Welcome to MilkyWay@home

NVIDIA Tesla T4

Questions and Answers : Unix/Linux : NVIDIA Tesla T4
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile PhilTheNet

Send message
Joined: 24 Sep 14
Posts: 4
Credit: 50,712,386
RAC: 0
Message 75295 - Posted: 9 Apr 2023, 6:12:39 UTC

Hello,
The "NVIDIA Tesla T4" are no longer recognized by Milkyway, although they worked fine until recently (and work with other programs like NumbersFields).

GenuineIntel Intel(R) Xeon(R) CPU @ 2.30GHz [Family 6 Model 63 Stepping 0] (2 processeurs)
NVIDIA Tesla T4 (4095MB) driver: 525.85
Linux Ubuntu Ubuntu 20.04.5 LTS [5.10.147+|libc 2.31 (Ubuntu GLIBC 2.31-0ubuntu9.9)]

Thks
ID: 75295 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[AF>EDLS]zOU

Send message
Joined: 31 Mar 08
Posts: 22
Credit: 84,159,673
RAC: 0
Message 75298 - Posted: 9 Apr 2023, 11:29:02 UTC - in response to Message 75295.  

Does Boinc still detects it ?
Do you have any explicit message in Boinc from the project ?

Failing that information it's all going to be conjecture unless a recent change in project code/app removed support.
But you would get a project message I expect.

None of the apps has been updated since 2019.

https://milkyway.cs.rpi.edu/milkyway/apps.php
ID: 75298 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[AF>EDLS]zOU

Send message
Joined: 31 Mar 08
Posts: 22
Credit: 84,159,673
RAC: 0
Message 75299 - Posted: 9 Apr 2023, 11:31:27 UTC

Also, your computer shows driver 525.85.

Nvidia website shows this:
Data Center Driver For Linux X64

Version: 460.106.00
Release Date: 2021.10.26
Operating System: Linux 64-bit
CUDA Toolkit: 11.2
Language: English (US)
File Size: 171.61 MB
ID: 75299 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3326
Credit: 521,770,357
RAC: 55,081
Message 75301 - Posted: 9 Apr 2023, 13:13:29 UTC - in response to Message 75299.  
Last modified: 9 Apr 2023, 13:14:49 UTC

Also, your computer shows driver 525.85.

Nvidia website shows this:
Data Center Driver For Linux X64

Version: 460.106.00
Release Date: 2021.10.26
Operating System: Linux 64-bit
CUDA Toolkit: 11.2
Language: English (US)
File Size: 171.61 MB


I have an Nvidia P104 mining gpu and it curnches just fine at PrimeGrid, Einstein and Ammicable Numbers, I haven't tried it here yet,it's also an older gpu and Nvidia also picked the 470 series of drivers to install when it first found it. However I upgraded the Nvidia software to the latest version 531.68 and it still works great, meaning I'm not sure if the Nvidia driver version means alot or the changes from 470 to 531 haven't been enough to kick it off line. I'm using Windows and since the OP is using Linux 525 version is the one they say is the latest in their Drivers section when you try to update that way. Yes you can update manually but that's another topic unless you want to go there.
ID: 75301 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Joseph Stateson
Avatar

Send message
Joined: 18 Nov 08
Posts: 291
Credit: 2,461,693,501
RAC: 0
Message 75303 - Posted: 9 Apr 2023, 13:38:52 UTC - in response to Message 75295.  
Last modified: 9 Apr 2023, 13:51:49 UTC

I just had my 20.04 Ubuntu drop my gtx-1060 after a restart of BOINC. Coincidence that this happened minutes after reading your post here. I came right back to add my 0.02c opinion.

I had RX-570 and GTX-1060 running just fine. I made a minor change in cc_config..xml and restarted BOINC minutes ago. System only shows the AMD card now.

I ran clinfo and the Nvidia is not reported.

I ran nvidia-smi and there is a library conflict. I am guessing I did an update some time ago and got a new library that Nvidia does not like after restarting BOINC.

root@dual-linux:~# nvidia-smi
Failed to initialize NVML: Driver/library version mismatch


Your Tesla has the same gpu chip, TU104, as RTX-2080 super according to tech power up. That new driver should have worked but then, this is Ubuntu which sucks



[edit] This was fixed by running "upgrade" and rebooting
ID: 75303 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
San-Fernando-Valley

Send message
Joined: 13 Apr 17
Posts: 256
Credit: 604,411,638
RAC: 0
Message 75305 - Posted: 9 Apr 2023, 15:14:37 UTC - in response to Message 75295.  

This is what I found at NVIDIA support (probably looked wrong - I really don't know):

Search
Name Version Release Date CUDA Toolkit
Data Center Driver for Ubuntu 20.04 NVIDIA Certified 525.105.17 March 30, 2023 12.0
Release Highlights:
Release notes, supported GPUs and other documentation can be found at:
https://docs.nvidia.com/datacenter/tesla/index.html
Data Center Driver for Ubuntu 20.04 NVIDIA Certified 515.105.01 March 30, 2023 11.7
Data Center Driver for Ubuntu 20.04 NVIDIA Certified 470.182.03 March 30, 2023 11.4
Data Center Driver for Ubuntu 20.04 NVIDIA Certified 450.236.01 March 30, 2023 11.0
Data Center Driver for Ubuntu 20.04 NVIDIA Certified 525.85.12 January 30, 2023 12.0

Just my unprofessional thought:
... if you have been using the 525.85.xxx driver, then why not try the 525.105.xxx ?
cheers
ID: 75305 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile PhilTheNet

Send message
Joined: 24 Sep 14
Posts: 4
Credit: 50,712,386
RAC: 0
Message 75313 - Posted: 10 Apr 2023, 9:20:07 UTC

I think I found it, that would be OpenCL: 3.0 not recognized
I did a test with the :
apt update
apt install nvidia-driver-525
apt install -y clinfo
clinfo
and now the OpenCL is recognized and Milkyway recomputes with the GPU

I will test on other Ubuntu machines

:)
ID: 75313 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3326
Credit: 521,770,357
RAC: 55,081
Message 76237 - Posted: 11 Jul 2023, 11:46:17 UTC - in response to Message 75313.  

I think I found it, that would be OpenCL: 3.0 not recognized
I did a test with the :
apt update
apt install nvidia-driver-525
apt install -y clinfo
clinfo
and now the OpenCL is recognized and Milkyway recomputes with the GPU

I will test on other Ubuntu machines

:)


WOO HOOI!!!
ID: 76237 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : NVIDIA Tesla T4

©2024 Astroinformatics Group