Bitcoin Forum

Hey all,

Last night my screen began blinking on and off with some red pixelation.  It became so bad that eventually there was more screen off time than on.  I contacted NVIDIA tech support thinking it was a GPU issue and after 3 hours of chat tech support with them on my laptop, I gave up and went to bed.  Here's what I know in a nutshell:

1) I have a Corsair HX1000i PSU, ASRock BTC H81 MB, 8 GB ram, (4) MSI 1070 GTX 8 GB GPUs, and a celeron cpu.
2) NVIDIA tech and tried musical chairs with the riser seating (BTW they said USB risers are NOT supported by these boards) to no avail
3) I have flashed my BIOS and updated the GPU drivers to the latest
4) Each GPU works individually
5) No issues found in msinfo according to NVIDIA
4) Currently I have (2) cards up and running.  As soon as I add in the additional cards, the problem comes back.

This leads me to think there's either a MB issue (I have a help request to ASRock), a power issue, or a riser issue.  I really despise not knowing where to take it from here and having essentially have $1K sitting around doing nothing.  Anyone that can help me get up and running I'll even send a token of my appreciation.  If this is in the wrong topic area, I apologize in advance.

Thanks

What was going on before that? Has this rig been happily mining for days/weeks/months, or is it a new system you've just built?

Do you use self-powered risers ? did you connected the additionnal molex power connectors on each side of the PCI-express slots on the motherboard ? What powerlevel you use ? Sound to me a power issue, feel free to share more infos :)

I started back in late august with just (2) 1070s.  At that point, it ran well.  When I added on the 3rd, I did notice some minor blinking on occasion when I probed the overclock limits of each board but went away after a restart and toned back Afterburner.  After the 4th GPU came on board it started happening more frequently but again was fixed with a restart.  As of last night, the restart no longer worked which is when I started going over everything, cleared the drivers, and went to NVIDIA.

I'm using these risers: https://www.amazon.com/gp/product/B06XGTM694/ref=oh_aui_detailpage_o05_s00?ie=UTF8&psc=1

I'm using the molex connectors that came with the PSU; each cable feeds 2 risers.  Then each GPU has it's own dedicated cable back to the PSU.  I also do have both molex connectors attached to the MB to stabilize PCI at the voltage per ASRock instructions.  I normally sit around 72% power with the Afterburner.

I've got 2 more risers coming that I'll try swapping out of the non functional GPUs tomorrow.

Ok sound like your setup is ok,

How do you power the 8+6 pins gpu power input ? 1 cable with a Y separation ? 2 cables ?

Have you tried to use for example 60% powerlevel ? does the issue happen before mining or at mining start ?

Your psu should be enough, with this configuration you should reach about 650w at wall.

Maybe a bad riser, but it seems odd it crash the rig completly. Are you using gen1 pci in motherboard bios ?

Answered in order:


My PSU has enough individual outputs on it that I run a separate cable from the PSU to each GPU.  I felt this would be the safest way to avoid voltage drop despite the snaggle of cables.  I'm getting tempted to see what voltage each one is seeing.

At this point, I can't even get past the Windows login screen with 4 GPUs on let alone changing my undervoltage levels  :-\

I sure hope so!  I picked the 1000W unit as I thought it would be enough to max out the MB with PCI slots.  The highest draw I've seen at 72% is roughly 575W at the wall using a Kil-o-watt meter.

I don't believe so but I'm sure I can check.  The MB comes with 1 x PCIe 2.0 x16, 5 x PCIe 2.0 x1  I'd have to assume it was stock 2.0.  Hell I didn't even know that was a selectable thing, figured it was a hardware protocol.

Well if it locks out even before you enter windows that's odd indeed. Which version are you running ? W10 Enterprise patched anniversary ?
I don't think gen1 would help, but doesn't hurt since it save up a bit of watts (5 to 10)

Also the 1070 you use have 6pin + 8 pin power input, did you powered them both with different cables right ?

Windows 10 Pro-Version 1703

Negative; I have a single PCIE cable coming from the PSU that branches into the 6 and the 8 Pin.  Is that incorrect?

Ok, well i prefer to use 2 cable not one splitted for more stable power, but it should be ok if you powered the risers and the motherboard, also you don't go pass login in windows, and the cards don't pull any power @ this point. Well maybe a bad riser then.

because I just had this issue with my mining rig last week, my money is on a bad riser.

I used MintCell on amazon and powering with a PCIe cable from PSU and fixed all my issues.

I'm going to try turning on turning on another GPU tonight and see if the issue happens.  Then when the next 2 arrive I'll swap out what I can.

It seems unlikely that both are bad, so what would you guys check next if swapping out the risers doesn't work?

I'd thoroughly check the GPUs. You said in your 1st post that "Each GPU works individually", but did you test them each with a display attached? It's just whenever there are some graphics glitches (and you said there was some blinking and color artifacts) there's always a chance that the actual GPU is the problem.

I would try that. I even had a Riser that when one card was plugged into it, it didn't read but when I have the GPU plugged into either a new or different riser it would work. I will always have a brand new set of Risers incase something messes up in the future that will be the first thing I check now.

I ended up doing a clean install of windows, and plugged each card in one at a time do the drivers could install shut down and reinstalled another card until I knew all the drivers were the same. (probably not needed because I'm 99.8% sure it was one of my Risers)

Then, when I was putting the rig back together, I powered on with 1 GPU on a riser and once I saw it working I shut it down and added the 2nd one and did that until all mine were working (6 total).

