3080 mined 2 weeks, then denys mining

[fk2]

Hi folks,

I have a small rig with 4 cards running:

- ryzen 1800x
- AX370 Gaming K7 Mainboard
- 1x 3090
- 1x 3080
- 1x 1080ti
- 1x 3070

two cards are sitting in the pcie x16 slots available on the board. two cards are connected with riser. I used these risers for months before, they should not be faulty. NVM I already switched risers meanwhile and that doesnt fix the problem:

the 3080 mined for two weeks now with memclock +1000, PL 70% and to make sure the VRAM does not throttle FAN set to 85% which resulted in tjunction of max 98°.

however, I installed the 4th card 3070 yesterday and since then I had a bluescreen and after that the 3080 shows some strange behaviour:

after reboot it shows u normal. when starting nicehash software everything looks fine. as soon as I start mining the stats like temp, fan speed, ec dissappears and so does the card in nicehash GUI.

In HWInfo I can also see several stats of the card dissappear. the card still does show up in device manager of windows. I tried to mine with the card in a complete new build and it worked. so I dont feel like the card is dead. unfortunately in my main rig i still cant get it to work again.

I tried the following:
- reinstall nvidia driver
- reset bios to defaults
- set above 4g decoding: enable and PCIe Gen: 2 @BIOS (had it working with defaults)
- swap risers, problem follows the card
- usefactory clocks by setting up afterburner once, save and reboot without afterburner loading.

anybody has a tip for me? I'm close to RMAing the card but since it is booting and surviving benchmarks I dont even know what to complain about, except mining which is not what i will complain about obv.

best regards

[batsonxl]

What happens if you discard 3070? is it still same?

[fk2]

hi batsonxl! I plugged out the 3070 first, same problem. after that I switched the riser of 3080, same problem. after that I put the 3080 in the riser of 3070 to check if it might be not the riser but the usbcable or pcie x1 slot. problem remains.

plugged it into the new rig, mining like there is no problem when on its own in a dedicated x16 slot. just checked for another BIOS but i already have the latest

e: uninstalled all drivers with DDU reboot and clean install latest 465 nvidia driver. uninstalled also afterburner and nicehash. problem remains

[batsonxl]

Looks like you have problem with windows.Ok sometimes windows really can cause headache when especially it is updates itself. If you have time install fresh windows disconnect internet and disable update and defender then try again. if you can connect only 1gpu to 16x rest keep on risers. you may run out of pcie lanes if you use 16x slots
Also you can use hiveos from usb disk without windows install at all.

[batsonxl]

And what PSU are you using for those?

[logicalray77]

Sounds like power issue, did you distribute your power accordingly?
Riser for 3080 should not share with any other risers, it`s power hungry.

[mak013]

It does not looks like as PSU problem. If it was so - the problem had to disappear after removing 3070. I can give advice to reinstall OS and add GPUs one by one. And the first GPU ought to be 3080.

[fk2]

PSUs: I am using two PSUs, secondary PSU is controlled through an adapter connected to the primary PSU.

Primary PSU is Corsair rmi1000W powering 3090 and 1080ti. those cards are the oldest addition so they sit on the motherboard dir3ctly.

Secondary PSU is Corsair 650W. It has two PCIE power cables. each one powers one of the other cards. since this is already using all connectors plus one Y adapter i power both risers by one seperate MOLEX powercable. Is this a bad idea?

I will reinstall today

[Wotan Wipeout]

Easiest way is to try simplemining or hiveos.
Might be a good idea to get 1 mining power supply.
More cables and less problems.

[varfast]

Easiest way is to try simplemining or hiveos.
Might be a good idea to get 1 mining power supply.
More cables and less problems.

Gotta be with him.

Sometimes double psu's cause problems it is better to have all the system running with 1 psu.

[fk2]

Well, I found the problem:

that was the PSU connector that powered the 3080. I used an Y splitter to get 3 ports powered but that was a stupid idea because I wasn't aware of the fact that unlike on linux, in windows clocks and powerlimits might reset which results in the rig pulling max wattage.

