straight extenders from cablesaurus.
 

[P4man]

These are the extenders I have :

http://www.delock.com/produkte/gruppen/zubehoer/Delock_Riser_card_PCI_Express_x16_with_flexible_cable_left_insertion_89130.html

As RJK mentioned; those are the worst, because of the design. Its 2 slots and 2 sockets.
Without extender, so plugging a card straight in to the motherboard, you have a single non-soldered socket. With a regular extender cable you have 2. With your clumsy extender there are 4 per device!