perhaps in the futur ... need more developement for making AP wu with BLC data i believe ..

---

quick question.

Are AP tasks ALWAYS Arecibo non-VLAR? it looks like all of my AP task in my list are this format, but i wasnt sure if it was possible to have BLC or VLAR variants of AP tasks.

Yes. No developer to write the BLC AP application. So will never happen.

---

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

A proud member of the OFA (Old Farts Association)

quick question.

Are AP tasks ALWAYS Arecibo non-VLAR? it looks like all of my AP task in my list are this format, but i wasnt sure if it was possible to have BLC or VLAR variants of AP tasks.

Yes. No developer to write the BLC AP application. So will never happen.

Unless someone wants to sponsor them?

---

Grant
Darwin NT

Well we did raise a lot of money on the drive for Parkes http://setiathome.berkeley.edu/forum_thread.php?id=82316&postid=1912560

---

20 of them turned up here for the 16th UTC.

I'm ready and waiting to get my rigs into some Parkes data. :-D

Cheers.

is there a benchmark tool that works for running AP jobs on nvidia opencl?

i downloaded the linux AP benchmark tool V2.0.23 from lunatics site here: http://lunatics.kwsn.info/index.php?action=downloads;cat=44

and using a GPU workunit with the GPU app, it fails to run giving the following errors:

-------------------------------------------------------------------------
Starting benchmark run...
-------------------------------------------------------------------------
Listing wu-file(s) in /testWUs :
ap_15no18ab_B1_P1_00004_20181116_20829.wu

Listing executable(s) in /testAPPs :
astropulse_7.08_x86_64-pc-linux-gnu_opencl_nvidia_100_cmd

Listing executable in /refAPPs :
astropulse_7.08_x86_64-pc-linux-gnu_opencl_nvidia_100

Listing additional reference results in /testData/refResults :
find: ‘./testData/refResults/ref-pulse.*.out’: No such file or directory
find: ‘./testData/refResults/ref-pulse.*.out’: No such file or directory
No additional reference results in /testData/refResults found.
-------------------------------------------------------------------------
Current WU: ap_15no18ab_B1_P1_00004_20181116_20829.wu

-------------------------------------------------------------------------
Running default app with command : ... astropulse_7.08_x86_64-pc-linux-gnu_opencl_nvidia_100
cp: cannot stat 'pulse.out': No such file or directory
cp: cannot stat 'pulse.out': No such file or directory
cp: cannot stat 'indices.txt': No such file or directory
Elapsed Time: ........................ 0 seconds

-------------------------------------------------------------------------
Running app with command : ........... astropulse_7.08_x86_64-pc-linux-gnu_opencl_nvidia_100_cmd
astropulse_7.08_x86_64-pc-linux-gnu_opencl_nvidia_100_cmd failed to run on your system. Continuing with next (if any)...

-------------------------------------------------------------------------
Done with ap_15no18ab_B1_P1_00004_20181116_20829.wu.

=========================================================================

Done with Benchmark run! Removing temporary files!

this script is quite old, from 2013. i'm not sure when AP started up on NV GPUs.

anyway, does anyone know of a benchmark tool that works for AP in this scenario? what I'd like to do is have a tool that can run a set collection of AP WUs on the same app, but only varying cmdline options. you should be able to do this by just renaming one app slightly different like i've done, then define the different cmdline options in the command line text file.

the latest MB benchmark tool does work for MB tasks on GPUs. and on a whim i tried to use it with the AP apps and WU, but it also doesn't work. (i didn't really expect it to, just tried anyway)

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

I had a AP wu pay out 0.01 credits.....ap_11no18aa_B5_P0_00174_20181112_29125.wu_0

---

100% radar blanked. Normal.

---

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

A proud member of the OFA (Old Farts Association)

I had a AP wu pay out 0.01 credits.....ap_11no18aa_B5_P0_00174_20181112_29125.wu_0

It was fully blanked and that happens now and again, but not bad for 23 seconds work. ;-)

Cheers.

Why is the test app executable named differently?

