BOINC crash & destroy GPU

Questions and Answers : Unix/Linux : BOINC crash & destroy GPU
Message board moderation

To post messages, you must log in.

AuthorMessage
DanHansen@Denmark
Volunteer tester
Avatar

Send message
Joined: 14 Nov 12
Posts: 194
Credit: 5,881,465
RAC: 0
Denmark
Message 1419646 - Posted: 24 Sep 2013, 11:56:47 UTC

Hi,

Can this be true? I'm running several linux servers only doing the BOINC client. on one computer the graphic card has been destroyed twice!

My question is, if it's a computer with onboard graphic card, is it better to run BOINC without GPU due to the fact that there's no cooling of the GPU on most motherboards!?!?
If this is the case, then how do I set it up, not to ue the GPU?
I now that I can set it up in here, on my account. But I'm running only the BOINC client on my Linux computers and I noticed different ways to handle the setups.
With the boinccmd tool I can "--set_gpu_mode" but I haven't been able to make this work.
Then there's "--read_global_prefs_override" and "--read_cc_config". I simply don't know the priority of these settings and which command overrules which.
I'm sorry if I'm talking b....... here, but what I really need is someone to set me straight and tell me how to:
1. use the boinccmd tool to stop the BOINC-client using GPU on a computer.
2. where and how to setup some standard rules which new computers will follow, if possible. I'm not sure, but I read something about setting up the computers to follow settings from the SETI@home account.

I tried BAM, but this went totally wrong due to the fact that I created an account but allready had one. Then I coulden't gather the computers to the account I wanted. I tried everything, but this issue is wellknown in the forum, so I stopped using BAM..

Hope to hear from some of you guru's out there...

Kind Regards,
Dan Hansen
Denmark
ID: 1419646 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22184
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1419667 - Posted: 24 Sep 2013, 13:54:04 UTC

If you don't want to use the GPU then the simplest way is to use your account preferences and set "Use GPU type xxx" to "No".
For example, the PC I'm working on just now has an ATI GPU built into the motherboard and an Nvidia card, so I've set "Use ATI GPU" to "No", and the "Use Nvidia GPU" to "Yes".
Looking at the computers on your account none are using any form of GPU, so either there aren't any fitted, or there is one there, but BOINC has determined that its not suitable for S@H, or you've already disabled GPU processing.


Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1419667 · Report as offensive
DanHansen@Denmark
Volunteer tester
Avatar

Send message
Joined: 14 Nov 12
Posts: 194
Credit: 5,881,465
RAC: 0
Denmark
Message 1419938 - Posted: 25 Sep 2013, 2:56:53 UTC

Hi Rob,

Thanks for your response ;)

So in a way, you agree with me to set onboard GPU to "No"!?!? ;) And yes, that was my idea too, to set the computers with "real" graphic card with active coolers to use GPU as well.. And thank you for that ;)

1.
When you say, that none of my computers use GPU, can this be because of the settings I made recently? I recently made the setup shown beneath, yesterday i think it was*. Is this why you can't see that the graphic card in this computer?
http://setiathome.berkeley.edu/show_host_detail.php?hostid=6981880
*
Use CPU
Enforced by version 6.10+ yes
Use ATI GPU
Enforced by version 6.10+ no
Use NVIDIA GPU
Enforced by version 6.10+ no

Because, 2 Nvidia cards, with only ordinary cooling ribs (no fan) died ;) And when I uninstalled the second card, it was burning hot.

So it is "http://setiathome.berkeley.edu/prefs.php?subset=global" I use to set this up, right!?! Not the "Global setting". But, will the computer adjust to this setting? When I run "boinccmd --get_cc_status" this is what I get:

root@mycomputer:/myhomedir/username# boinccmd --get_cc_status
network connection status: don't need connection
CPU status
not suspended
current mode: according to prefs
perm mode: according to prefs
perm becomes current in 0 sec
GPU status
not suspended
current mode: according to prefs
perm mode: according to prefs
perm becomes current in 0 sec
Network status
not suspended
current mode: according to prefs
perm mode: according to prefs
perm becomes current in 0 sec


Isn't it suppose to show it here? Show that the GPU status is set to "NO" ???

2.
If we setup rules this way, then the computer is suppose to follow this setup from next time it communicates with the the server at SETI/Asteroids etc.(updating), right? Then what happenes if I change settings locally on one computer running this command?:

boinccmd --set_gpu_mode never (set GPU run mode for given duration)

Will the computer follow this setup always? Or, will it change when communicating with the the server at SETI/Asteroids etc.(updating)
I've never been able to really understand the rules! And this is why I'm begging for somebody to please show me how ;)

Then there's the "global setting" Here we are told:
http://setiathome.berkeley.edu/prefs.php?subset=global

These apply to all BOINC projects in which you participate.
On computers participating in multiple projects, the most recently modified preferences will be used.