If you need any help shoot me a PM. Ill be up late tonight and understand the feeling of troubleshooting over and over. Still research what you can so you can learn, but if I can help speed anything else (learning wise/troubleshooting) I will try to help

Correct, NVIDIA asked me to do exactly that.  I ran the system 1 card at a time using the HDMI outputs and they all functioned normally.  Tech support said it was very likely not to be the GPUs but I'm not going to rule it out until this is fixed.

Well I'm back up to 3 cards again without distortion.   ;D  I plugged in the 2 cards not functioning prior, one at a time, and found one being the problem child.  I'll replace the riser on that one tomorrow and see if that does the trick.

Do you only have one system you can test these on? If you're planning to continue mining and will be adding more cards in the future I strongly suggest to build some kind of a test rig. Doesn't have to be anything fancy, an old system with just one pci-e slot and some simple PSU will do just fine. Whenever you have some issues with your main rig(s) you just pull the parts in question and install them into that test rig to find the source of the problem. Having a spare rig like this simplifies the diagnostics greatly.

Currently yes this is my only rig besides my very simple desktop doing BOINC.  I was using my mining profits to grow my equipment so I'm at this point now since this summer.  Well I threw some extra at it but you get the point.

I think that's a very good idea to have a test rig before installing in the main one.  Would definitely point out problem hardware prior to bringing it online. 

I've had nothing but issues with risers with molex connections.. I switched to 6-pin and haven't had an issue. Thankfully my risers had both types of connectors.

Yep, and you can assemble that test rig for next to nothing. All you need from the platform is a PCI-e x16 slot, so it can be anything as old as intel's socket 775 or amd am2. You can run win10 even with 2gb of ram, I guess, and obviously any old hdd will work too. Can source a system like this off ebay or some local classifieds for as low as $20. Then you add a PSU (doesn't have to be very powerful or efficient, I'm using some old cheap 460W coolermaster) and that's it.

I'm also running any new card I buy through the test rig first. Just to figure out the maximum overclocks and check the stability etc, make sure everything works. Then it goes to one of the rigs.

*UPDATE*

Replaced the riser in question.  I've been stable in mining ETH for the last hour.  No blinking!!!  :o  I'll report back if anything changes.

Great, risers are often your sources of troubles, always good to have a few spare laying around  8)

glad you are up and running. adter all that troubleshooting im sure you feel like you can fix any issue now. glad it was something easy like a riser. I have a hand full on just in case now.

Waiting for the new year before I start building Rig #2. 6-8x RX580 8GB cards (they just keep getting cheaper)

Different motherboards will also kind of determine the overclock potential for the GPUs.

So benching your cards in a test system will give you a good reference point, but it most likely wont be identical when you swap it to its home system.

No they won't. Why would they? For CPU overclocking yes, there might be slight differences due to different VRM setups and the BIOSes using different approaches to configuring some parameters (different vdroop and LLC settings, for example), but what does the motherboard have to do with the GPU overclocking? All that matters (besides the obvious silicon lottery luck) is the power quality (PSU), and even that doesn't matter too much provided that there's enough power (unless the PSU is total crap). Never ever have my GPU OC results depended on the platform the card is used on, and that's 100+ GPUs and at least 20+ boards over the years.

What makes you think so? Have you ever had (repeatable) cases of one card running fine with set OC on one board and not being able to run with the same OC on another? With exactly the same PSU and not a faulty board? Excluding some nonsense of course, like reference rx 480 cards drawing more than 75W from the PCI-E slot, thus crashing on some boards and running better on others. In normal conditions, without anything like that?

And the blinking/red screen is back.  Now I'm truly frustrated.  ASRock has offered to RMA my board but frankly that would mean my rig is down for weeks.

Any suggestions  :'(

Have you tried putting 1 GPU directly on the MB, the rest 3 only on raisers?  
How much do you Overclock the memory?

No because the 1070 it block one of the PCI slots so I could only put 5 cards in instead of 6.  What would be the benefit of doing this? I'm open to it at this point.

I do overclock, around 620 on afterburner.  I also jumped power limit to 75 this morning to see if that would help.

Sorry, I am confused.  You said you have 4 GPUs -  (4) MSI 1070 GTX 8 GB GPUs,   now 6.  

Try to disable memory overclock, see how it goes for a test.

Sorry didn't mean to confuse  ;D  My end goal was to eventually have 6 in total.  That's what I sized my MB and PSU for.  The rig CURRENTLY has (4) 1070s.

I'll see how it looks when I get home from a day of mining.  I'll give the clock test a try later.

Somehow I have the feeling what you see is related to the GPU, not the MB.   That is why I suggested to remove the memory OC and just see how it goes.
If those are with Micron memory, which probably they are,  620 is a bit more than it can take long-term.

So may be try this
Keep keep TDP at 70-75% or whatever is preference for you.  (not less than 60% because hashrate drops a lot then).
Try a while with no memery OC.  See results
Increase OC of memory to 350. See results
Increase OC of memory to 500. See results.

don't OC core to more than 90-100.

So a day of TZC mining went just went, no issues.  Here are my current OC settings:

Power Limit: 73
Temp Limit (linked to PL): 71
Core Clock: 0
Mem Clock: 620
Fan: Auto