In that case, a 3080 with 320W TDP is too much for a single PCIe 8Pin cable that specifications are capped at 288W

Lucky me, it is just the PSU and not the whole rig or even worse, my whole house...

[philipma1957]

Well, I found the problem:

that was the PSU connector that powered the 3080. I used an Y splitter to get 3 ports powered but that was a stupid idea because I wasn't aware of the fact that unlike on linux, in windows clocks and powerlimits might reset which results in the rig pulling max wattage.

In that case, a 3080 with 320W TDP is too much for a single PCIe 8Pin cable that specifications are capped at 288W

Lucky me, it is just the PSU and not the whole rig or even worse, my whole house...

yep I have a lot of simple mining rigs or linux the clocks do not runaway, 1 or 2 times in 5 years with hundreds of cards involved.



I have a hand full of windows rigs (5) and over the years with under 100 cards used maybe 35 runaway clock drops to top power.

[fk2]

I have still some questions here. I used one of the PCIe cables that has two x 62 pin at the GPU end. Just like mentioned here, second picture: https://www.gpuminingresources.com/p/psu-cables.html

according to this, 288W per cable is fine. according to the TDP of the 3080 it is 320W. When card reaches 320W it pulls like 50W off the riser. 270 left, which should be inside the 288W with a single cable?

nevertheless: I picked  up a second PSU as currently I cant get my hands on a single 2400w psu atm. This one has 3 of those cables so I powered 3080 with 2 of them and the 3070 with one. now Im paranoid that I need 2 more cables, 3 dedicated for 3080 and 2 for the  3070.

I thought about that the CPU EPS12V ports aint used and that with such an adapter: https://www.caseking.de/kolink-adapter-6-pin-pcie-auf-eps12v-8-pin-cpu-anschluss-schwarz-10cm-zuad-765.html

I could supply 2 extra cables for cpu. I know it says adapters are shit in general but this one should be one of the good guys?!

[philipma1957]

I have still some questions here. I used one of the PCIe cables that has two x 62 pin at the GPU end. Just like mentioned here, second picture: https://www.gpuminingresources.com/p/psu-cables.html

according to this, 288W per cable is fine. according to the TDP of the 3080 it is 320W. When card reaches 320W it pulls like 50W off the riser. 270 left, which should be inside the 288W with a single cable?

nevertheless: I picked  up a second PSU as currently I cant get my hands on a single 2400w psu atm. This one has 3 of those cables so I powered 3080 with 2 of them and the 3070 with one. now Im paranoid that I need 2 more cables, 3 dedicated for 3080 and 2 for the  3070.

I thought about that the CPU EPS12V ports aint used and that with such an adapter: https://www.caseking.de/kolink-adapter-6-pin-pcie-auf-eps12v-8-pin-cpu-anschluss-schwarz-10cm-zuad-765.html

I could supply 2 extra cables for cpu. I know it says adapters are shit in general but this one should be one of the good guys?!

it may be good.  meter it with a meter and be sure it matches your pcie cables.

[fk2]

Can you please tell me what to measure with a multimeter exactly and how can I make sure it fits my cabling?  atm Im also thinking about what I can do to prevent a clock reset in win10/afterburner/nicehash app

[Commie]

I have a hand full of windows rigs (5) and over the years with under 100 cards used maybe 35 runaway clock drops to top power.

Just curious, did you use any mining OS on Windows or operated all rigs manually? Also, did you use Afterburner for overclocking?

[philipma1957]

I have a hand full of windows rigs (5) and over the years with under 100 cards used maybe 35 runaway clock drops to top power.

Just curious, did you use any mining OS on Windows or operated all rigs manually? Also, did you use Afterburner for overclocking?

okay

windows 7

windows 10

using afterburner
using trixx
using zotac
using asus

all of them have dropped settings over the years.

smos almost never does.

@fk2. tommorow I will link

cables and meter that should work.

the best thing to make sure for windows settings to stick is

never update

never boot

and have indentical gpus.

 

[Commie]

I have a hand full of windows rigs (5) and over the years with under 100 cards used maybe 35 runaway clock drops to top power.

