All things Nvidia GPU

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3080
Credit: 4978237686
RAC: 1468955

Keith, With your script

Keith,

With your script for Nvidia GPUs, I've decided to take a look at my lonely settings in NVIDIA X Server Settings for my 3080Ti - and they're not working!  My GPU is running cool at 62॰C with a full load of E@H, but I notice that though I set the fan speeds for 'whatever' and hit APPLY, it still shows "Failed to set new Fan Speed!"  NVIDIA-SMI shows GPU fan speed as 58% when I've set the NEW fan speed to 75% (or so I thought).

When I go to the PowerMizer and set new settings for Clocks and Memory Transfer it sets when I hit 'Enter' but the Clocks and Memory Transfer don't change and it still reads "Failed to set new Fan Speed!" in the lower left.

Do I still need to get Coolbits?  Or should I reload Nvidia X Server Settings? 

George

Proud member of the Old Farts Association

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4977
Credit: 18787829256
RAC: 7479480

Persistence mode is not used

Persistence mode is not used for performance.  It is used to maintain the card settings if you stop BOINC and restart it.

If you are happy with how the cards run automatically in P2 reduced performance state then run then that way. The cards will always run in Power State P2.

You don't have to do anything.

Power Mizer does not control power.  It only controls whether you are allowing the card to handle things itself automatically.  PowerMizer mode=1 is "Prefer max performance" in the Nvidia X-Server Settings app choose dialog at the bottom.  Prefer max performance tells the card to boost as high and as often as the temps and power limits allow and keep them there.

You don't have any control over the power state with a compute load on the card.  So all cards are deliberately hamstrung by the drivers to run with reduced clocks on the memory.  The penalty can be as high as 2000Mhz or as little as 600Mhz depending on the card generation.

That affects crunching speed directly, a lot or not so much depending on the project's gpu apps.

The whole point of the script is to push the cards back to an "effective" P0 state while crunching.  It sets the card up with the clocks the card should and would be running if the card was running a game.  The clocks the card runs at for all the publicized specs on the box or the specs all the gaming review sites mention.

 

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4977
Credit: 18787829256
RAC: 7479480

You've already forgotten how

You've already forgotten how you fixed that once already.  It is a permission issue.  I assume you reloaded the OS and lost your setup.

Yes, you need the coolbits tweak to enable the the fan and clock settings to be set manually.

Yes, you need to enable permission to set the fan speeds.

So first, look at your /etc/X11/xorg.conf file and see if you have thermal control set for the card. You need to find this in your Screen section.

Option         "ThermalConfigurationCheck" "True"
    Option         "Coolbits" "28"

If you don't see that then you need to run the coolbits tweak and reboot.  Then you will be able to set clocks and fan speeds.

Because you are on 22.04 now you need to change the permissions of the xwrapper. Here is the link to remind you.

Xwrapper config file changes needed for fan control in Ubuntu 22.04.1

edit the /etc/X11/Xwrapper.config file with nano

add 

allowed_users=anybody

needs_root_rights=yes

below the existing 

allowed_users=console entry

Save the file changes and reboot.  You will be able to set fan speeds now.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6474
Credit: 9602338290
RAC: 5898566

I was looking through the

I was looking through the listing(s) of various Nvidia gpus.

Would someone point me at the model that crunches the fastest for the type of crunching we do at e@h?

This is a "price is no object" type of question.

A most bang per video card type of question.

I understand I can't afford one or more of them but...

Tom M

 

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

mikey
mikey
Joined: 22 Jan 05
Posts: 12711
Credit: 1839114599
RAC: 3624

Tom M wrote: I was looking

Tom M wrote:

I was looking through the listing(s) of various Nvidia gpus.

Would someone point me at the model that crunches the fastest for the type of crunching we do at e@h?

This is a "price is no object" type of question.

A most bang per video card type of question.

I understand I can't afford one or more of them but...

Tom M

I would guess the easy answer is the latest ones ie 4090 Nvidia is selling them for about $1100US

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3971
Credit: 47341292642
RAC: 65957756

the 4090 is fast. but suffers

the 4090 is fast. but suffers from memory bandwidth bottlenecks. so it's only a little faster than the 3080Ti or 3090 in reality.

"Price is no object" winner would likely be the H100, if you've got about $30,000 laying around, or $100+k since you'll probably have to buy a whole system with several of them in it.

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6474
Credit: 9602338290
RAC: 5898566

Thank you. That is what I

Thank you. That is what I needed so I could properly lust after an Nvidia GPU upgrade.

Tom M is

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Boca Raton Community HS
Boca Raton Comm...
Joined: 4 Nov 15
Posts: 244
Credit: 10650835586
RAC: 17053189

We will be able to see how

We will be able to see how the 4090 does in reality very soon. We have two on order right now that will be going into the student built systems that will be running Ubuntu. I was given an ETA of a few weeks from the vendor but we will see if that's true in a few weeks... 

These will be the student's first Ubuntu systems so it should be fun. 

mikey
mikey
Joined: 22 Jan 05
Posts: 12711
Credit: 1839114599
RAC: 3624

Boca Raton Community HS

Boca Raton Community HS wrote:

We will be able to see how the 4090 does in reality very soon. We have two on order right now that will be going into the student built systems that will be running Ubuntu. I was given an ETA of a few weeks from the vendor but we will see if that's true in a few weeks... 

These will be the student's first Ubuntu systems so it should be fun.  

WOO HOO!!!

mikey
mikey
Joined: 22 Jan 05
Posts: 12711
Credit: 1839114599
RAC: 3624

Tom M wrote: Thank you. That

Tom M wrote:

Thank you. That is what I needed so I could properly lust after an Nvidia GPU upgrade.

Tom M

Tom have you seen this at EBAY:

NVIDIA Tesla T4 16GB GDDR6 PCIE 3.0 x16 GPU Graphics Accelerator Card

Opens in a new window or tab

Refurbished · NVIDIA · NVIDIA Tesla T4 · 16 GB

$999.00

Top Rated Plus

or Best Offer

Free 4 day shipping

Free returns

I have NO IDEA of the credibility of the seller so this is a 'did you see what I saw thing' not an endorsement of anything else. Don't the Tesla T4's work VERY well here?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.