Which is close enough to my current problem for me to reply to you in hopes there's a solution someplace...
This box is a Linux Mint v20.3 box that has two card PCIE slots. Before I ever got 'coolbits' / xorg working one of the cards fans gave up. While that card was out I got xorg & coolbits working while the box was single card. I got the replacement card in today, adjusted my xorg device entries, installed the card and with one minor BusID change (I still had prior one in) I got to my screen and ran my script to set things:
# card on top w/o monitor
/usr/bin/nvidia-smi -i 1 -pm 1
/usr/bin/nvidia-smi -i 1 -pl 190
# added 2nd Zotac 3070 5/24/2023, coolbits only functional on gpu0(Device0)
#
/usr/bin/nvidia-settings -a "[gpu:0]/GPUPowerMizerMode=1"
/usr/bin/nvidia-settings -a "[gpu:0]/GPUGraphicsClockOffset[4]=90"
/usr/bin/nvidia-settings -a "[gpu:0]/GPUMemoryTransferRateOffset[4]=138"
/usr/bin/nvidia-settings -a "[gpu:1]/GPUPowerMizerMode=1"
# /usr/bin/nvidia-settings -a "[gpu:1]/GPUGraphicsClockOffset[4]=90"
# /usr/bin/nvidia-settings -a "[gpu:1]/GPUMemoryTransferRateOffset[4]=138"
Last two lines were just ignored, so I commented them out to remind me for now.
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 1226 G /usr/lib/xorg/Xorg 146MiB |
| 0 N/A N/A 1755 G cinnamon 32MiB |
| 0 N/A N/A 5171 G ...gnu/webkit2gtk-4.0/WebKitWebProcess 26MiB |
| 0 N/A N/A 10050 G /usr/lib/firefox/firefox 97MiB |
| 0 N/A N/A 17585 C ...6_x86_64-pc-linux-gnu__BRP7-cuda102 714MiB |
| 0 N/A N/A 18953 C ..._64-pc-linux-gnu__opencl_nvidia_101 348MiB |
| 1 N/A N/A 1226 G /usr/lib/xorg/Xorg 4MiB |
| 1 N/A N/A 17663 C ...6_x86_64-pc-linux-gnu__BRP7-cuda102 714MiB |
+---------------------------------------------------------------------------------------+
All to show that coolbits is only applying to GPU0 (Device0). In xorg it is coded in the single Screen Section. I will run off now and try putting it on each device but I think I've been down this road w/o success. Any suggestions?
If you have installed a card after you ran the coolbits tweak, you need to rerun it. You should see thermal control sections for each card in the xorg.conf file in /etc/X11
I'd clear out the existing xorg.conf by copying back over the original xorg.conf.backup file of the original bare installation and then rerun the coolbits tweak.
1) Saw a post of yours from long ago and ordered a hdmi dummy plug. Arrives tomorrow.
2) There is a lot of good info in this thread. I've gotta make to time to read the 50 pages I haven't yet.
3) Mint v20.3 doesn't come with an etc/X11/xorg.conf file and we could never get it to start the Xserver with one placed there. It builds it on the fly from parts and pieces in usr/share/X11/xorg.conf.d. I'll call it a virtual xorg.conf until I can figure out where it writes it out at.
4) I took a backup so I can see what "sudo nvidia-xconfig --thermal-configuration-check --cool-bits=28 --enable-all-gpus" does to me on this box.
RESULTS: 4) above. Surprisingly I didn't get a black sceen and it built a legit etc/X11/xorg.conf It didn't get me coolbits control on GPU1 but I'm going to try adding the virtual display thing to the screen it hooked GPU1 (Device1) to. And try again tomorrow with hdmi dummy.
Mint is weird. Should behave exactly like Ubuntu and Debian which it is derived from.
But it doesn't a lot it seems.
I've never had any card not be enabled by the coolbits tweak. I've run as many as four at a time on hosts before with only one card connected to the monitor. And all got fan and thermal control.
And none of them needed a dummy plug. The only time I've needed a dummy plug was on headless SoC's to get video output for VNC or RDP sessions and to keep crunching on the embedded gpu in the SoC.
Mint is weird. Should behave exactly like Ubuntu and Debian which it is derived from.
But it doesn't a lot it seems.
I've never had any card not be enabled by the coolbits tweak. I've run as many as four at a time on hosts before with only one card connected to the monitor. And all got fan and thermal control.
And none of them needed a dummy plug. The only time I've needed a dummy plug was on headless SoC's to get video output for VNC or RDP sessions and to keep crunching on the embedded gpu in the SoC.
I have some homemade ones in a drawer because Windows required them for awhile but I'm not sure it does anymore. It's kinda like having to load the gpu drivers twice, once for each gpu, but not one installation takes care of both cards as long as they are both Nvidia. And for me it even works with a miner gpu and a fairly cheap 1gb card I use just for the display.