Just curious, did you use any mining OS on Windows or operated all rigs manually? Also, did you use Afterburner for overclocking?

okay

windows 7

windows 10

using afterburner
using trixx
using zotac
using asus

all of them have dropped settings over the years.

smos almost never does.

@fk2. tommorow I will link

cables and meter that should work.

the best thing to make sure for windows settings to stick is

never update

never boot

and have indentical gpus.

 

I already see the reason why it keeps happening in your case. I've experienced that before too, but without bad consequences. OC software (any) is a) not that reliable, b) isn't really designed for our purposes and c) adds an extra risk factor as we need to load another piece of software in order to mine. It was one of the reasons I've switched to specialized mining software instead of writing endless batch files,  eliminating the very need for Afterburner and such. Nowadays all my OC settings are configured and controlled by Awesome miner that I use (and I'm sure other mining OS/SW can do it too), so it's the same program that first sets overclocking values and then starts miners. If it crashes them mining process simply won't start (but I'm yet to see that crash).

[Commie]

Can you please tell me what to measure with a multimeter exactly and how can I make sure it fits my cabling?  atm Im also thinking about what I can do to prevent a clock reset in win10/afterburner/nicehash app

The only thing to measure there is amperage and I don't think multimeter is the right tool for it (yes it can measure amps but you'll have to break the circuit to do so). Clamp meter is way better for that.

It looks like you use a lot of connecting points in your setup, which is no good. Eliminate all of your adaptors. Make/order custom 16awg cables if you need to (but if you use modular PSUs high load might melt/burn connectors on PSU side). Don't power your risers off Sata. Don't run 270W loads via single cable, that's a road to disaster. At the moment one of my rigs rund on a "backup" PSU as I took main ones off for service, I'm short of GPU cables so I run 1 cable per GPU. Even with 240W total per GPU the cable and plugs are significantly warm. Not hot, but quite warm. I wouldn't want to think what happens if suddenly 300-350W will run through them.

[fk2]

Need to clarify some things:

Eliminate all of your adaptors

besides the Y adapter from 8pin to 2 6+2 pins that caused the meltdown I did not use any other adapters only what came with the PSUs. After damage I only use what came with the PSUs. The risers are MOLEX powered and to be on the safe side it is now 1 cable exclusive to one riser, MOLEX connector from the PSUs.

All PSUs are from corsair, I have read in their forums they use AWG16 from the first cable. I found on their forums one of the usual pigtail cables are specified up to 300W.
Right now Im testing the following:

- Reset all GPUs to factory settings
- Measure temps on the connectors with infrared tool

So I found this german article about stressing the NVIDIA 12 Pin connector: https://www.igorslab.de/tester-in-gefahr-wieviel-watt-ueberlebt-der12-pin-connector-der-nvidia-geforce-rtx-3080-und-rtx-3090-ohne-zu-schmelzen/

I find the following info useful:

- Temps are rising pretty fast within 10 minutes
- up to 144°C and connectors start smelling but not melting already

So far im mining for 25 Minutes and the hottest connector I can measure is the ones from the 1080ti and the 3070 around 58°C. 1080ti is connected with 2 pigtail cables, 3070 only on one. with 220W TDP being the lowest power consumption card in my rig I went for one cable on the 3070 though i know recommendation in general is also 2 cables.

 I thought the connectors on the PSU sides are the hottest since this was where my meltdown occured. But these are cool @ around 35°C.

Lets assume one pigtail cable can make up to 300W before it starts melting. The TDP of 3080 is 320W. That would explain how it happened but is it really like 20W above specification can lead to temperatures in the 3 digit area? Is it like once you go above the specification it only gets warmer continously?

e: another question comes to my mind:

lets say we have a 1200w bequiet PSU with 3 exclusive PCIe pigtail cables and a total of 6 connectors. we want to power 2x 3080 with 3 8pin connectors each. Is it okay to both connectors of the first two cables into each slot 1+2 of the GPUs and take one connector of the third cable into each 3rd GPU slot?

trying to visualize here: