I assume I'm hitting a resource limit ... but not sure which one.

I have a Dell studio XPS 8100 with a core i-7 860 2.8 Gh and 12 Gb mem running win-7 64 bit SP-1

I'm running BOINC 6.12.34 and Lunatics_Win64_v0.41_setup.

I am running 4 AP work units.

I had an nvidia 560 TI with 448 gpu's running for a long time on the PCI-2.0 x 16 slot and was able to get
about 31,000 Cobblestones out of it.
I also have an nvidia 9500 gt 32 gpu's running on the open pci slot getting about 525 cobblestones.

I had some good luck at the casino so I bought a nvidia 770 with 1536 gpu's. I reinstalled lunatics for Kepler and set

processpriority = abovenormal
pfblockspersm = 8
pfperiodsperlaunch = 200

GPU-Z shows avg loads at

GPU Core clock 1188.9
GPU Mem clock 1752.8
GPU Temp 63.0 Deg C
Gpu load 93%
Mem contr load 45%
power 58.2%

after 2 weeks now I am getting about 31,000 cobblestones

so

I still have 50% cpu available and I still have 4 AP processes running at the
full 12.5% CPU available on the 4cpu's (and 4 HT whatevers)

Is my "limit" the CPU power, the PCI-E IO limitation, some other bad configuration?

I realize this may not be enough configuration info but any "best guess" would be appreciated

Ed F

I had some good luck at the casino so I bought a nvidia 770 with 1536 gpu's. I reinstalled lunatics for Kepler and set

processpriority = abovenormal
pfblockspersm = 8
pfperiodsperlaunch = 200

Very good, If you're been running MBv7 that is, you have seven hosts with GTX770's, all but one don't do MBv7, only Astropulse v6.

The one that does only has 16 MBv7 tasks eithier valid or pending.

Please tell us which host you're referring to.

Claggy

I'm not sure how to answer the question ... the MBv7 is currently suspended so I could get better numbers on how AP was running (reduce the number of variables in the "test")

The current question is about AP ( I know the mbcuda.cfg refers to the MBv7 only).

I don't know if the PCI-e2.0 x16 is overwhelmed by the traffic imposed by 4 AP's running (It would surprise me if it were THAT SLOW!!)

my assumption is that the CPU is the limiting resource. Each cpu appear to be spending about 3/4 of available time busy in krnl mode and 1/4 busy in user mode (according to my reading of the task manager graphic) - (assigning each ap process to a cpu i.e.(0,2,4,6)

There is a very reasonable possibility the slowness may be due to my missing something needed to take advantage of the nvidia 770 ...

it just seems strange that there is NO improvement between the 560 TI and the 770 ... some resource is at the wall ..

Any ideas would help!!

Ed F

There is a very reasonable possibility the slowness may be due to my missing something needed to take advantage of the nvidia 770 ...

it just seems strange that there is NO improvement between the 560 TI and the 770 ... some resource is at the wall ..

Any ideas would help!!

Ed F

All your GTX770 hosts are running the default AP parameters ment for maximum compatibility across GPUs from the 8400GS upwards,
the ReadMe_AstroPulse_OpenCL_NV.txt gives you everything you need to know about supplying parameters that are more likely to utilise your GPUs.

Claggy

Sigh!

I can't find "ReadMe_AstroPulse_OpenCL_NV.txt"

Is there a link I can use??

Ed F

It's actually AstroPulse_OpenCL_NV_ReadMe.txt .

                                                                   Joe

Sigh!

I can't find "ReadMe_AstroPulse_OpenCL_NV.txt"

Is there a link I can use??

Ed F

It should be in the docs folder in your seti project directory, if you're used the default install location it will be:

C:\ProgramData\BOINC\projects\setiathome.berkeley.edu\docs

Claggy

Save the following as app_config.xml:

<?xml version="1.0"?>

-<app_config>


-<app>

<name>setiathome_v7</name>


-<gpu_versions>

<gpu_usage>0.5</gpu_usage>

<cpu_usage>.04</cpu_usage>

</gpu_versions>

</app>

</app_config>


and place the file in your "c:\ProgramData\BOINC\projects\setiathome.berkley.edu" folder (without the quotes} and the read your local config files and your local preferences under the Advanced menue of the BOINC Manager.

With all those GTX 770s you should be able to apply for your own Zip Code. You should be doing 2 or 3 times what you are doing now, at least I hope you will.

-Bill

---

-Bill

It's actually AstroPulse_OpenCL_NV_ReadMe.txt .

