Hi all, I recently bought a GTX 980 and most of the app_config.xml example files seem out of date to some degree w/ the new version of s@h. I was wondering if someone could post an example that would be reasonable to use with my card, and considering that I have a Sandy Bridge core i3 (i3-2120T).

Thanks

---

<app_config>

<app>
<name>setiathome_v7</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.2</cpu_usage>
</gpu_versions>
</app>

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.2</cpu_usage>
</gpu_versions>
</app>

</app_config>


Watch with GPU-z and win task manager and readjust if necessary.

---

If you wish to crunch multiple WUs at a time:

----------------------------------

<app_config>
<app>
<name>astropulse_v7</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v8</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>0.50</gpu_usage>
<cpu_usage>0.04</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v7</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>0.50</gpu_usage>
<cpu_usage>0.04</cpu_usage>
</gpu_versions>
</app>
</app_config>

---------------------------------------

This configuration makes SETI@Home crunch 2 WUs at a time. (AP v7, v7, and v8)


TL

---

TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join Calm Chaos

<app_config>
<app_version>
<app_name>setiathome_v7</app_name>
<plan_class>cuda50</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
<app_version>
<app_name>setiathome_v8</app_name>
<plan_class>cuda50</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
<app_version>
<app_name>astropulse_v7</app_name>
<plan_class>opencl_nvidia_100</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
</app_config>

He's running a 980, so he should be able to run 3 work units on that card.

He should install the lunatics app and choice the cuda 50 for that card.

Thanks guys, that's hugely helpful. By the way, I'll see if things improve by trying these config files, but I just installed GPU-Z and see that the load is only around 50-60%. I have it configured right now to run three GPU threads simultaneously with each one also using 10% of the CPU. I figure with the correct settings the load should be close to 100%? Also I can see the GPU load jumping to 0% sporadically.

---

Are you running CPU work units at the same time?

Edit..

How many total work units are you running?

Right now I have running:
1) 3x of: .1 CPUs + 0.33 NVIDIA GPUs
2) 4x just using the CPU's 4 cores

Really my first time crunching with a GPU so that might be really wrong.

---

The reason the GPU drops to zero is you are starving the GPU for CPU time.

Even thou you put it at 0.1 CPU, that is just an estimate.

The GPU is going to use what it needs, and in most cases it's closer to 25-35% of a CPU core.

So when you have 4 CPU work task going at the same time you are not allowing any time for the GPU task.

You need to cut down on some of the CPU work.

Couple of ways to do that, I prefer to use a <project_max_concurrent>

I'll modify the one I posted to allow for 3 GPU work units and 3 CPU work units. I would try that first and look and see how the times are. If you continues to see the GPU fall to zero then change the value to 5.

<app_config>
<app_version>
<app_name>setiathome_v7</app_name>
<plan_class>cuda50</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
<app_version>
<app_name>setiathome_v8</app_name>
<plan_class>cuda50</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
<app_version>
<app_name>astropulse_v7</app_name>
<plan_class>opencl_nvidia_100</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.33</ngpus>
</app_version>
<project_max_concurrent>6</project_max_concurrent>
</app_config>

Thanks - By the way I notice that Lunatics installs its own app_info.xml file with all of its own parameters - is it OK to then add your own customized app_config.xml file to the same folder? Not sure if they might conflict or something.

---

Lunatics will install it's own app_info

You can put the app_config into the same folder and boinc will read the app_config values and over ride the values in the app_info.

There is no problem with having them both in the same folder.

I do that all the time without any problem.

Thanks! Hugely useful info. I have it running pretty well right now, and plan to continue to tweak and optimize.

---

This one caught me on app_info.xml. The <count>.33</count> isn't acceptable in app.info for V8. It must be <count>.3</count>. Is app_config different in this instance?
Are there any instructions for building the app_config? I'd like to switch over. Is app_info still required to load the executables and DLLs?

---

Delta-V

I can't explain nor comment on the problems you had with the app_info.

You could try installing this app_config and see if it works (without changing the values in it)

If it work, great.

If it doesn't then you could try changing the value to 0.3 but that isn't how it usually works.

It works in my system and it appears to work for the OP. (at least we haven't heard from him saying he is having problems)

You definitely need the app_info to point the apps to where the libraries and executables are located.

Let us know how it goes

---

I didn't have any problems with .33. I'm currently running it at 1 full CPU core per GPU thread as that's been helping increase the GPU load.

---

Would that last example be the best for the new rig I'm constructing too, Zalster?

---

Akio,

That last reversion of the app_config.xml was for his machine alone.

It's set up for 1 gpu running 3 work units and 3 work units on the CPU.

So that is why the project max concurrent is set at 6.

For your rig, if you will need to keep at least 2 cores frees.

There is also a question if you plan on hyperthreading your cpu and if you will be running any work on the CPU as well, so the value of the project max concurrent will be different.

For your rig, if you will need to keep at least 2 cores frees.

There is also a question if you plan on hyperthreading your cpu and if you will be running any work on the CPU as well, so the value of the project max concurrent will be different.

I have my i7-960 (4 cores) with HT enabled. It takes a while to complete tasks, but that's the way of things. I planned on doing the same thing with my new i7-5930K (6 cores). What are the pros and cons of maxing out the CPU usage (including GPU assists)? Should I not worry so much about utilizing the CPU and concentrate more on the GPUs doing the work?

---

With higher end Intel chips you can run some CPU work but I would not max out the CPU. How many? You have to run test and find what is the best ration for your configuration.

With lower to mid range Intel chip I would use them to support the GPUs only.

The amount of work done by 1 GPU can easily out do any CPU in the same amount of time.

There's no 1 answer as each computer have a lot of variables that will influence how they crunch.

So for my i7-960, how would I modify my settings to only run GPU work and just have the CPU assist them?

---

Go to your preference setting on your account page and deselect "use CPU"

That way only the GPUs will be used to run work units.

If you do that then you can use the modified app_config that I posted as you will only be running 3 work units on both of your GPUs, for a total of 6 work units.

There should be enough free cores after that to support those GPUs

I can't really talk about the i7 960, all I have to go off is the comparison against other i7s from some review pages.

From what I see, you might be better off turning off the hyperthread and just running it as a pure 4 core. (that is up to you)

You can try it both ways, with and without hyperthread and see how the times are.