Hey there @julioleon! I just bought the pieces for a 3090ti Flame build. I decided to go for a VGA Supernova 1600p power supply. Yes, 1600 watts! Between the GPU, CPU (Intel 12900k) and all the cooling needed for it. a 750-watt PS isn’t going to be enough. Here’s the Amazon link for it:
Definitely go more than 750w. There is a problem(?) with the 30 series GPUs where they have some crazy high transient loads. This means that for just a moment the GPU can draw 500w or something. If your power supply can’t handle that it will trip the overcurrent protection and turn off the machine.
From my experience, Flame isn’t that likely to produce a load llke that but if you are using other software it might. I ran into this issue when doing some Unreal renders on a 750w PSU which had been totally fine in Flame for months beforehand.
Hey Julioleon, I have a 1200W power supply for my system. I used to have a 3090 and a 2080TI in the same system until I swapped out my 2080TI into my unraid server so I wanted to make sure I had enough power. I agree 750 is kind of pushing the limit for what a 3090 system should have for power.
What version of the drivers do you run now on Rocky Linux 8.5? Could you please provide step by step instructions on how to correctly update the drivers? Thanks!
Please refer to this post where I have written instructions: Symbol lookup error: /lib64/libGL.so.1: undefined symbol: _glapi_tls_Current - flame install with RTX 3080 - #3 by friendlyape
I am currently using the default DKU ones as they are what work for me, but I have validated up to 495.46
I believe it was @friendlyape that gave me the tip but I just install Rocky Linux and then install the DKU 17.0 and personally I just installed the most recent nvidia driver from their website after the DKU installs.
Just keep in mind you need to install the new GPU driver in text mode so you’re not loaded into the Linux desktop with “ systemctl isolate multi-user.target” and use the install flag “ --install-libglvnd” when installing the gpu driver. So ./NVIDIA-xxxx.sh --install-libglvnd
This is what I did and it has been working for me but my system may be different from yours.
thanks a lot - I am looking at nvidia drivers page and 495.46 - Release Date: 2021.12.13 looks like the latest driver in the new feature branch as for production branch that number is 515.48.07 - Release Date: 2022.5.31.
Production Branch Production Branch drivers provide ISV certification and optimal stability and performance for Unix customers. This driver is most commonly deployed at enterprises, providing support for the sustained bug fix and security updates commonly required.
New Feature Branch New Feature Branch drivers provide early adopters and bleeding edge developers access to the latest driver features before they are integrated into the Production Branches
I used production but that’s just me.
I am getting error with that install flag ERROR: Invalid commandline, please run --help for usage info
As root in multi-user mode i use: ./NVIDIA-Linux-x86_64-495.46.run --install-liglvnd
Verifying integrity…OK
Uncompressing NVIDIA bla bla…
Then the error above
Just noticed that your original post has mistake in the command
yea sorry I edited the post, thought it did it fast enough that you would see the update.
ok I am up and running with the new driver but I noticed that nvidia control app did not install hmmm? looks like the desktop and especially browser playing videos inside is not choppy with original driver I had this issue now that is gone.
I guess settings app is installed just not icons - its all running fine now thanks! Also 15-20 frames better RAW performance for CUDA on this latest driver
this original post has that mistake in the flag command - it should be --install-libglvnd
maybe update that original post there
Good catch! not sure how I made the mistake twice haha. updated.
Looking at the dates of the production vs new feature drivers is it safe to assume that all new featured by now have propagated to the production driver?
Be very careful when not using the NVIDIA driver part of DKU. There are very good reasons we qualify drivers and not use latest and greatest.
understood! Yea I used the newer driver understanding there are risks but I needed access to optix rendering in Blender and the DKU driver did not enable that for me.
I would then use dual boot option since we are aware of differences in our products between the driver part of DKU and the latest available driver.
Could you go into more detail what those differences are I guess it has to do with quadro cards only? Many use RTX 3090 and I guess that never gets tested anyways.
I cannot provide you more details about the issue but like I wrote, sticking with the driver part of the DKU is the way to go to avoid issues.
We do not qualify gaming card so hard to say if they have the same issues with the latest version of the driver.