---

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

A proud member of the OFA (Old Farts Association)

I thought Raistmer had written the OpenCl apps for the GPUs. He might be able to point you in the right direction.

---

Why is the test app executable named differently?

Yes. You can see the app names in the blurb from the terminal I posted.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

I thought Raistmer had written the OpenCl apps for the GPUs. He might be able to point you in the right direction.

it's not a problem with the app itself. the app works fine when run in BOINC as normal. I pulled the apps for benchmarking from the project folder where they normally reside.

it seems to be a problem with the benchmark script executing the apps. it seems like it's not properly creating some needed reference files or something.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

Why is the test app executable named differently?

Yes. You can see the app names in the blurb from the terminal I posted.

Oh sorry, I didn’t realize you asked WHY lol.

They are named differently so that later they can be differentiated from each other for the purposes of adding different commandline options for each. They are the same exact app though.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

Ahh, gotcha. I knew each app would be identified with its designation, Test or Reference, but if the purpose of the benchmark runs was to test different command lines, then you will have to name each application differently in the comlineoptions.txt file to apply the different command line options for testing. I would have just run a single case test of the AP application first as reference first to see if it ran correctly before going ahead for the benchmark runs with different parameters.

Did you run the original 2.01.08 benchmark app or the new benchmarkV3 app?

---

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

A proud member of the OFA (Old Farts Association)

I tried both scripts. Neither works correctly. And just exits almost immediately. I’m fairly certain that the V3 script is made for MB, not AP. Looking through the code, the AP script looks to create a bunch of different supporting files, that aren’t made for the MB tasks under the V3 and other MB scripts.

I did try running the script without any commandlines. And I also tried running it without renaming the app (with them both the same), still nothing.

The WU I was testing also has been cruched by the system and took ~400s or so,so it’s not a 100% blanked file either.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

got 19 AP wu for the 16th ^^

---

Yeah, I see that now. I thought for some reason all the script needed was the name of executable and its command line parameters.

---

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

A proud member of the OFA (Old Farts Association)

19 arrived here for the 17th UTC.

Cheers.

so with Keith's help I was able to get the AP benchmark to "kind of" work. for my purposes at least.

but some interesting results.

since the 2070 performance seems similar to a 1080ti, i've been more or less trying to apply similar settings to the AP command line. and following the general rule of unroll = <number of SMs on GPU>. well the RTX 2070 has 36 SMs so I applied -unroll 36, and that actually runs slower than the cmdline arguments provided by default in the CUDA special app package.

default arguments = -sbs 256 -unroll 14 -oclFFT_plan 256 16 256 -ffa_block 2304 -ffa_block_fetch 1152

these actually run pretty well out of the box, but i wanted to squeeze some more work out of the card if i could.

so i tried these: -sbs 2048 -unroll 36 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1 -hp

but with the significant variance in AP task run times from WU to WU, it was hard to gauge if it was actually faster or not by just looking at the runtime of real WUs that come and go.

putting these two sets of cmd line arguments through the benchmark tool (on the same app; ap_7.08_r2751.....) with the set of clean AP WUs from the Lunatics site yeilds the following results:

Default command lines (-unroll 14 .....)
Clean_01LC - 7 seconds
Clean_02LC - 11 seconds
Clean_05LC - 22 seconds
Clean_10LC - 41 seconds
Clean_20LC - 81 seconds

Test command lines (-unroll 36 ......)
Clean_01LC - 7 seconds
Clean_02LC - 11 seconds
Clean_05LC - 24 seconds
Clean_10LC - 43 seconds
Clean_20LC - 85 seconds

and running it against a couple of real AP WUs that i pulled from my task list showed conflicting results, but from what i've been told, the different levels of blanking can cause unpredictable results in how long it takes the WU to finish.

anyway. can someone more well versed in the AP app and it's cmd line options chime in? maybe someone can direct Raistmer over here? since the Turing cards have half the cores per SM than Pascal/Maxwell does, should we instead run unroll values equal to half of the SM count? anyone want to guess what might be more appropriate arguments for the other options as well besides unroll?

i think next i'm going to try these:
-sbs 1024 -unroll 18 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1 -hp

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours