KWSN Orbiting Fortress Forum Index KWSN Orbiting Fortress
KWSN Distributed Computing Teams forum
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

The early implementation of GTX-690...

 
Post new topic   Reply to topic    KWSN Orbiting Fortress Forum Index -> Ye Olde Help Scrolls
View previous topic :: View next topic  
Author Message
Putting_things_on_top
Duke
Duke


Joined: 14 Oct 2009
Posts: 435
Location: Frostbite Falls, Minnesota, USA

PostPosted: Fri Aug 03, 2012 10:35 pm    Post subject: The early implementation of GTX-690... Reply with quote

I have just posted a 'how-to' for getting a GTX-690 to work on Win-7 on the Stanford forum.
It might be helpful for any of the Kepler series cards, too (if you're experiencing BSODs).

http://foldingforum.org/viewtopic.php?f=38&t=21049&start=90#p221398



Oh - don't be surprised if they move my post elsewhere:
Some officious content nazi noticed that I mentioned "client-type=beta" and immediately replied that my post should be moved to the "beta" topic.
#Bad Talk
_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC


Back to top
View user's profile Send private message Visit poster's website
Putting_things_on_top
Duke
Duke


Joined: 14 Oct 2009
Posts: 435
Location: Frostbite Falls, Minnesota, USA

PostPosted: Fri Aug 03, 2012 11:04 pm    Post subject: Reply with quote

And just in case they scuttle my post, I'll copy it here.
---------------------------------------------------------------------------------------------------------------------------------------------------------
I recently purchased a GTX-690 for a new F@H rig, and I'd like to share some of the obstacles & observations I have gone through in order to get this beast to work.

The GTX-690 appears to Win7-OS as a dual GPU card in SLI. GPU-Z (and other hardware monitors) will show 2 'cards'.

First thing I had to do was to 'break' the default on-board SLI.
The NVidia control panel (driver version 301.42) has an option "Disable multi-GPU mode" - I selected this option.
On the same panel, I changed the PhysX processing to "CPU" (I doubt that F@H uses PhysX - could be wrong).

----EDIT---------- this may be an incorrect assertion on my part --------
Next thing...
After careful & exhaustive trolling on the geek-ware sites, it struck me that the PCI device class/subclass identifiers for each of the GPUs might be different.
So (and since I have no other GPUs on the new rig), I pared-down the GPUs.txt file to just the essentials for my rig:
Code:
0x10de:0x1188:3:0:GK104 [GeForce GTX 690]
0x10de:0x1188:3:2:GK104 [GeForce GTX 690]
Note the "3:0" and the "3:2" entries. See http://pci-ids.ucw.cz/read/PD/03
----END-EDIT------------------------------------------------------------------

Now (in my best impression of Paul Harvey)..."Page 2":

In order to get the V7 client to even recognize these cards, I had to set "client-type=beta".
Yes, I realize that this setting has the potential to cause minor problems, but what the heck!
So, once that was done: VOILA! I'm folding on 690s.

Well, not so fast there, Buck-O. BSODs!!!!
The BSODs were tossing stop-code 0x0000000000000101 (see http://pcsupport.about.com/od/findbyerrormessage/tp/stop_error_list_0xc1_0x11d.05.htm)
I can't remember where I found the article, but a very knowledgeable person said that that "...the 'secondary processor' is quite often referring to a GPU...".

We need to understand that the Kepler GPUs have this new automatic "GPU Boost" feature.
As the card-thermals allow, the GPU will automatically up-clock or down-clock itself.
This - in my opinion - is the foundation for these 0x0000000000000101 BSODs.
During F@H processing, there is some coordination going on between the GPUs and the CPU.
If the GPU is altering its own clock rate while there is an outstanding timing-signal, that signal may either time-out (too late) or respond prematurely (too early - CPU not ready for the reply).

I found that even our "Gamer" friends have some minor issues with this variable clock-speed.
See http://www.overclock.net/t/1267918/guide-nvidia-inspector-gtx670-680-disable-boost-fixed-clock-speed-undervolting

But, I found a tool that can set the clock-speed for the Kepler GPUs to a fixed speed. It's called NVidia Inspector mentioned in URL above Smile
Download it safely from here: http://www.softpedia.com/dyn-postdownload.php?p=169925&t=4&i=1

For my purposes, I created a Win7 task (thru task scheduler) triggered upon Windows-startup.
I created a .bat file to use the command-line mode of NVidia Inspector:
Code:
nvidiaInspector -setGPUClock:0,2,1020 -setPowerTarget:0,115 -forcePState:0,2 -setFanSpeed:0,85
nvidiaInspector -setGPUClock:1,2,1020 -setPowerTarget:1,115 -forcePState:1,2 -setFanSpeed:1,85
Remember to define the task paramater "start in" as the directory where the nvidiaInspector executable resides.

A word to the developers & beta-testers:
The variable clock-speed of the Kepler series is - in my experience - something that needs to be addressed (to prevent BSODs).
It would be unreasonable to expect most F@H participants to be capable of this level of jury-rigging.
I don't know what the answer is, but it should be looked into with a fair amount of urgency.
When donors are willing to pop $400-$1100 per card, they expect it to be "plug-n-play"!
Additionally, using the beta client-type just allows us [currently] to execute Fermi WUs on a Kepler.
Kepler has roughly 3x the cores than its Fermi counterpart has, so its performance seems a little underwhelming (under-utilized).