                                                                   Joe

Not in the Lunatics_Win64_v0.41_setup it isn't, it might have changed since (the filename in my project directory matches yours, but it's dated this year),
I got that file name from an extracted Lunatics_Win64_v0.41_setup directory.

Claggy

Save the following as app_config.xml:

<?xml version="1.0"?>

-<app_config>


-<app>

<name>setiathome_v7</name>


-<gpu_versions>

<gpu_usage>0.5</gpu_usage>

<cpu_usage>.04</cpu_usage>

</gpu_versions>

</app>

</app_config>


and place the file in your "c:\ProgramData\BOINC\projects\setiathome.berkley.edu" folder (without the quotes} and the read your local config files and your local preferences under the Advanced menue of the BOINC Manager.

With all those GTX 770s you should be able to apply for your own Zip Code. You should be doing 2 or 3 times what you are doing now, at least I hope you will.

-Bill

Quite useless, for one thing it's been opened and pasted in something other than Notepad, so the formatting is all wrong, and two, EdwardPF is running Boinc 6.12.34:

http://boinc.berkeley.edu/wiki/Client_configuration#Application_configuration

Application configuration

This mechanism allows you to specify scheduling parameters associated with specific applications or app versions. It is available with 7.0.40+ client versions.

Claggy

It's actually AstroPulse_OpenCL_NV_ReadMe.txt .

                                                                   Joe

Not in the Lunatics_Win64_v0.41_setup it isn't, it might have changed since (the filename in my project directory matches yours, but it's dated this year),
I got that file name from an extracted Lunatics_Win64_v0.41_setup directory.

Claggy

I wish I could avoid foot in mouth syndrome, it doesn't taste good!

Claggy is of course right.

                                                                   Joe

I have app_info.xml out there but I sure could use a lesson (more documentation or examples) on app_config.xml. ... lots of stuff out there but what is reasonable for a 770??

Is anyone with an nvidia 770 using app_config.xml with AP?? and willing to share the file??

Ed F

Claggy is right you're using Boinc 6.12.34. You have really good hardware. Perhaps you should think about upgrading Boinc to take better advantage of the hardware's capabilities?

-Bill

---

-Bill

And the Seti@Home apps, standard or Lunatics 0.41.

-Bill

---

-Bill

how is BOINC 6.12.34 slowing me down??

Ed F

how is BOINC 6.12.34 slowing me down??

Ed F

From Claggy,
"Application configuration

This mechanism allows you to specify scheduling parameters associated with specific applications or app versions. It is available with 7.0.40+ client versions."
Without running a more recent BOINC application, you can't tweak the application parameters to make better use of your hardware.

---

Grant
Darwin NT

how is BOINC 6.12.34 slowing me down??

Ed F

From Claggy,
"Application configuration

This mechanism allows you to specify scheduling parameters associated with specific applications or app versions. It is available with 7.0.40+ client versions."
Without running a more recent BOINC application, you can't tweak the application parameters to make better use of your hardware.

Well, you can through app_info.xml, but it's not as convenient.

On the other hand, v6.12.34 has its advantages - I still run it on Windows XP machines: it's the last version which can run GPUs with BOINC running 'as a service'. I fondly believe that ought to be more efficient (shorter start-up times, for a start), but I'm not prepared to spend time generating evidence to back up that belief.

I/we have seemed to have lost what I thought was the main question here ...

The question centers on the 2 orig lines:

I had an nvidia 560 TI with 448 gpu's running for a long time on the PCI-2.0 x 16 slot and was able to get
about 31,000 Cobblestones out of it.

after 2 weeks now I am getting about 31,000 cobblestones

The question was/is:

With 4 AP work units running (consuming 4 CPU's and 4 "HT CPU's" free) ...

what resource is causing the nvidia 770 to perform at the same level as an nvidia 560 TI??
Am I CPU bound?
Am I PCI-E v2.0 x 16 bound
Am I software configuration bound?

I.E can a computer of this hardware configuration and age be expected to perform better than 31,000 cobblestones regardless of GPU power?

ed F

ok now ,,,

I have been running for only 2 days now with the new set up BUT nothing has changed!

I have ap_cmdline_win_x86_SSE2_OpenCL_NV set to


-unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -hp

and priority is indeed set to "high" and the computer can chew an average
AP in just over 30 min's the computer is still crunching at about 30,000 cobblestones per day ... again ... the nvidia 770 is almost exactly the same as the old nvidia 560 TI!!

I have tried -unroll 24 -ffa_block 16394 -ffa_block_fetch 8192 -hp for a few hours and all the WU in that time ran at the same speed!


Antone out there with a nvidia 770 that can help me know its true potential??

Ed F