Asus TUF 3080 TI fluctuating hashrate

Thanks for the reply @everon1218 . I’ve tried 2600 mem but cant get passed 76-77mhs…
I’m using 1350 core, 2300 mem, 0 pl atm and I get jumping from 86-90 even tho i set “hashrate-avr”: 60 in miner Extra config arguments.

I’m guessing it’s not good as I’ve seen it at one point going 52mhs??? Don’t know how 74% unlock is still holding :smiley: Really mind boggling card

Edit: oh and trying drivers 470.86 now

Zimbaba,
Thanks for answering. You just wait a little bit of hour.
When you are dropped on hashrate, the the miner auto reconfigures itself.

At first, the miner sets the LHR unlock to 74.5, next to 74, now its to <>73.5.
The miner changes its configs automatically - i.e. auto reconfigures itself alone, Im just waiting and watch for temps.
Im not using “hashrate-avr” option. Still on 470.82 nvidia driver.

The miner is configurated in auto mode by default.
Im on that configuration almost 20 hours.

Thanks for sharing! The model is quite capricious, really :slight_smile: The very finesse

ah I got the case of throttling for sure, atm 1250/2300 with 35C on the GPU @ 100% fan + airflow 2 fans by the window :slight_smile:
which means <90C mem and I keep between 89-90 mhs, rarely dropping to 88mhs. Guess it’s time to change the pads after all. Fcuk warranty

Did you change your thermopads? It is a first step to do after buying any 3080TI’s.

I will send a command tomorrow morning, so you can check if throttling

Use hiveos shell:
nvidia-smi -q | egrep “Thermal Slowdown”
image

@Grea, thanks. Its much more informative

1 Like

nvidia-smi -q -d PERFORMANCE

2 Likes

Thanks for this very useful!
I’ve noticed that all my card nvidia-smi -q -d PERFORMANCE are on P2, should i force P0 in overclock or this doesn’t matter for 30xx cards?

tried p0, with lower clock to adjust for the +400 when using the mode, no changes. This card is not throttling according to nvidia-smi -q -d PERFORMANCE but it simply can’t go above 2300 mem clock, and with 1700-2300 mem clock it’s fluctuating mh. I’m out of ideas really :smiley:

Here are the results: (Outside temp ~10C)
Placed the rig next to the opened window ~88 hash (avg I’d say jumps from 86-90) on 1250core / 2300mem (24h doing fine - core temp 34-38)
Closed the window hash immediately dropped to never reaching above 85 hash (core temp 45)

So this card is throttling for sure but nvidia-smi -q -d PERFORMANCE doesn’t recognize it doing so.

While setting it up first time and trying out settings in windows I’ve noticed that on 40C core, mem temp was at 90C.

So am i wrong to think so, what’s going on here? I’d appreciate any explanation

New version has the throttling for nVidia and limited memory data for HBM/HBM2 GPUs exposed.

use the tool from Hive Shell:

nvtool --throttle
nvtool --memtemp (only shows temp for HBM based memory, not GDDR6

0.6-212@211124

  • Added display temperature of memory for Nvidia GPUs equipped with HBM/HBM2 memory e.g. A100, CMP 170HX, etc
  • Updated nvtoolto v1.57 (added memory temperature reporting using option --memtempfor GPUs with HBM/HBM2 memory; added option --throttleto show throttle reason which also reported by nvidia-infotool, so you can look all info using it)

I dont have explanation but maybe you need more control over your card under Linux. I am almost in the same situation with the window but I dont even think to close it :smile: 10-15 cm opened

Im trying get more control over linux, and found that article. It has a lot of commands and sources. Im not sure how to implement this article in HiveOS but will try.

Opened shell in a box and ran: image
Its working on hive…

Next lines added 7 days later.
Found another interesting post in nvidia forums with a lot of questions which I have too:

And the last answer

The key insight is that when you apply an overclocking offset, you are shifting the curve that you see in MSI Afterburner. You can’t move individual points, but that’s rarely interesting. You then set a max frequency which is equivalent to flattening the curve in Afterburner. Combine that with lowering the power limit and you get real undervolting, but by indirect means.

which goes to:

1 Like

Another solution came up today RTX 3080 TI HASHRATE FIX - YouTube
Not tried yet.

Just a quick video that will point you to how to fix the Ethereum hashrate on your RTX 3080 TI. I have not been stuck at 77 MH/s myself, however after applying it to my Inno3D RTX 3080 TI ICHILL X4, I went from 89 MH/s to above 90 MH/s.

The main thread on reddit: https://www.reddit.com/r/gpumining/co…
Direct link to vbios that fixes the hashrate: https://www.techpowerup.com/vgabios/2…

Please note that it can brick your card and that updating the bios is your own responsibility.

1 Like

Cheers bro, was just about it to post.
I did flash it and the hash seems a bit higher and less fluctuating…

Cheers, hope you all are well.
I closed the window and 2 days already working fine with the same temps. :heart_eyes: :sunglasses:
Recently updated the post from Dec 5-th, 3 posts above

  • Undervolting nvidia gpus in linux -

looking the volt curve function like in MSI Afterburner under Windows, and
I think its a key point of overclocking the gpus to achieve best performance.

…2 days later, I just found that there is option for Absolute core clock in HiveOS, in Overclocking option (e.g. in the miner) so now just hope to lower the watts and temps… so many researches to lower the voltage under Hive and Linux. If some of you pointed about that, Im getting to the point just now.

While testing the overclock settings in Hive, except the main ROXTerm, open another 2 Terminals. In the first one run
nvidia-smi -h
Scroll down to QUERY OPTIONS and check out -d (display) commands. Leave it open, for reference.

In the second terminal use the statement from above
nvidia-smi -q -d <QUERY OPTIONS>

While tweaking in Hive, check out what is going on with the card by running the Query command.
Its possible to check the Clock Limits with
nvidia-smi -q -d SUPPORTED_CLOCKS
and few other useful options as VOLTAGE.

This topic was automatically closed 416 days after the last reply. New replies are no longer allowed.