GTX 460 superclocked

[Ghost0210]

just going through the tasks I hadn't reported, hoping to get definitive times from my testing with mulitple tasks running

Ok, after running through the tasks I haven't yet report and the online testing I done with whichever random VHAR tasks Boinc decided to throw at me next (I'll have to set up some real test conditions when I get the time)

So, first of all I need to slap myself for not doing this before I first posted and my apologies Frizz23 
Revised and accurate average times:
Like I said, all VHAR and all different tasks:

1 task at a time::each task took ~ 200 seconds to complete
2 tasks at a time::each task took ~330 seconds to complete[revised time]
3 tasks at a time::each task took ~520 seconds to complete[revised time]

So tasks completed in an hour using the above times:
1 task at a time:: 3600 / 200 = 18 tasks an hour
2 task at a time:: 3600 / 330 * 2 = 21.8 tasks an hour
3 tasks at a time:: 3600 / 520 * 3 = 20.7 tasks an hour

So the best throughput (highest number of tasks per hour) on my 465 works out to 2 at a time, could well be different on your 460 if you decide on one, but I agree with Jason, its a good card and as nVidia release more documentation on it, it will only get better

[sunu]

You guys that run 2 or 3 tasks at a time in your fermi cards, what RAC does that give to your cards?

As for the question about the 768MB card I would advise against that. 768MB memory seem quite small nowadays. Maybe 2-3 years ago yes, but not now. Get the higher memory model.

[Frizz]

How can I see in the WU XML file whether or not a WU is a VHAR?

Shorties (are they always VHARs?) usually run in 5:50 on my GT 240. So it's about 10 per hour.

Somehow I can't believe the 465 is only 2x my cheapish GT 240 ...!?

[Jason G]

You guys that run 2 or 3 tasks at a time in your fermi cards, what RAC does that give to your cards?

On the 480, with x32f 3.1, running on a E8400 Oc'd to 3.6GHz, I just passed 21000 today.  The CPU is running AKv8b with occasional AstroPulse... I think it should start to level off soon, but don't know for sure.  I'm finishing two midrange tasks on the 480 in about 9-12mins, or two shorties in ~3 mins.   single task times were ~7-8 mins & ~2mins for Mid AR & shorty respectively.

As for the question about the 768MB card I would advise against that. 768MB memory seem quite small nowadays. Maybe 2-3 years ago yes, but not now. Get the higher memory model.

  I agree.  For me it's the memory bus width though.  With this much processing power at hand memory bandwidth will nearly always be the primary concern, and nVidia wisely have included some special cache control mechaisms in the driver API ( which we can't switch to yet, until I have the complete set of freaky powerspectrums, still on track for christmas 2010 , later to be replaced with a freaky powerspectrum cascade kernel to maximise cache benefits)

[Jason G]

How can I see in the WU XML file whether or not a WU is a VHAR?

Shorties (are they always VHARs?) usually run in 5:50 on my GT 240. So it's about 10 per hour.

Somehow I can't believe the 465 is only 2x my cheapish GT 240 ...!?

As you get faster on GPU, a larger proportion of shorty run time is CPU time    That's a fairly fixed overhead between tasks & largely machine dependant.

You can search for the word angle in the task file using a text editor.  The true angle range of the task is there.  It's easier just to look at the stderr task output after the task reports (and the servers are up) though.  There is an exact threshold above which is considered VHAR, and that makes shorties.  I forget exactly what that threshold is, around 1.1 I think.  Joe keeps that kind of information around here  .  Above that, whatever it is, they all tend to take pretty much the same short time to process.

[Frizz]

It's easier just to look at the stderr task output after the task reports (and the servers are up) though. 

Yeah bummer ... now since the servers are offline I'm starting to poke around :-)

[Miep]

It's easier just to look at the stderr task output after the task reports (and the servers are up) though. 

Yeah bummer ... now since the servers are offline I'm starting to poke around :-)

There's always client_state to rummage around for stderr (after the task finished. before that you'll have to poke in the slot directories)

[Josef W. Segur]

I am planning to build a system with :
1x GT 240
1x GTX 460

How do I have to modify my app_info.xml to run those two cards with x32f?

The GTX 460 can run more than one task in parallel, right? Was it 3? How do I have to configure this?

I see you have answers for both questions in isolation, but nobody has pointed out a conflict. AFAIK, if you have a GT 240 in the system it won't be practical to run more than one task at a time on the GTX 460. It's a shortcoming in BOINC, there's no way to set up a separate app_version for each card.
                                                                                      Joe

[Raistmer]

It's a shortcoming in BOINC, there's no way to set up a separate app_version for each card.
                                                                                      Joe

What if one app version uses fermi plan class and another one - only cuda23 ?
Will cuda23 tasks be scheduled on both GPUs ?

[And most important - if fermi tasks will be scheduled only on 460 GPU, then it can do 3 tasks at once at least sometimes while 240 will do one task at once always]

[Richard Haselgrove]

I'm not certain, but I'm fairly sure that <plan_class> isn't used at the client scheduler level for determining device allocation.

That is, <plan_class> cuda23 --> device 1
<plan_class> cuda_fermi --> device 0 ?

I don't think so.

[Raistmer]

That is, if non-FERMI GPU installed and app_info has only fermi plan class app_version entry, "fermi" tasks will be still scheduled on non-FERMI GPU ?? If so it's plain bug in BOINC.
plan class should limit number of usable devices.
That is, both devices available for cuda23 application, but only single FERMI device is available for fermi application.
[And even such approach can be wrong. CUDA 23 app can't work OK on FERMI GPUs
That is, different plan classes should have different device lists indeed.]

[Frizz]

I'm not certain, but I'm fairly sure that <plan_class> isn't used at the client scheduler level for determining device allocation.

But how does this work for hosts crunching MBs with NVIDIA and APs with ATI?

[Raistmer]

LoL, it's not so easy to get such config working at all. Ask Claggy

[Frizz]

I have such a setup running too

[Raistmer]

I have such a setup running too

Successfully? What OS/Catalyst ? I failed to launch such config because ATI's lack of support for Win 2003 Server OS, Claggy had bad failures for different apps when both GPUs installed... Interesting what makes your config so successful