I know that there is a new GPU core being developed for Kepler.
I am (as well as many others) a bit anxious waiting for its arrival! Shocked
_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC




Last edited by Putting_things_on_top on Sat Aug 04, 2012 10:35 pm; edited 1 time in total
Back to top
View user's profile Send private message Visit poster's website
Concrete-mixing Moose
Prince
Prince


Joined: 30 Apr 2012
Posts: 567
Location: The Joyce Grenfell Home for the Distressed

PostPosted: Fri Aug 03, 2012 11:12 pm    Post subject: Reply with quote

Thanks for the work around info. Have you tried the card on any other projects, and if so, got the same problem?
#ni-1
_________________

Save my home - click every day on the picture!
Back to top
View user's profile Send private message Send e-mail
Gemjunkie
Prince
Prince


Joined: 03 Jul 2010
Posts: 3519
Location: Earth, lately

PostPosted: Sat Aug 04, 2012 7:34 am    Post subject: Reply with quote

Nice work. Cool


#ni-1
_________________




(older, before split CPID)
Back to top
View user's profile Send private message
Putting_things_on_top
Duke
Duke


Joined: 14 Oct 2009
Posts: 435
Location: Frostbite Falls, Minnesota, USA

PostPosted: Sat Aug 04, 2012 10:07 pm    Post subject: Reply with quote

Yet another officious content nazi (on the Stanford forum) has scolded me for presuming that the "3:0" and "3:2" in the GPUs.txt file is the PCIe device-class/subclass numbers.
I have no logical reason to doubt the person, but the tone was rather "snarky".

Alas, I may already be perceived as persona non grata by some self-appointed group of "hall-monitors" on that forum.
They seem to have their precious little secrets - and although most of those secrets are public knowledge anyway: "...first rule of Fight Club: no one talks about Fight Club!"
Oh well, it seems that I may have inadvertently pissed in their punch-bowl. Embarassed

Thankfully, one of their admins ("bruce") is much more even-tempered and civil in his tone/presentation.
I've seen his moderating influence on that forum for a while, now.
And I - for one - appreciate his efforts!
He can be firm when necessary, but usually strives for accommodation & tolerance.

Long story short: Stanford is "working on it (Kepler)"...but they never give details or ETAs!


_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC


Back to top
View user's profile Send private message Visit poster's website
Putting_things_on_top
Duke
Duke


Joined: 14 Oct 2009
Posts: 435
Location: Frostbite Falls, Minnesota, USA

PostPosted: Sat Aug 04, 2012 10:54 pm    Post subject: Reply with quote

Concrete-mixing Moose wrote:
Thanks for the work around info. Have you tried the card on any other projects, and if so, got the same problem?
#ni-1

Just speculating here, but I think that this would affect performance, and risk similar BSODs regardless of the project.
Trying to synchronize a CPU (which uses a constant clock-rate) with a Kepler GPU (which now has an automatic, variable clock-rate) can be an enormous challenge.

This was not a problem with previous GPU architectures, since the GPUs did not up-clock themselves automatically.
They would - under severe conditions - radically down-clock to prevent thermal incineration.

But with Kepler, it's kinda like someone constantly fiddling around with MSI Afterburner or EVGA PrecisionX while the GPU is under heavy or full load:
Bad things will happen! Twisted Evil


_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC


Back to top
View user's profile Send private message Visit poster's website
Gemjunkie
Prince
Prince


Joined: 03 Jul 2010
Posts: 3519
Location: Earth, lately

PostPosted: Sat Aug 04, 2012 11:06 pm    Post subject: Reply with quote

You'd kinda think someone woulda thought of that. Rolling Eyes
_________________




(older, before split CPID)
Back to top
View user's profile Send private message
Putting_things_on_top
Duke
Duke


Joined: 14 Oct 2009
Posts: 435
Location: Frostbite Falls, Minnesota, USA

PostPosted: Sun Aug 05, 2012 1:40 am    Post subject: Reply with quote

Gemjunkie wrote:
You'd kinda think someone woulda thought of that. Rolling Eyes

Yes, I am aghast at the astonishing level of "not seeing that one coming" in the DC/HPC developer communities.

And I'm afraid that the ignorance is even more prevalent than I feared.
Many of the gamer sites are filled with wailing & moaning about the so-called lack of stable overclocking with the GTX-6xx series.
Have they NOT been paying attention?
Traditional overclocking methods are not relevant with this card series, and may (in some instances) be counter-productive or even destructive.

Did you ever see the movie "Idiocracy"?
If you have, you know how I feel at times!
A simultaneous combination of and #2gunfire and
#ni-1
_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC


Back to top
View user's profile Send private message Visit poster's website
Gemjunkie
Prince
Prince


Joined: 03 Jul 2010
Posts: 3519
Location: Earth, lately

PostPosted: Sun Aug 05, 2012 2:42 am    Post subject: Reply with quote

While not a perfect parallel, it reminds me of the old frustration of buying hot dogs in packages of 10, and only being able to buy hot dog rolls in packages of 8. Each group focuses on their product and doesn't care what the other is doing or how the products work together.
_________________




(older, before split CPID)
Back to top
View user's profile Send private message
Display posts from previous:   
Post new topic   Reply to topic    KWSN Orbiting Fortress Forum Index -> Ye Olde Help Scrolls All times are GMT - 5 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group
Optimized Seti@Home App | BOINC Stats