Issues with 2 GPU setup: 7790 + 5870 (invalid nonce - HW error)

[plonk420]

to save some power, i added a 7790 to my 5870 for a mini dual card setup.

it pretty much seemed to take an OS reload to even get just the 7790 working (and THAT didn't even work until i manually installed OpenCL via the MSI file from the CD that came with the card).

after a fair bit of uninstalling and reinstalling, i finally got the cards to even show OpenCL support in GPU-Z (i think it took the ULPS registry entry to do this). however, even then, all clients were crashing: CGMiner and BFGMiner; GUIMiner won't even start.

FINALLY, after i put the 7790 in the top PCI-e slot (out of 2 16x-length slots), would *anything* get a result: CGMiner, however, the 5870 only gives "invalid nonce - HW error." the 7790 works just fine. and all the other apps still only crash.

my most recent fix attempt included hooking the monitor to the 5870, it's exactly the same.

i'm at a loss.

any ideas?

currently running
Athlon II X4 630 on GA-970A-D3
Mushkin 2x1GB DDR3-1333
Server 2008 R2 x64
PC Power & Cooling Silencer 750 watt
CCC 13.5 beta (but also most recently tested on either "12.101.2" (whatever came on the disc) or 12.101.2.1-130416a from the website--i can't remember)

a side note just to verify the GPU doesn't seem faulty: Furmark works on the 7790 as well as the 5870.



additional finding while waiting for my ability to post:

i got finally Phoenix2 working alongside CGMiner, but the hashrate is 10m/hashes lower than it should be, but more concerningly, i'm getting ~3-6% rejections (CGminer has 0.3%). it would also be nice to be able to use just 1 app.

[ISAWHIM]

Your driver/cgminer is bugged... Just try reloading the "saved" card profile, and restart the miner program.

If the problem does not resolve itself, then you have to adjust the settings until you no longer get hardware errors.

I see this a LOT after a short run, or card-cooling, or just an odd profile setting for that card.

One setting will not work for all cards, unless you go "completely safe". Also, running two cards is not the same as running one cards settings. You will have to adjust for the SHARED bandwidth/slots/threads/workload... (Lower intensity, less threads, lower thread values, individual clocks/voltages/fan-control... etc...)

Using any single 16x slot may make it 16x... but using both, may degrade them to 8x each... or only one is 16x, the other turns into 8x... etc... (Less bus-speed/volume)

Could also be that a pin has "dust" on it, thus, no contact... or the card is not completely seated right... (pin touching two contacts, or not touching "enough".)

[GenTarkin]

cgminer 3.1.1 seems to have this issue for me on random gpu's too, it happens over time after hours of mining, randomly start to get HW errors - invalid nonce.
I went back to older cgminer version and problem went away...

[plonk420]

Your driver/cgminer is bugged... Just try reloading the "saved" card profile, and restart the miner program.

what do you mean "saved" card profile?

Using any single 16x slot may make it 16x... but using both, may degrade them to 8x each... or only one is 16x, the other turns into 8x... etc... (Less bus-speed/volume)

Could also be that a pin has "dust" on it, thus, no contact... or the card is not completely seated right... (pin touching two contacts, or not touching "enough".)

the 5870 slot runs at 4x (which someone i know from IRC says should even run a 5970 just fine). i'll have to play with thread and other settings, i guess.

would i be getting 94+% accepted shares if it was a bad pin connection?

cgminer 3.1.1 seems to have this issue for me on random gpu's too, it happens over time after hours of mining, randomly start to get HW errors - invalid nonce.
I went back to older cgminer version and problem went away...

what's a good version? (or at least for you)

also, thanks for the suggestions!

[-ck]

As I've said elsewhere, previous versions of cgminer would create garbage hashes without hardware errors if you're mining at high difficulty on scrypt. If the new version produces more hardware errors, it's only because they're all visible now.

[plonk420]

how about bitcoin? that's all i'm mining.

edit: i've tried all the way back to 2.6.6 and i'm still getting errors. tried 1 gpu thread and intensity 1 and 2. still no improvement. looked at shaders, but it looks like it's LTC only.

[HellDiverUK]

2.11.4 works OK for me with a 7950, 6970 and 5870 all in one machine (FX8320@4.5GHz, 990FX SABERTOOTH).

[ISAWHIM]

I use afterburner... I have profiles setup for "1 = SAFE-LOWPOWER", "2 = BITCOIN", "3 = SCRYPT", "4 = TESTING", "5 = DEFAULT"

CGminer attempts to "control your card", using settings you have in the BAT file, or the conf file... (Or default settings if you have none, the ones set by the drivers/control-center/afterburner.)

Those often conflict with the "card drivers" or "overclocking tools" settings. They do not "initialize" correctly. (Eg, setting voltage low, while the clock is still high, causes it to fail, thus, HW ERROR...)

This results in constant failure upon reloading cgminer... until you fire-up your drivers-controllers, or overclocking-controllers (afterburner), and "Reload some profile" that is "safe"... If you catch it at the right time, the voltage MAY not actually set first, letting the clock lower first...

EG... I have a profile that is 1100-CoreVolt 900-CoreClock 1600-MemVolt 300-MemClock...
But my profile for CGminer, in my BAT file, is... 1200-CoreVolt 1075-CoreClock ----MemVolt 150-MemClock

So... when the core-volt is set UP, that 900-MemClock is functional, until cgminer can raise that to my higher settings also... If I had the core-voltage higher, and matched to a 1180-CoreClock... then tried to lower the voltage first, to 1100 while the CoreClock was still at 1180, that would cause a HW error, because the cards fail loading the profile as they crash in the setup, and thus, return to a default "safe" value, or remain half-set.

Resolved by simply "loading" and "applying" my safe profile in afterburner, then re-running CGminer... fixes the HW errors... unless I have a setting that is causing a HW error, then that has to be fixed in my profile. Because something-else I SET that is setting, is failing.

[Twingo]

I have issue with 5xxx and 6xxx cards paired with 7870 in my setup, the problem is guiminer would reject the shares from 5870 cards saying it's hardware failure, did a search on goggle and found that lots of people are having same problem, I ended up with building another rig and throw the 5xxx and 6xxx cards in, and use the first rig for 7870 and 7950s, I think it's the latest drivers for 7xxx cards are causing the 5xxx and 6xxx cards to get wrong results, this is definitely a driver problem. I'm currently using 11.11 for 5xxx and 6xxx, 13.5 beta for 7xxx cards, both combinations have been working with no problems for a few weeks.