Suspend GPU work while computer is in use?
Enforced by version 6.6.21+ yes


Again, there's a new way to setup rules for the computers and the GPU.
I can see that this is for setting up "when" to use GPU and not activating or deactivating it. Never the less, it's pretty confusing.

3.
So, we have:
A.: The settings made directly on the host/computer using "boinccmd tool"
B.: "--read_global_prefs_override" which tell the core client to read the [PrefsOverride global_prefs_override.xml] file.
C.: Global settings on the server at e.g. SETI http://setiathome.berkeley.edu/prefs.php?subset=global
D.: Preferences/Project preferences at e.g. SETI http://setiathome.berkeley.edu/prefs.php?subset=project

Well, I would so very much like to find a site/place where all the rules are explained. And especially how the priorities are. Which rule is used, is it just the latest modified or is there some kind of "master" setting?

I hope that my explaination is understandable! I'm doing my very best to show what my problem is.

4. Last question, and this I guess is a easy one for the guru ;)
Will performance be better/will the processing of data increase, if a Pentium4/Linux Computer has 2Gb Ram instead of 500Mb? Is Ram an issue for the boinc-client?


I'm looking so much forward to hear from you. Very much hoping I didn't "scare you of" ;)


Kind Regards,
Dan Hansen
Denmark

Computer Systems:
INTEL Pentium4/Ubuntu Server 12.04.2
BOINC-client v.7.0.27, v.7.2.14, v.7.2.27
ID: 1419938 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 1420159 - Posted: 25 Sep 2013, 18:29:24 UTC - in response to Message 1419646.  

Can this be true? I'm running several linux servers only doing the BOINC client. on one computer the graphic card has been destroyed twice!


Yes, it can be true. SETI@home - as well as any other distributed computing application - tends to push the hardware to it's limits. Increased usage of the hardware translates directly into increased heat production. If you do not have adequate cooling for your components, they will experience an much shorter lifespan.

However, GPU computing has been a boon for people interesting in crunching as much as they can. No CPU can reach the performance of a GPU. Though, as Rob said, since none of your computers are showing crunching-capable GPUs recognized by BOINC, it is doubtful that BOINC directly destroyed your GPUs if it isn't pushing them.

The only way I can see BOINC indirectly destroying your GPUs is if the CPUs are making the inside of the case so hot (again, back to inadequate cooling), then I can see your GPUs getting burned out. The obvious solution is to make sure you have good enough cooling to keep the temperature in your cases down to a reasonable level. Keep an eye on CPU and case temps.

When you say, that none of my computers use GPU, can this be because of the settings I made recently? I recently made the setup shown beneath, yesterday i think it was*. Is this why you can't see that the graphic card in this computer?


No. Even if you set those options in your preferences, BOINC would still be able to detect your GPUs and display them in your account.

The more likely reason is that either A) your GPUs are not crunching-capable so BOINC doesn't detect them (and hence couldn't have destroyed them) or B) you are not loading the OpenGL component needed for GPU crunching on Linux. (Sorry, I don't know how to add OpenGL for Linux as I don't use Linux.)

Will performance be better/will the processing of data increase, if a Pentium4/Linux Computer has 2Gb Ram instead of 500Mb? Is Ram an issue for the boinc-client?


SETI@home launches one app per logical CPU in your system (e.g. a Pentium 4 with HyperThreading will have two logical CPUs). Each SETI@home app uses about 64-96MB of RAM. SO if you have a Pentium 4 w/HT, you'll be using approximately 192MB of RAM total. Having more RAM does not help crunching performance at all.
ID: 1420159 · Report as offensive
DanHansen@Denmark
Volunteer tester
Avatar

Send message
Joined: 14 Nov 12
Posts: 194
Credit: 5,881,465
RAC: 0
Denmark
Message 1420523 - Posted: 26 Sep 2013, 14:16:25 UTC

Hello Americano ;)

Thank you for responding..

Yes, it can be true. SETI@home - as well as any other distributed computing application - tends to push the hardware to it's limits. Increased usage of the hardware translates directly into increased heat production. If you do not have adequate cooling for your components, they will experience an much shorter lifespan.


