olealgoritme / gddr6 Goto Github PK

Linux based GDDR6/GDDR6X VRAM temperature reader for NVIDIA RTX 3000/4000 series GPUs.

C 77.88% CMake 10.12% Shell 4.54% Nix 7.46%

gddr6's Introduction

GDDR6/GDDR6X GPU Memory Temperature Reader for Linux

Reads GDDR6/GDDR6X VRAM memory temperatures from multiple supported NVIDIA GPUs found in a host Linux system. These findings are based on reverse engineering of the NVIDIA GPU Linux driver.

Prerequisites

Kernel boot parameter: iomem=relaxed

sudo vim /etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash iomem=relaxed"
sudo update-grub
sudo reboot

Disabling Secure Boot

This can be done in the UEFI/BIOS configuration or using mokutil:

mokutil --disable-validation

Check state with:

$ sudo mokutil --sb
SecureBoot disabled

Dependencies

libpci-dev

sudo apt install libpci-dev -y

Installation (cmake)

git clone https://github.com/olealgoritme/gddr6
cd gddr6
./build_install.sh
sudo gddr6

Supported GPUs

RTX 4090 (AD102)
RTX 4080 Super (AD103)
RTX 4080 (AD103)
RTX 4070 Ti Super (AD103)
RTX 4070 Ti (AD104)
RTX 4070 Super (AD104)
RTX 4070 (AD104)
RTX 3090 Ti (GA102)
RTX 3090 (GA102)
RTX 3080 Ti (GA102)
RTX 3080 (GA102)
RTX 3080 LHR (GA102)
RTX 3070 (GA104)
RTX 3070 LHR (GA104)
RTX A2000 (GA106)
RTX A4500 (GA102)
RTX A5000 (GA102)
RTX A6000 (AD102)
L4 (AD104)
L40S (AD102)
A10 (GA102)

gddr6's People

Contributors

Stargazers

Watchers

gddr6's Issues

RTX 6000 ADA

A suggestion to add RTX 6000 ADA support using the following line:

{ .offset = 0x0000E2A8, .dev_id = 0x26B1, .vram = "GDDR6", .arch = "AD102", .name = "RTX 6000 ADA" },

How do one deteran the Memory Mapping and VRAM Temperature Reading?

Hi. I have A5000 and 5000ADA' how would one get the offset?

support for 3080 LHR

Add

    { .offset = 0x0000E2A8, .dev_id = 0x2216, .vram = "GDDR6X", .arch = "GA102", .name =  "RTX 3080 LHR" },

for support of LHR versions. Tested, works.

Error at line 57, file /home/username/gddr6/lib/src/gddr6.c (1) [Operation not permitted]

the code fails at this point everytime I try to run it

Random output on RTX 3070 (FE) on driver 535.113.01

Here are the reported temps on my Nvidia RTX 3070 (FE) running on driver version 535.113.01

Device: RTX 3070 GDDR6 (GA104 / 0x2484) pci=1:0:0
VRAM Temps: |  24°c | @ 0x0000ee50
VRAM Temps: |  43°c | @ 0x0000ee50
VRAM Temps: |  26°c | @ 0x0000ee50
VRAM Temps: |  56°c | @ 0x0000ee50
VRAM Temps: |  90°c | @ 0x0000ee50
VRAM Temps: |   9°c | @ 0x0000ee50
VRAM Temps: |   1°c | @ 0x0000ee50
VRAM Temps: |  50°c | @ 0x0000ee50
VRAM Temps: |  85°c | @ 0x0000ee50
VRAM Temps: |   4°c | @ 0x0000ee50
VRAM Temps: | 118°c | @ 0x0000ee50
VRAM Temps: |  36°c | @ 0x0000ee50
VRAM Temps: |  71°c | @ 0x0000ee50
VRAM Temps: |  78°c | @ 0x0000ee50
VRAM Temps: |  69°c | @ 0x0000ee50
VRAM Temps: |  94°c | @ 0x0000ee50

Seems like a random number between 0 and 127.
I modified so it shows the .offset it reads from and not flush the output.

Only shows one GPU temp when multiple GPUs in system

Hi, really sorry to ask for more, but would it be possible to observe the vram temps of multiple gpus? In my dual gpu setup (for ai/ml work), it appears if there is gpu 0 and gpu 1, it only pulls the vram temps of gpu 1. I will look at the code and see if I can make the changes myself, but thought I might as well plug the request in here as well. Thanks in advance, in case you have time to figure it out!

Support for 3090 Ti

I have two 3090 Ti FEs in my workstation and was trying to use this tool but it reports "No compatible GPU found". I see that 3090 Ti is not in the list of supported GPUs. Can we make it work for 3090 Ti?

Multi Gpu Order and coresponding temperatures wrong

Looks like the order of the output is reversed compared to nvidia-smi output.

Here is the output of gddr6:

user@test_server:~/tmp/vramtest/gddr6$ sudo ./gddr6
Device: RTX A2000 GDDR6 (GA106 / 0x2531) pci=8:d:0
Device: RTX A2000 GDDR6 (GA106 / 0x2531) pci=6:1b:0
Device: RTX A2000 GDDR6 (GA106 / 0x2531) pci=6:10:0
VRAM Temps: | 76°c | 74°c | 70°c |

And here is the output of nvidia-smi:

Thank you for you help.

A5000 Mobile

Requesting help on adding the A5000 mobile. I sent you a friend request on discord