PrimeGrid Races 2018

[Orange Kid]

Ryzen2 2700x
Running 2 x 8
First two in
6895 and 6735 seconds run time.
52477 and 50915 seconds cpu time.
Just plug and play stock settings.

 

[Markfw]

Estimate down to 19 hours, but is that real ?? Is something wrong ?

 

[Ken g6]

OK, I am in. But one day per task ?????

16 tasks at a time on a 2700x@4.2 ghz, memory 3600 with subtimings set, my fastest single-core box, but with 8c/16t

That's because you didn't set up multithreading with an app_config.xml file. The net amount of work done will be similar.

ah! poop!

I knew I wouldn't be around near GO! time. Starting at lunch was less than successful. I was hoping on good timing so that round2 would fully count. Alas, very little luck.

I'm in, but I'm very late.

Three hours late isn't very late in a five day race.

 

[Markfw]

That's because you didn't set up multithreading with an app_config.xml file. The net amount of work done will be similar.
Three hours late isn't very late in a five day race.

OK, what do I put in it ? Windows 10 Ryzen 2700x@4.2 And where does it go ? (directory)

 

[StefanR5R]

Path on Windows is:
C:\ProgramData\BOINC\projects\www.primegrid.com\app_config.xml

@Orange Kid's setup should look like this:

<app_config>
    <app_version>
        <app_name>llrSR5</app_name>
        <cmdline>-t 8</cmdline>
        <avg_ncpus>8</avg_ncpus>
    </app_version>
</app_config>

After you saved this file, shut down and restart the boinc client.

PS,
remember, all of the LLR subprojects of PrimeGrid are biased towards Haswell and its successors.

 

[TennesseeTony]

ThreadRipper is horrible at this. My Haswell 5820K using 4 threads is getting a task done in two hours, but TR with 5 threads is taking 3.5 to 4 hours. I've bumped the TR up to 8 threads, but according to Orange Kid's results, it looks like 8 threads might equal 4 threads of the Haswell.

Off to work now, but I'll check times when I get home (with 8 threads), and give it one last chance perform at 50% CPU usage.

 

[Ken g6]

ThreadRipper is horrible at this. My Haswell 5820K using 4 threads is getting a task done in two hours, but TR with 5 threads is taking 3.5 to 4 hours. I've bumped the TR up to 8 threads, but according to Orange Kid's results, it looks like 8 threads might equal 4 threads of the Haswell.

That sounds about right. Remember, modern Intel processors can handle 256 floating point bits per core, while TR can only handle 128. Intel put in more hardware that's mostly useless except for prime projects. (And certain other benchmarks.)

 

[StefanR5R]

As usual, SETI.Germany have stats graphs:
https://www.seti-germany.de/pg_challenge/stats/team/201804/132

 

[Tejas_2]

I'm partially in, another 6 or so hours before reaching full speed.

 

[biodoc]

I finally got my P2 in but my Ryzen will stay on another project.

 

[Orange Kid]

The Ryzen run times are all over the board. The only thing that seems consistent is the CPU time. Got to wonder what is doing this, but I'm 1600 miles away, so it is what it is.

Run time(sec)CPU time(sec)
27,603.00 ...….52,756.61
13,660.00 ...…..52,784.94
41,281.00 .........51,845.69
6,895.00 ......….52,477.66
20,653.00 .........52,007.91
20,491.00 ......….52,099.45
13,828.00 ...…...52,846.86
46,401.00 .........52,492.86
6,738.00 ...........50,915.22
48,184.00 .........52,602.34
27,441.00 ...…...52,814.53
34,491.00 ......….52,502.28

 

[Orange Kid]

This is my app_config.xml
Don't know if the "<app></app>" part is needed, but it works for me.

<app_config>
    <app>
        <name>llrSR5</name>
       <max_concurrent>2</max_concurrent>
    </app>
    <app_version>
        <app_name>llrSR5</app_name>
        <cmdline>-t 8</cmdline>
        <avg_ncpus>8</avg_ncpus>
    </app_version>
</app_config>

 

[indydude345]

Would my app_config.xml be the same for a 6 core Ryzen with SMT? Just wondering.

 

[StefanR5R]

The Ryzen run times are all over the board. The only thing that seems consistent is the CPU time. Got to wonder what is doing this, but I'm 1600 miles away, so it is what it is.

Did you take these run times from your account at the PrimeGrid web site? If so, then those are wrong. The bookkeeping is totally buggy as soon as you run the PrimeGrid applications multithreaded. The PG admins say the upstream boinc code is at fault.

This is my app_config.xml
Don't know if the "<app></app>" part is needed, but it works for me.

The <app></app> part is not needed. For your setup, it does no harm but also has no obvious benefit.

Today in the morning I wondered why one of my hosts ran fewer tasks at once than it should. At a second look at the app_config which I edited yesterday late at night, I saw that I had a <project_max_concurrent> tag in there which was good for a prior race, but not anymore for this one. Long story short, better not use any <max_concurrent> or <project_max_concurrent> tags unless really needed for a special purpose.

Would my app_config.xml be the same for a 6 core Ryzen with SMT? Just wondering.

Replace each of the two "8" by "6" then.

But I am not aware of recent & precise tests whether or not SMT is beneficial for PrimeGrid LLR applications on Zen.