OK, I see.. All the new computers I use are RACK-mouted in 2U cases with 4 8x8 fans and the CPU cooler is a special 2U copper cooler with fan/active. Even though the standard intel cpu cooler fits, I decided to do this properly. All is industrial* hardware for a server running linux/boinc-client and nothing else.. I'm just not buying new CPU's, but motherboards, ram, fans, psu, cases etc. is new hardware. Like the idea to run those old cpu's "all the way to the grave" ;)
It's just that I have 1 old standard desktop computer left, running boinc as well and this is the one which killed 2 graphic cards. I think you are right about the heat being the reason, because, even though I have 2 ekstra fans mounted in that case, its pretty hot according to LM-sensors. And when this computer's graphic card is not fitted with a active cooler, the cooling ribs will make the card same temp. as inside the case. And inside this case its hot because of the CPU running 100%. So I think you are pretty right about this..

No. Even if you set those options in your preferences, BOINC would still be able to detect your GPUs and display them in your account.


Thanks, for that one - I nneded to hear that. Haven't been able to find that info anywhere..

SETI@home launches one app per logical CPU in your system (e.g. a Pentium 4 with HyperThreading will have two logical CPUs). Each SETI@home app uses about 64-96MB of RAM. SO if you have a Pentium 4 w/HT, you'll be using approximately 192MB of RAM total. Having more RAM does not help crunching performance at all.


OK, now we are getting somewhere! So I dont need to fill up the RAM blocks. I have minimum 500Mb RAM, but it was my plan to use 2Gb. Well, that's not necessary I can see.. nice!
I installed SSD in all the new RACK-mounted computers to get maximum performence.
But, there's one thing I need to ask you.
e.g. a Pentium 4 with HyperThreading will have two logical CPUs

Which P4 is this? which model? Is it from e.g. those P4 "Extreme" CPU's??? If it is, yoiu just made my day! I just bought about 10 P4's which fit the socket 478 and more than 6 of them are "Extreme" 3.0 and above. I'm so much looking forward to hear your answer, because if this is the case, something is about to happen in the RACK ;)

However, GPU computing has been a boon for people interesting in crunching as much as they can. No CPU can reach the performance of a GPU.


OK! I see! Well, this changes everything then! Do you think boinc will accept it, if I install a Low Profile APG graphic card? I've been looking for a not so pricy* 2U graphic card. This is possible because of the 2U RACK cases I use - they are designed to have a Low Profile APG card installed. Like this one:
http://www.proshop.dk/Grafikkort/ASUS-GeForce-GT-610-2GB-DDR3-2342329.html
I'm just having trouble finding the Low Profile Bracket ;) But it would really make the "crunching" a lot better, right? Is a GPU really better than a CPU ?, I didn't know that!?!?

Right now I'm installing the Xvision cooling controller from "???" - I don't remember the label/manufacturer. But there's a little problem. The 4 fans in the case are all connected to a little PCB/circuit board* which has a old/normal type 4-pin plug/connector - and this the controller can't control. I think the solution to that issue is to unplug the 4 fans from the PCB/circuit board and connect them directly to the controller. The fan plugs/connectors are still there...

Well, this was all ;) Hope to hear from you asap...

Again, thank you for all your help!

Kind Regards,
Dan Hansen
Denmark
ID: 1420523 · Report as offensive
Profile arkayn
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4438
Credit: 55,006,323
RAC: 0
United States
Message 1420575 - Posted: 26 Sep 2013, 16:25:41 UTC - in response to Message 1420523.  



OK! I see! Well, this changes everything then! Do you think boinc will accept it, if I install a Low Profile APG graphic card? I've been looking for a not so pricy* 2U graphic card. This is possible because of the 2U RACK cases I use - they are designed to have a Low Profile APG card installed. Like this one:
http://www.proshop.dk/Grafikkort/ASUS-GeForce-GT-610-2GB-DDR3-2342329.html
I'm just having trouble finding the Low Profile Bracket ;) But it would really make the "crunching" a lot better, right? Is a GPU really better than a CPU ?, I didn't know that!?!?



Kind Regards,
Dan Hansen
Denmark


As far as I know, there are no current CUDA cards that use the AGP interface. The card you linked to is a PCIE 2.0 card.

ID: 1420575 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 1420600 - Posted: 26 Sep 2013, 17:48:46 UTC - in response to Message 1420523.  
Last modified: 26 Sep 2013, 18:12:49 UTC

But, there's one thing I need to ask you.
e.g. a Pentium 4 with HyperThreading will have two logical CPUs

Which P4 is this? which model? Is it from e.g. those P4 "Extreme" CPU's??? If it is, yoiu just made my day! I just bought about 10 P4's which fit the socket 478 and more than 6 of them are "Extreme" 3.0 and above. I'm so much looking forward to hear your answer, because if this is the case, something is about to happen in the RACK ;)


To be clear, Intel's HyperThreading is not two real CPUs. It splits the resources of a single CPU and presents it as two CPUs to the OS. As far as the OS is concerned, any CPU with HyperThreading looks like the chip has two CPUs.

