Topic: 7x 7970's, PCI-E 16x-connected cards not working (Read 12450 times)

Had the same prob on a new GA-Z87x-OC board > mining on 2 cards worked fine > after adding 3de or 4th card WinOs did not find the card and the board went unstable during boot.
After setting the pcie or peg Gen config from auto to Gen1 in the bios > all went stable and solved the problem for me. 

Try increasing your PCIE latency from 64 to 96 in the bios. Worked for me when I added my last 7970 when nothing else would.

If you want to convert a 16x riser into a 1x riser, yes you could convert it by cutting wires. Also make sure your motherboard plays nice.. some boards require pin shorting to be able to use a 1x riser in certain slots (i had to sometimes short presence detect on 1x slots and 16x slots of various boards)

huge migraine right now will try to help later

Don't remember seeing this option in the BIOS, but will check again! Thanks for the heads-up!

I did attempt to set everything up in Linux initially (Ubuntu 12.04) but I remember having problems getting it set up properly with the drivers and having the cards show up as being available for mining. I suppose I could sink some time into it and use Ubuntu 10.10 like I used to. Hm.

Also, if I set everything up with Linux, I doubt I could underclock the RAM as much as I do in Windows (thanks to a "bug" in the beta versions of Afterburner). I guess I could BIOS edit since the 7970's don't do hash checking like the 6XXX series and below did (not a lot of people know this.. including the guy that makes RBE, sadly), but thats very uncharted territory at this point, and not many people have attempted it. Also, I've set the SATA type to AHCI in the BIOS as opposed to ACPI.. is that fine? Or should I configure things under IDE?

Did that. Disabled everything but USB, then disabled USB and gave that a try. Didn't help.

Seems reasonable. When all cards are plugged in via risers, I noticed that the two cards that weren't available for mining were (incidentally?) the two 16x connected cards. Could I just cut the part of the cable that makes the whole thing a 16x cable so that there are only enough connected wires for it to function as a 1x to 16x, and use the cable as a 1x to 16x cable? I'm willing to sacrifice a cable to make this all work! For science!

The risers aren't defective; I unplugged two 1x to 16x cables from two other cards, and after a reboot all of the remaining cards were recognized and functional (including the 16x cards, which were not functional before).

But I'm not underclocking/undervolting, quite the opposite, I'm majorly overclocking and majorly overvolting (1231mV, 1270MHz core).

Way ahead of you; I'm using this bad-boy:

Sounds like this is a dedicated mining rig. Go in your BIOS and disable EVERY non-essential piece of hardware that you wont be using. Audio. Firewire. All but 1 NIC. Serial ports. Parallel Ports. Serial ATA Ports/controllers except 1 if using a hard drive. All but 1 CPU core if using a multi-core CPU. USB ports, unless you need a keyboard to get in for something. All of that is consuming "resources" I'm assuming that can be used for the hardware you actually want to use, i.e. video cards. Also, using 16x > 16x risers may be causing a problem, because your board is trying to negotiate to 16x speeds. Some boards, when negotiating to max slot speed, will turn off other slots entirely in order to have the full amount of bandwidth available (PCI-E lanes are not unlimited). Try using 1x risers across the whole shebang. If you have limited 1x risers, use the 16x risers in your PCI > PCI-E adapters since they are always 1x speed.

Also make sure your risers are not defective. I had a bunch of 1x/16x > 16x risers, and they all went bad because of the stress of plugging them into the cards or bending the cables, would either cause cards to freeze the system when a load was applied, or device manager would show errors, or the cards wouldn't even detect at all. I use 1x > 1x risers with the edge sanded off in order to accept anything bigger than 1x, and they have all worked no problem (have 10 in use now), and are effortless to plug a card in.

Edit: I didn't really read much of the article before. You say 16x > 16x doesnt work, but 1x > 16x does. It could be a bad riser or bad speed negotiating. Use the 16x> 16x risers in your PCI > PCI-E adapters since they always work at 1x. Powered shouldn't matter if you are doing underclock/volt. Also, I already have experience with 6 cards on 1 board as well as using PCI > PCI-E adapters (I'm using one in a 4 PCI-E slot board now, now have 5 slots).

Oh, almost forgot. You say you are using 2 PSU's. If you can, power on the PSU that ISN'T powering the motherboard FIRST, then the one that does SECOND. I mean, kind of worth a shot I guess. I've forgotten to plug in PCI-E power cords to a card before and they wont detect, or they malfunction. If the card doesn't have proper power before the motherboard tries to detect it, that might be a possible problem too.

