Ubuntu 20.04 GTX 1080ti clock issue

[reb0rn21]

So old rig that worked fine on 16.04

Tested all 390, 435 and 440 driver as soon I apply power limits let say 160W and start miner clock drop as mad to ~500-600Mhz

depending on miner used, kawpow is droping it most, power state is same P2 just clock drop as mad, aslo power use does not drop its limited to 160W

i think even mining at 250w the hash is a lot worse

gtx1070 i have in other rigs have no issue 110w limit is 14mhz and 1080ti 160w limit is ~11mhs (insane)

[rikuu]

Do you really need Ubuntu 20.04? I never upgrade the OS/drivers if everything works

[JayDDee]

Different Cuda/compute version? Try compiling with compute 5.2 max.

[Grim]

This isn't an os issue its a 1080ti specific issue.

Afaik the card changes to P5 if you set the power limit too low.

[reb0rn21]

Its on most/all miner gminer, nbminer, performance state is fine P2 clock just drop as insane and power use/draw is still huge

cuda version has nothing to do as miner is precompiled, and i tried older drivers that are not 10.2 cuda its same behavior

I aslo notice on GTX1070 if card go over 65C temp or so it drop the clock a bit, but here with gtx 1080ti temp is no issue

But its strange as GTX 1070 in different rig same ubuntu configurations works just fine... PL limited to 110w and it hashed kawpow ~14Mhs

[rikuu]

The core clock you experience on your GTX 1070 above 65ºC is probably due to increased power consumption caused by power leakage at those temps, making the GPU to lower its core clock to stay within TDP.

[JayDDee]

This isn't an os issue its a 1080ti specific issue.

Afaik the card changes to P5 if you set the power limit too low.

Yet it's the same 1080ti but a different OS.

The focus seems to be on performance based on the power limit, not the actual power usage.
I don't think we have a clear picture of the problem.

[reb0rn21]

This isn't an os issue its a 1080ti specific issue.

Afaik the card changes to P5 if you set the power limit too low.

Yet it's the same 1080ti but a different OS.

The focus seems to be on performance based on the power limit, not the actual power usage.
I don't think we have a clear picture of the problem.

Problem is at 160W power limit gtx 1080ti hash only 11Mhs on kawpow and it should do ~20+
also the clock drop to 500-600 Mhz which i never seen in ubuntu 16.04 what ever algo to drop under 1200Mhz at 160W

And no @rikuu
Its not power leaking as in ubuntu 16.04 same rig works at 75C without issue
They could have changed some rules about temp and power and made some shit storm, but it looks like more issue with the ubuntu 20.04 in general

[JayDDee]

Problem is at 160W power limit gtx 1080ti hash only 11Mhs on kawpow and it should do ~20+
also the clock drop to 500-600 Mhz which i never seen in ubuntu 16.04 what ever algo to drop under 1200Mhz at 160W

The only thing you're really testing is the effect of power limit. Test with power usage.
I have a 1080ti and Ubuntu 20.04 and I didn't notice any performance problems but I didn't change the power
limit and I don't use pre-built Linux binaries either.

[reb0rn21]

That does not change nothing if its prebuild, i have compiled miners too clock drop more then expected happen on them also

I am not sure of performance at default power limits as I never used any card on them...
I just tested it, as I have AORUS GeForce® GTX 1080 Ti Xtreme that are pre overclocked at 250w clock was ~1600Mhz mining kawpow

[JayDDee]

That does not change nothing if its prebuild, i have compiled miners too clock drop more then expected happen on them also

I am not sure of performance at default power limits as I never used any card on them...
I just tested it, as I have AORUS GeForce® GTX 1080 Ti Xtreme that are pre overclocked at 250w clock was ~1600Mhz mining kawpow

Without comparative data it's pretty hopeless to identify the exact cause of the discrepency.
But looking at the results at face value it's hard to conclude there is a problem on 20.04.
You reduced the power limit by about a third and the clock dropped by about a third.
There's nothing suspicious about that.

[Grim]

The clock drops to 500 Mhz for the MEMORY, right?

EDIT:
I just tried a low power setting with a 1080ti mining kawpow in win 10.

The card dropped to P-State 5 and locks the memory to 810 mhz. (202,5mhz x4)

[reb0rn21]

The clock drops to 500 Mhz for the MEMORY, right?

