When I have CUDA work I can run 2 WU's on my 980 faster that it would take to run 1 at a time.


But with these SoG WU's it's faster on my 980 to run them one at a time. I noticed for a few days I got all CUDA work now they all end with SoG

---

I'm not aware of the difference between CUDA 32,50 and so on and these SoG WU's .

Will I have to monitor my box to change my XML file around depending on what WU's I get?

---

There is no such thing as a CUDA or SoG workunit. What you are seeing is that application that was assigned to the task when the host requested work.

The SoG application is based on OpenCL which requires a bit more CPU support than do the CUDA applications.
If you have been using the app_config.xml to run 2 tasks at a time then you could start by setting the CPU usage for one CPU per task.

Additionally the SoG has tuning parameters that cam be added to the mb commandline txt to adjust how it runs.
You should have a "ReadMe_MultiBeam_OpenCL_NV_SoG.txt" in your \projects\setiathome.berkeley.edu\ folder that will describe the parameters.

For my GTX 750 ti I run 2 at once with the SoG app and the performance is pretty near the same as the CUDA5 app.
Except for GBT VLAR tasks. Which the SoG application handles much better.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Only difference i've found really is that SOG " likes" to have a whole thread on the cpu , apart from that no difference :-)

---

Life is what you make of it :-)

When i'm good i'm very good , but when i'm bad i'm shi#eloads better ;-) In't I " buttercups " p.m.s.l at authoritie !!;-)

Hi,

You can take a look at the Computing - statistics - top computers - #2.

There is a CUDA app running on a (actually four) GTX980. Look at the task times. Find some vlar tasks to compare... EDIT: that may be hard because of rescheduling. You can take a look at the #1 and scale the vlar times considering they would take about 1.5x the time on a GTX980.

It may serve as an example of what to expect from a CUDA on Windows some time in the (near?) future.

Petri

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

It may serve as an example of what to expect from a CUDA on Windows some time in the (near?) future.

Petri

Quit teasing us, Petri. :^}

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

SoG is Raistmer's app for handing the new data from Green Bank Observatory data. Those data sets were larger than the traditional Arecibo data. It was found that the old cuda apps weren't as good as running those. Raistmer developed the SoG app over time to compensate for these differences but it is an OpenCl application. It too was slower but over time evolved into a much faster app that greatly increase productivity. It had to be fine tuned to older and slower GPUs over the year but has gotten to the point where it runs pretty much stable on most people's computers without having to modify their setting at the expense of 1 CPU core per work unit.

Petri's is referring to his special cuda app that he has been developing over the same period of time. I acknowledge that it is faster than SoG, and he has made great strides in improving it's inconclusive rate (much lower than it used to be) but it's still running only on Linux. As such, it requires much more input from the end user in knowing how to install, code,modify, etc than a standard end user can do. They are attempting to port it to Windows and are making strives but it is still not something the average person can run.

Hope this gives you a better understanding of the applications.

Zalster

---

I find that the "use_sleep" parameter is very useful for SOG. Instead of nearly 100% of a CPU being used by the app, Arecibo WUs use about 20% of a CPU over the life of a WU's computations, Green Bank, about 30-35%. Thus I set no_cpus to 0.5 in my app_config.xml for GPU. My impression is that I get more work done (980s and 1080s running 3 WU at a time).

---