As far as which models have HyperThreading: it depends on what revision core they have, not specific speeds. The first revision of the Pentium 4, known as Willamette, did not have HyperThreading. All Pentium 4s from the next revision core, known as Northwood on to the Prescott and Cedar Mill (the last Pentium 4) all supported HyperThreading.

The first Pentium 4 Extreme Edition was known as Gallatin, and it too supported HyperThreading, as well as every Pentium 4 Extreme Edition and Pentium Extreme after it.

However, GPU computing has been a boon for people interesting in crunching as much as they can. No CPU can reach the performance of a GPU.


OK! I see! Well, this changes everything then! Do you think boinc will accept it, if I install a Low Profile APG graphic card? I've been looking for a not so pricy* 2U graphic card. This is possible because of the 2U RACK cases I use - they are designed to have a Low Profile APG card installed. Like this one:
http://www.proshop.dk/Grafikkort/ASUS-GeForce-GT-610-2GB-DDR3-2342329.html
I'm just having trouble finding the Low Profile Bracket ;) But it would really make the "crunching" a lot better, right? Is a GPU really better than a CPU ?, I didn't know that!?!?


Let me be more specific. BOINC won't work on just any old GPU. The GPU must be what's called a General Purpose Graphics Processing Unit (GPGPU) in order for it to support the right API that allows it to be crunched on. GPGPUs weren't manufactured until the PCI Express (PCIe) bus came about, and GPU manufacturers never back-ported their designs for the AGP bus. Some people have found some regular PCI-based cards that support GPGPUs, but these are generally very low-end crunchers and the performance, while still better than a CPU, is sub-par compared to their PCIe counterparts.

[Edited to add] Now that I think about it, I believe AMD/ATi had Radeon HD 4xxx series cards that were made for AGP. If you can find a low-profile Radeon HD 4xxx series card and load the correct drivers, it may work.

Going further, I see many of your servers are running Linux. As far as I can remember, there is no official Linux GPU app to even utilize them for crunching. You would have to install a third-party application to get Linux to crunch on GPUs.
ID: 1420600 · Report as offensive
DanHansen@Denmark
Volunteer tester
Avatar

Send message
Joined: 14 Nov 12
Posts: 194
Credit: 5,881,465
RAC: 0
Denmark
Message 1421266 - Posted: 28 Sep 2013, 4:24:16 UTC

To "Volunteer Tester"

First of all, thank you very much for your answer!
Now this is an answer I can use! You certainly set things straight here! And thank you for that.
Regarding Linux computers, yes, I have a BOINC project going, and all computers are only running the BOINC-client, nothing else. OK, I'm using 1 windows machine as well, but this is an office desktop, where I wanted to test this GPU stuff and see if I could get it to accept the graphic card and show it in the "GPU" column under "computers/hosts"

Regarding the Intel P4 CPU, thanks for all the information!! Now I know ;)

Regarding the graphic cards and GPU's I now know what to do. And I mean exactly what to do!
I no longer have to worry about my RACK mounted Linux computing motherboards socket478/P4 with onboard graphic card being destroyed, because it's not possible at all!
And the desktop computer/GPU I just tested will not be accepted either, it's a NVIDIA Quadro FX 1400 which, I guess, is to old ;) Problem solved!

Thanks a lot for your full answer - it really helped me..

Kind regards,
Dan Hansen
Denmark




ID: 1421266 · Report as offensive
Chris Berry

Send message
Joined: 3 Jul 99
Posts: 1
Credit: 7,139,279
RAC: 0
United States
Message 1433183 - Posted: 25 Oct 2013, 13:44:46 UTC - in response to Message 1421266.  

Dan

Nvidia has a list of CUDA/OPENCL capable vidio cards. Just google nvidia gpu list. I use linux and Fedora with the rpmfusion nvidia drivers with few problems. Fedora however has not updated the Boinc Client to 7.2.27 which correctly handles the CUDA and the OPENCL detection. I may have to switch to Ubuntu until that is fixed.

Chris Berry
ID: 1433183 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1433254 - Posted: 25 Oct 2013, 15:26:53 UTC - in response to Message 1433183.  

7.2.27

7.2.27 isn't out yet. I think you meant 7.0.27 and if so, that's a bad version as it still contains the memory detection bug, where when the GPU is being detected the available memory on it can be detected as several petabytes worth of RAM. Better is for people to use 7.0.65 where possible.

Furthermore, to know which Nvidia cards do OpenCL is very simple: All CUDA capable GPUs do OpenCL. Meaning that from a GT8400 till the latest GTX780, all do OpenCL.
ID: 1433254 · Report as offensive

Questions and Answers : Unix/Linux : BOINC crash & destroy GPU


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.