3080 – GPU Driver Error, no temps

[pavanery]

I have the same problem with the 3080 Gainward Phoenix card. Initially I put it on my rig, it made me lose 24 hours trying to adjust everything! So I put it in a separate machine, testing it with a riser, testing it directly in the motherboard slot, I tried all kinds of OC configuration, my temperature is normal, below 50C; So I have realized that it may be a problem with the production of the card or limitation of the BIOS of it with OC, because when I use it without any OC it works normal, something happens with this model that hiveos doesn’t know how to work on. I haven’t tested it on another system, but it seems to me that the hiveos script that collects temperature information and the fan goes into a loop trying to get information and can’t.

I have 3080 from zotac that runs the OC quite high and is stable at only 213W; This Gainward to work with any OC needs to be 240W.

3080 Gainward Phoenix model supports up to 2050 MC; 1050CC with 240w PL in my tests: 93MH/s

This limitation of OC does not make sense, I use 25% more in zotac and it delivers 100MH/s stable.

Any gain above that for Gainward, stable, is to be happy!

[Gardenal]

I have 3 x RTX 3080 Gainward Phoenix, each one works in a different way! The GPU1 and the GPU3 is the same model MICRON GDDR6X - 94.02.42.00.8E and the GPU2 is …8F.

[pavanery]

@Gardenal , But are all 3080s running without errors?

[Gardenal]

@pavanery , Yes, in the GPU1, if I put the memory over 2150 it starts to brake and I see errors in hiveos!

[pavanery]

@Gardenal Exactly, I see that your gpu1 is a twin sister of my haha You have to keep your OC low or start all sorts of problems, I’ve already updated the bios but it’s still the same. Maybe I will change the bios to the own version of gainward, to see if it improves.

My bios is 94.02.42.00.8F, I will test it later with your 8E to see if it resolves, in your case you could change the bios from gpu1 to 8F to see if it corrects this OC problem.

BIOS 94.02.42.00.8F · is from Palit, not the original from Gainward.

How long has it been running without any errors?

[Gardenal]

@pavanery
I noticed that GPU 2 is different externally from the others, its a tag GS! In the box has a sticker " GS : GOLDEN SAMPLE - BORN TO KILL"!

GPU2 works fine, low PL and temperature and nice mh, its in bios 8E!!!
GPU3 works fine, but more PL, its in bios 8F!!!
GPU1 in this config -250, 2150, 245 works fine, 6 days works without errors!

When you change the bios, please tell me your result!

[Gardenal]

[pavanery]

@Gardenal Great feedback man! Mine is the non-GS model. Yes, when I make the modification, if it works, I send it here for you to try too.

Which miner are you using? Are you using autofan?

[Gardenal]

@pavanery
T-Rex, OS VERSION 0.6-203@210414, full update! Nvidia drivers installed version 460.67 (CUDA 11.2).
Yes, I using autofan.

[SKYBOB66]

i updated yesterday to the 460.67 firmware and fix it, but now im getting some invalid shares errors in 24hrs i have about 15 bad shares, but its better than offline, since my rig its in my office. im forcing in the morning the hiveos update to see if it fix the shares

my AutoFan its set off

[highflyer]

Your problem is you are using autofan on 3000 series cards which don’t show memory temp in hiveos so autofan is less than useless. Turn it off and set the fans to 85%-90% minimum full time, this is mining they need that much anyways. The memory core is running extremely hot like 110 degrees at those fan levels and throttling, this is a very easy fix. Up the fans cool off the cards because the core runs cold with Ethereum it uses all memory to process so if the core is even slightly hot know that your memory junction temps are on FIRE!!! So, just turn OFF AUTOFAN, I keep seeing people trying to use it and burning up their brand new cards. Once the fan’s cool off you will get full power levels and speed back. Problem solved. If you don’t believe me just google it, this is a very well known problem with the 3000 series and autofan simply doesn’t work with them because HiveOS doesn’t show mem temps for them. Only windows HWINFO app has memory junction temps and believe me you wouldn’t like what you see with fans that low. I would turn her off, let her cool down completely. Also add a 100 watt box fan to it, takes a lot to cool off 90’s. Also the fans don’t reach the memory junction part of the cards they simply aren’t made for mining, gaming they don’t reach past 100 degrees basically ever, so to even have a hope of cooling off the memory junction temps I cannot overstress this you need to turn off autofan and up the speed to 85-90% heck I some do 95-100% and just replace the fans they give em to you for free warranty or not usually. Better to replace a bad fan ever year or two than burn out a card. Finally, your current GPU target temp is set to 90 which is way too low for 3000 series if it actually indeed did work, they throttle at 110 and usually run if you get it right around 100-102.

[Devilz]

So, any update or fix about this issue?

[mad]

Hello all hope your doing fine,
On my rig I have 2x3080 zotac AMP holoblack.
And 4x3070 msi ventus 2x. Power supply 1800watt, Asus z490-p, i3 10gen, 8gb ddr4

I was getting the message “GPU driver error, no temp” quite often until I downgraded Hive image to 0.6-203@210403 and Nvidia drivers to 460. 67.

I changed Thermal pads for one of my 3080 after accidently broken a blade of the right fan I glued it back on with hard glue ( it’s stable like nothing happened)

The rig will run with no errors until it reaches
8 hour and then again it displays the error message and reboot.