I would have to disagree with this.  I have 5 rigs with 8 gpus running 24/7 for about 8 months now. very stable.

as for the original topic.  i have 3 rigs with 5x7970's.  I tried to put a 6th card in one and could NOT get it to work.

 I could see all 6 cards in list of devices, but only 5 would be enabled at a time.


I hope this does not stop me from getting 4x7990s working... cant wait for that.

Do yourself a favour and configure it on a linux machine.

Use kernel parameters to disable ACPI and self-assign IRQ's to each GPU (ACPI can stuffup horribly for some reason with so many GPU's)

Ive had 8 GPU setup like this temporarily but it really puts alot of strain on the board and I could literally feel the slots heatup with all 8 running thus I dont really support this for a stable longterm setup.

That's a bummer, I guess everything is auto-configured without choice of changing it (curses motherboard manufacturer)

One last thing I remember is maybe playing with the PnP OS option if it exists: See this KB article.

No IRQ settings in the BIOS that I could see. And I'm not sure how (or if I should) change IRQ settings in Windows.

I take it you didnt have IRQ settings to mess with?

Welp, looked through my available BIOS options, and while I see PCI-E clock options (increase from 100MHz to something higher, or lower), I see no PCI latency timer option. Should it be in my BIOS? Disabling audio controller and USB3.0 controllers haven't helped. On boot, there are still 2 GPU's that are giving a code 43 in Device Manager.

Would powered riser cables fix this? Would raising the PCI-E clock above 100MHz help? Or am I just stuck using a second computer for two of the cards?

You assume correctly! Using two StarTech PCI1PEX1 PCI to PCI Express Adapter cards to make the magic happen.

Interesting.. anything in particular I would have to do with IRQ settings? Never had to mess with IRQ stuff before. Will look into this in the next few minutes, will report back.

Hm, I already disabled the Audio controller, but you're right, I might as well disable some other stuff while I'm in there. Will report back!

Latency timing eh? Sounds reasonable! Will look into this in just a few minutes!

ATI drivers cap out at 8. You can't use 5 dual GPU cards (5970, 6990, etc), but you can use up to 8 single CPU cards.

IIRC windows wont see more then 5 gpus

Did you change the latency timer in BIOS? I've read on this forum that this needs to be increased from the default (64) for 5+ cards on gigabyte boards. If I remember correctly, the theory was that you need to set the timer to 16x# of GPUs. I'm not sure about that but you should try increasing it. I'd try 96 and some higher settings as well. 7x16=112, so try that too if you can. 

I've heard of other people with those Gigebyte boards saying that the MB can only allocate a certain number of devices. Try going into your BIOS and disabling anything you don't need. I'd start with the audio, 2nd SATA controller, and USB3.0. Disable anything you don't need, and then try it.

I assume you're using some kind of adapter to convert those PCI slots to 1x PCI-E.

I looked at your motherboard block diagram in the manual and it looks like it should work.

I remember vaguely reading somewhere before that someone tried the same thing you did and having to go in and play with their IRQ settings to get it to work.

Well, all 7 GPU's do show up in the Device Manager, but only 5 work properly. The other 2 have status messages of code 43, that Windows prevented them from being used because they reported errors.

I disconnected one of the 16x to 16x riser cables from a card, disconnected a 1x to 16x riser cable, and plugged the card into the 1x to 16x riser cable, resulting in only 6 cards being connected. I did this to make sure that the card itself wasn't to blame. Sure enough, when Windows booted and finished installing the drivers for the card, I could control it via software (my test: adjust fan speed to see if controllable). What was odd was that the other 16x to 16x card could now also be controlled via software, and one of the original 1x to 16x cards, which was working fine before, now could not be controlled via software. Its as if the motherboard can only work with 5 devices at one time, which is absolutely ridiculous. Also, I checked through the manual a few minutes ago, didn't see anything about impaired PCI lanes under certain scenarios.

I've updated the BIOS, updated Windows, updated systems drivers, cleanly uninstalled and reinstalled the AMD 11.12 drivers since everyone is saying they support 7970's and up to 8 GPU's, and it seems like Windows (or the motherboard) just does not want to let me work with more than 5 GPU's.

I would be very curious to hear from others that have put together 6x or 7x GPU rigs!

This sounds like your answer in short.   Your saying the ones that previously didn't operate, do and the ones the worked before are now non-functioning?   Is this correct?

Dalkore

Welp, oddly enough, I've just switched around some cables to make sure that the cards aren't to blame, and something odd happened: the card not working is one of the cards which was working previously, and the card connected via 16x to 16x is working just fine (can adjust fan speed and clock rates). WHAT IS THIS. :|

Anyone out there ever hook up 7x GPU's and have it work?