Ubuntu 12.04 LTS - success; Tesla S1070 - half failure

Moderators: slegrand, Site Moderators, PandeGroup

Ubuntu 12.04 LTS - success; Tesla S1070 - half failure

Postby openvoid » Mon Apr 30, 2012 7:25 pm

Hello, got successfully rununng client on Ubuntu Server 12.04 LTS.
Everything is done by viewtopic.php?f=54&t=6793
Exept one small thing - wine refused to search .dll's wrapper and symlinks in system32 folder, so i just copied cu*.dll* nv*.dll to working folder of folding client.
This working even with additionaly installed in different directory cudatoolkit 4.2 for other calculation software, and both added to ld.so.conf
Noticed FahCore_11.exe uses around 40% of some core, which drops to around 20% when second GPU is loaded by another calculation.
Don't tested yet how much it affect folding speed.

The sad part:

Have a big problem with Tesla S1070 with PCI-E x8 interface cards.
With only single card it works ok. Any combination of interface cards, cables, slots.
But when i connect with two cards it halts system or hangs just few secounds after computation started.
Tried following
GA-Z68A-3DH-B3 motherboard. It have 3 PCI-E x16 connectors, one in x4 mode, two in x16 or x8 mode, depending on both populated or first only.
Well, probably those x8 mode is somehow incorrectly routed on motherboard, so i tried
Asrock Z68 Extreme7 motherboard with 5 PCI-E x16 with special chip to warrant enough lanes, one PCI-E in x4 mode, one x8, and two x16
received following result
only one card is visible after boot, and only in sli preffered slots both visible but hangs system when calculation starts.

Can someone suggest something?
openvoid
 
Posts: 4
Joined: Mon Apr 30, 2012 7:08 pm

Re: Ubuntu 12.04 LTS - success; Tesla S1070 - half failure

Postby Hyperlife » Mon Apr 30, 2012 11:00 pm

openvoid wrote:Hello, got successfully rununng client on Ubuntu Server 12.04 LTS.
Everything is done by viewtopic.php?f=54&t=6793
Exept one small thing - wine refused to search .dll's wrapper and symlinks in system32 folder, so i just copied cu*.dll* nv*.dll to working folder of folding client.
This working even with additionaly installed in different directory cudatoolkit 4.2 for other calculation software, and both added to ld.so.conf
Noticed FahCore_11.exe uses around 40% of some core, which drops to around 20% when second GPU is loaded by another calculation.
Don't tested yet how much it affect folding speed.

Thanks for letting us know -- I need to experiment with 12.04 and update my guide. Are you using Wine 1.4?

openvoid wrote:The sad part:

Have a big problem with Tesla S1070 with PCI-E x8 interface cards.
With only single card it works ok. Any combination of interface cards, cables, slots.
But when i connect with two cards it halts system or hangs just few secounds after computation started.
Tried following
GA-Z68A-3DH-B3 motherboard. It have 3 PCI-E x16 connectors, one in x4 mode, two in x16 or x8 mode, depending on both populated or first only.
Well, probably those x8 mode is somehow incorrectly routed on motherboard, so i tried
Asrock Z68 Extreme7 motherboard with 5 PCI-E x16 with special chip to warrant enough lanes, one PCI-E in x4 mode, one x8, and two x16
received following result
only one card is visible after boot, and only in sli preffered slots both visible but hangs system when calculation starts.

Can someone suggest something?

Did this work under previous versions of Ubuntu? I have gotten multiple cards to run, but they were both older cards (9600 GSOs). Are both cards visible in /dev (i.e. /dev/nvidia0, /dev/nvidia1)?
Image
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: Ubuntu 12.04 LTS - success; Tesla S1070 - half failure

Postby openvoid » Tue May 01, 2012 5:28 am

Hyperlife wrote:Thanks for letting us know -- I need to experiment with 12.04 and update my guide. Are you using Wine 1.4?


Yes, its wine 1.4 from ppa in the setup post.
I suggest most short and correct way to add ppa for guide:
add-apt-repository ppa:ubuntu-wine/ppa
Though this command is available from additional package python-software-properties, and should be installed first.

Hyperlife wrote:Did this work under previous versions of Ubuntu? I have gotten multiple cards to run, but they were both older cards (9600 GSOs). Are both cards visible in /dev (i.e. /dev/nvidia0, /dev/nvidia1)?


The S1070 is not a new card (2008). It's 1U rack of 4 x C1060 (T10) cards which bridged to 2 x PCI-E interface cards.
Tried this on 10.04 also with same result - system hang. Was thinking its because of PCI-E x8 slots are actually x4 on my server board. Then went to shop and bought consumer ATX motherboards, mentioned above, but that was not a solution yet.

From every interface cards two bridged GPUs are visible, /dev/nvidia0, /dev/nvidia1 (after cudainit script). It hangs after 2-3 secounds when any calculation on GPU starts, on any GPU, when all four 0..3 cards are visible.

UPD: Well found true reason of hangs.
Unfortunately its hardware failure. When port with faulty hardware is connected - it hangs. When only one good port is connected - it works. I could find it earlier but for some joke of fortune, i was continiuosly connecting only good port when doing lots of experiments.
openvoid
 
Posts: 4
Joined: Mon Apr 30, 2012 7:08 pm

Re: Ubuntu 12.04 LTS - success; Tesla S1070 - half failure

Postby ABBAS » Wed Aug 22, 2012 6:25 am

Can you tell me how could you install Nvidia driver ?
ABBAS
 
Posts: 27
Joined: Sat Apr 28, 2012 6:10 pm
Location: Cairo, Egypt


Return to unOfficial Linux GPU (WINE wrapper) (3rd party support)

Who is online

Users browsing this forum: No registered users and 1 guest

cron