No memory don`t drop the core clock do
| NVIDIA-SMI 440.64 Driver Version: 440.64 CUDA Version: 10.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 108… On | 00000000:01:00.0 On | N/A |
| 80% 61C P2 163W / 160W | 2654MiB / 11178MiB | 100% Default |
±------------------------------±---------------------±---------------------+
| 1 GeForce GTX 108… On | 00000000:02:00.0 On | N/A |
| 80% 60C P2 163W / 160W | 2580MiB / 11178MiB | 100% Default |
±------------------------------±---------------------±---------------------+
| 2 GeForce GTX 108… On | 00000000:03:00.0 On | N/A |
| 80% 56C P2 102W / 160W | 2580MiB / 11178MiB | 100% Default |
±------------------------------±---------------------±---------------------+
| 3 GeForce GTX 108… On | 00000000:05:00.0 On | N/A |
| 80% 58C P2 163W / 160W | 2580MiB / 11178MiB | 100% Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 2389 G /usr/lib/xorg/Xorg 18MiB |
| 0 2492 G /usr/bin/gnome-shell 69MiB |
| 0 3480 C ./nbminer 2553MiB |
| 1 2389 G /usr/lib/xorg/Xorg 14MiB |
| 1 3480 C ./nbminer 2553MiB |
| 2 2389 G /usr/lib/xorg/Xorg 14MiB |
| 2 3480 C ./nbminer 2553MiB |
| 3 2389 G /usr/lib/xorg/Xorg 14MiB |
| 3 3480 C ./nbminer 2553MiB |
±----------------------------------------------------------------------------+
ru7@ru7-System-Product-Name:~$ nvidia-smi -i 0,1,2,3,4 --format=csv --query-gpu=clocks.mem,clocks.gr
clocks.current.memory [MHz], clocks.current.graphics [MHz]
5514 MHz, 658 MHz
5514 MHz, 658 MHz
5400 MHz, 544 MHz
5514 MHz, 683 MHz

[reb0rn21]

I did more digging and as soon i start GPU cude appz Sw power cap change to active
GPU 00000000:05:00.0
Performance State : P2
Clocks Throttle Reasons
Idle : Not Active
Applications Clocks Setting : Not Active
SW Power Cap : Active
HW Slowdown : Not Active
HW Thermal Slowdown : Not Active
HW Power Brake Slowdown : Not Active
Sync Boost : Not Active
SW Thermal Slowdown : Not Active
Display Clock Setting : Not Active

My gtx1070 and 1060 rigs SW cap is not active even I use PL limit too... strange

[rikuu]

That is only telling you what is the limiting factor preventing the GPU from clocking higher. There is always one limiting factor: Max clocks reached, max voltage reached, max temps reached, etc... When the GPU encounters one of these limits, it clocks itself down to stay within its determined limits.

In your chase, "Power Cap" is active, therefore the GPU thinks that it reached its power budget limit. You could verify this using a clamp amp meter to make sure that the GPU is doing the right thing.

Your 1070 and 1060 may each another limit before the power limit (like temps or frequency limit)

[reb0rn21]

Again you get card downclock to 550-600Mhz at 160W PL?
Ppl are mininng up to 20Mhs at kawpow at 160W and i get 11Mhs.......

Also its software power cap which i had no issue with 16.04

[JayDDee]

The only way to figure this out is to go back to 16.04 and do detailed control testing and gather lots of data.

Try different power limits, different algos, different mining SW so you get the full picture with both OSs.
Then you can compare the results and see where thing deviate.

Anything else is just idle speculation.

[reb0rn21]

But what to gather on 16.04 and 18.04 I have none power cap, only power cap is if I push 250w and temp hit 85C, also issue is with all miner its just depend of power demand they need from GPU

I was told to try older kernels <5.4, tested 5.3.18 same...

Its clean system so issue is not there, OC, PL all works OK issue is just sw pow cap activation, its even active at default PL but you don`t see huge clock drop as it burning 250w

[JayDDee]

But what to gather on 16.04 and 18.04 I have none power cap, only power cap is if I push 250w and temp hit 85C, also issue is with all miner its just depend of power demand they need from GPU

I was told to try older kernels <5.4, tested 5.3.18 same...

Its clean system so issue is not there, OC, PL all works OK issue is just sw pow cap activation, its even active at default PL but you don`t see huge clock drop as it burning 250w

I suggest you start fresh and forget everything so far, don't make any assumptions, test any theories.
Start with the hashrate and power usage with default settings. If they don't match the problem has nothing to
do with the power cap. Try different algos and miners to see if they match. Try different power caps to see if they match.

[reb0rn21]

I have man....... default work better but i never used card at 250W nor I want

200w PL is also bad as same speed as GTX1070 at 110w, LOL insane

there is defo some issue with PL limits and sw cap control with 1080ti and ubuntu 20.04

I am now mining some more light algo that kinda work but still it was not as it should be