Mint is weird. Should behave exactly like Ubuntu and Debian which it is derived from.
But it doesn't a lot it seems.
I've never had any card not be enabled by the coolbits tweak. I've run as many as four at a time on hosts before with only one card connected to the monitor. And all got fan and thermal control.
And none of them needed a dummy plug. The only time I've needed a dummy plug was on headless SoC's to get video output for VNC or RDP sessions and to keep crunching on the embedded gpu in the SoC.
and if it doesn't get me there so be it. The top card (in standard atx mobo) is the one running a bit slower and it's the hotter one so I'll just live with it.
With about 135~150w current loading...
Every 11.0s: NV_Clocks.sh skip-MS7C91: Thu May 25 06:44:53 2023
I bought a EVGA 3080TI FTW3 for $650 on ebay which seems ~ reasonable ~ in this day and age. Seems like a sweet spot for compute/value with 10,000 compute cores.
Which is close enough to my
)
Which is close enough to my current problem for me to reply to you in hopes there's a solution someplace...
This box is a Linux Mint v20.3 box that has two card PCIE slots. Before I ever got 'coolbits' / xorg working one of the cards fans gave up. While that card was out I got xorg & coolbits working while the box was single card. I got the replacement card in today, adjusted my xorg device entries, installed the card and with one minor BusID change (I still had prior one in) I got to my screen and ran my script to set things:
Last two lines were just ignored, so I commented them out to remind me for now.
My clock checker at the moment shows:
nvidia-smi:
All to show that coolbits is only applying to GPU0 (Device0). In xorg it is coded in the single Screen Section. I will run off now and try putting it on each device but I think I've been down this road w/o success. Any suggestions?
Thanx, Skip
With coolbits and Thermal...
)
With coolbits and Thermal... stuff applied to the Device1 section:
Xorg.0.log shows:
What was your coolbits
)
What was your coolbits command line?
This is the one I use to have thermal and fan control set for all cards.
If you have installed a card after you ran the coolbits tweak, you need to rerun it. You should see thermal control sections for each card in the xorg.conf file in /etc/X11
I'd clear out the existing xorg.conf by copying back over the original xorg.conf.backup file of the original bare installation and then rerun the coolbits tweak.
1) Saw a post of yours from
)
1) Saw a post of yours from long ago and ordered a hdmi dummy plug. Arrives tomorrow.
2) There is a lot of good info in this thread. I've gotta make to time to read the 50 pages I haven't yet.
3) Mint v20.3 doesn't come with an etc/X11/xorg.conf file and we could never get it to start the Xserver with one placed there. It builds it on the fly from parts and pieces in usr/share/X11/xorg.conf.d. I'll call it a virtual xorg.conf until I can figure out where it writes it out at.
I hand built a 20-nvidia.conf there:
4) I took a backup so I can see what "sudo nvidia-xconfig --thermal-configuration-check --cool-bits=28 --enable-all-gpus" does to me on this box.
RESULTS: 4) above. Surprisingly I didn't get a black sceen and it built a legit etc/X11/xorg.conf It didn't get me coolbits control on GPU1 but I'm going to try adding the virtual display thing to the screen it hooked GPU1 (Device1) to. And try again tomorrow with hdmi dummy.
Xorg.0.log:
Mint is weird. Should behave
)
Mint is weird. Should behave exactly like Ubuntu and Debian which it is derived from.
But it doesn't a lot it seems.
I've never had any card not be enabled by the coolbits tweak. I've run as many as four at a time on hosts before with only one card connected to the monitor. And all got fan and thermal control.
And none of them needed a dummy plug. The only time I've needed a dummy plug was on headless SoC's to get video output for VNC or RDP sessions and to keep crunching on the embedded gpu in the SoC.
Keith Myers wrote: Mint is
)
I have some homemade ones in a drawer because Windows required them for awhile but I'm not sure it does anymore. It's kinda like having to load the gpu drivers twice, once for each gpu, but not one installation takes care of both cards as long as they are both Nvidia. And for me it even works with a miner gpu and a fairly cheap 1gb card I use just for the display.
Keith Myers wrote: Mint is
)
Well, I'll try the dummy plug with the
and if it doesn't get me there so be it. The top card (in standard atx mobo) is the one running a bit slower and it's the hotter one so I'll just live with it.
With about 135~150w current loading...
Skip
So far any price pressure the
)
So far any price pressure the rtx 4090 has provided doesn't seem to have driven the rtx 3090 ti price on eBay down much at all.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
I bought a EVGA 3080TI FTW3
)
I bought a EVGA 3080TI FTW3 for $650 on ebay which seems ~ reasonable ~ in this day and age. Seems like a sweet spot for compute/value with 10,000 compute cores.
3080Ti is certainly a sweet
)
3080Ti is certainly a sweet spot for Einstein.
_________________________________________________________________________