My overclock for the zotac 3080 are

• 200 core +1800 mem PL 230w fan 76%
  temp 50°c I get 96.5 mhs for both gpus

For the 3070 msi 2x OC are
1100 core 2600 mem pl 125w and 1 3070 pl set at 118w because it was getting hotter than the other 3070s even though they are the same gpus
I get 62. 37 mhs for 3x3070 Temps 48-53°C
and 61.5 mhs for the 3070 temp 55°C

Autofan is disabled, hashrate watchdog disabled also
I’m using Trex miner

I would like to know if my problem is a software issue maybe I should try a different invidia driver or OC settings or a hardware issue (power supply isn’t enough, Risers,ups…)

Will appreciate your help, please let me know what you think

Thank you

[shawn23232]

Ok, here are few things I tried that have worked for me.

I got 3 Gigabyte 3080 Gaming OC, two with bios of 94.02.42.40.33 and one with bios 94.02.42.00.4D.

Two of the 3080 with the same bios worked perfectly out of box for two weeks no problem clocking around 98Mh/s, then started to thermal throttle down to 80-85MH and weren’t using the 230 PL I set. Was only using around 200W. Changed 2mm thermal pads for both front and back, now its back to normal clocking 98MH/s for a week.
The other 3080 with BIOS ending in 4D is the one causing my rigs to restart all the times. Changed thermal pads just like two other 3080, and it still won’t go over 93Mh or 1300 for mem. Can’t recognize Fan speed, no GPU temp and can’t apply OC settings and all sort of errors if I put Mem above 1300Mhz. I have also noticed that this GPU could not flash BIOS for some reason. I have first saved the original BIOS and tried to flash it, it failed. Then I tried to flash other two good card’s BIOS, it failed as well.

My conclusion is to try to find the cards that’s causing the issue and set the Mem OC low, probably 900-1300. Then if its working and not restarting or giving you errors in miner terminal, try to flash its original BIOS and see if that works. If it doesn’t, it probably just means its one of the defect or bad cards like mine. Set it to low Mem rate and call it a day.

Add:

One other tip I got from my friend is to do a power shutoff and reboot with no OC settings. Let it run for 10-20miutes with no issues, then apply OC settings in one go and see if that works. Set a Delay for the OC profile, so next time your rig restarts, it will have 30-60s of delay before applying OC settings

Hope this will help someone, please report back here if it works.

[Johnnmarnell]

Hey how did it go? I am following you. Mine was doing well the other day around 100mhs but maybe the temperature outside it got hot even though in basement.

[jmw60546]

I have had to just manage the overclock settings to keep it stable. Each card has its own limits and it takes time.

The hot weather makes it hard to keep cool and I get more failures than usual as a result.

[Johnnmarnell]

I think just a HIVEOS issue if you are only getting the fan missing. If fan missing, just verify in machine fan is running and ignore it. If you are getting more GPU issues then pull back the overclock. Temp outside makes a big difference as well. I have my internal fans, then two extrenal fans blowing air in and out of my rig. on 4 of 6 I can get overclock mem up to 2550 and stable doing like 101 mhs. on one of them only 1700 and on another one only 1800. If I put the 1700 up to 1800 will eventually error out. IF I put 1800 up to 1900 same after awhile. You just have to find the stable speed. BUT if just fan mssing ignore it as long as you know it is running. I also found a cool way to make Hiveos reboot for me each time a GPU error does occur. If you need to know let me know. With 6 3080 I am at 598.5mhs. I really want to hit 600 lol.

[CristhianJP]

@Johnnmarnell

I also found a cool way to make Hiveos reboot for me each time a GPU error does occur. If you need to know let me know. With 6 3080 I am at 598.5mhs. I really want to hit 600 lol.

Hi bro, could you please tell me how did you do that?

Thanks in advance!

[Johnnmarnell]

I had a few issues lately with it as IFTTT is having lots of problems. What I did though is set my hashrate in Hiveos to restart after it drops below 500. This just means one of my workers goes offfline, OR if my GPU goes down. I set the restart to 1 minute and the reboot to 1 minute. Then I have it set to trigger the notification to my telegram account. So, anytime it goes down and is supposed to reboot I get a telegram notification. Then I bought a KASA smart plug on Amazon for under $20 and created an IFTTT (free) so that anytime I get a notification from Hiveos saying offline or whatever keyword they send it shuts my KASA plug off. And then added another notification after that to turn it back on. But, lately IFTTT has not been triggering well and think the company might just stink, lol. One last thing you have to link your IFTTT and allow it to access your telegram by creating a group in telegram with your HIVEOS and IFTTT allowed in that group. This is how you create the rule. Are you using Linux? If so, I found a cool notification program you can install directly into Linux and playing around with that but so far cant quite get it to work yet. It can not be this hard and shocked someone does not have a simple script like I do for all my windows machnines. If you want to know for windows I use a very simple script that checkes and makes sure it is always running.

[Johnnmarnell]

And then hate to say it but worst case, fine your overclock setting that causes an error. One of mine can hash 101 per sec, one only 95 per sec so each one you tweak on its own. Once you find that, lower the memory by like 3000 (so my max on the 101 /s is about 2700 overclock memory. I reduce that to 2400. You may lose a little, but then I am up for weeks and I restart it every week just for fun as in my head l think it is good for no reason. HA ha. But rather than keep it close to the max overclock and have it shutting down once in awhile. I can go from like 590/s with 6 gpu down to 582 with 6 GPU and when it is up all the time it ends up averaging out much better. So lose some but VERY stable. One of mine max out at 1700 I just set it to 1400 as that ends up being only a 1.5/s hash loss and still get 95 out of it. THEN DEF GET A SMART PLUG. Cause no matter where I am if I see hiveos down or get notificaiton and it does not work, I just can restart it on the app so never down more than maybe an hour.