On recent Intel CPUs, HT is somewhat beneficial if the number of tasks per processor and the number of threads per task were set to the optimum, or near to it. (Conversely, if the number of tasks and threads are not at the optimum, HT may be detrimental.) On older Intel CPUs (how old?), HT is likely detrimental.

--------
ATP's top dog apparently hasn't switched all of his gear to SR5-LLR yet, and thus, @emoga and I are still in front of him. :-)
I wonder how long this will last.

 

[biodoc]

But I am not aware of recent & precise tests whether or not SMT is beneficial for PrimeGrid LLR applications on Zen.

I did some tests on my 2700x before the race and found when I disabled SMT in the bios and set the app_config for 8 mt cores, all the task downloads failed. I tried several times but could not get a successful task download. I turned SMT back on and and then task downloads were successful. Weird.

This was the error:

<core_client_version>7.4.23</core_client_version>
<![CDATA[
<message>
WU download error: couldn't get input files:
<file_xfer_error>
  <file_name>llrSR5_299573850</file_name>
  <error_code>-200 (wrong size)</error_code>
</file_xfer_error>

</message>
]]>

 

[Ken g6]

The Ryzen run times are all over the board. The only thing that seems consistent is the CPU time. Got to wonder what is doing this, but I'm 1600 miles away, so it is what it is.

PrimeGrid doesn't seem to know how to calculate run times correctly. If you download more than one WU at a time, the run times are added together for later WUs. If you set <avg_ncpus>1</avg_ncpus> on a lower-end processor, and set your BOINC client to only use "one core", it will probably compute correctly, while using however many cores you set in the app_config. I'm not sure how to handle it when you actually need multiple WUs at a time.

@biodoc, you should report that on PrimeGrid's forums.

 

[StefanR5R]

@biodoc, weird indeed. I have no idea how the processor config could affect file transfers.

 

[Markfw]

I have a E5-2683V3 now running it with an app_config set for 4 (28 threads/4). But its not using it. I said read config files, and read local files, and update, and still I have 28 threads running. 19 hours each

Edit: figured it out. Damn windows... defaults to hide extensions, so it was app_config.xml.txt. renamed, and reread options, and BAM, 7 tasks running. But 19 hours ETA on each.

 

[StefanR5R]

The boinc-client should gradually correct the ETA while the tasks go on running with the new config.

 

[StefanR5R]

Oh wait, did you restart the client though? (Or restarted the tasks at least?)

If you only re-read the config but let the tasks keep running, then only the <avg_ncpus> part will catch and tell the client to run fewer tasks at once. These tasks will still only run using a single processor thread.

The <cmdline> part of the config, which is the actual part which modifies the application to use the given number of threads per task, is only applied when a task is started. That is, it catches only as soon as

• the boinc client is started, or
• the client launches a new task, or
  
• the client resumes a task which was suspended (if the option "leave non-GPU tasks in memory while suspended" is off).

The safest course of action after PrimeGrid's app_config.xml was modified, is to shut down and restart the boinc-client.

 

[biodoc]

@Ken g6 @StefanR5R I'm using a beta bios for the MB so I'm trying SMT disabled on TN-Grid to see if I run into task download issues.

 

[biodoc]

I did some tests on my 2700x before the race and found when I disabled SMT in the bios and set the app_config for 8 mt cores, all the task downloads failed. I tried several times but could not get a successful task download. I turned SMT back on and and then task downloads were successful. Weird.

This was the error:

<core_client_version>7.4.23</core_client_version>
<![CDATA[
<message>
WU download error: couldn't get input files:
<file_xfer_error>
  <file_name>llrSR5_299573850</file_name>
  <error_code>-200 (wrong size)</error_code>
</file_xfer_error>

</message>
]]>

I can't reproduce this error. I was attaching to PrimeGrid for the first time on this machine so maybe it was a temporary issue.

Spoiler: event log of download error

13-Jun-2018 14:40:29 [PrimeGrid] Requesting new tasks for CPU
13-Jun-2018 14:40:31 [PrimeGrid] Scheduler request completed: got 8 new tasks
13-Jun-2018 14:40:33 [PrimeGrid] Started download of llrSR5_299574020
13-Jun-2018 14:40:33 [PrimeGrid] Started download of llrSR5_299573732
13-Jun-2018 14:40:34 [PrimeGrid] Incomplete read of 48.000000 < 5KB for llrSR5_299574020 - truncating
13-Jun-2018 14:40:34 [PrimeGrid] Incomplete read of 49.000000 < 5KB for llrSR5_299573732 - truncating
13-Jun-2018 14:40:34 [PrimeGrid] Finished download of llrSR5_299574020
13-Jun-2018 14:40:34 [PrimeGrid] Finished download of llrSR5_299573732
13-Jun-2018 14:40:34 [PrimeGrid] Started download of llrSR5_299573931
13-Jun-2018 14:40:34 [PrimeGrid] Started download of llrSR5_299569064
13-Jun-2018 14:40:34 [PrimeGrid] File llrSR5_299574020 has wrong size: expected 35, got 0
13-Jun-2018 14:40:34 [PrimeGrid] Checksum or signature error for llrSR5_299574020
13-Jun-2018 14:40:34 [PrimeGrid] File llrSR5_299573732 has wrong size: expected 36, got 0
13-Jun-2018 14:40:34 [PrimeGrid] Checksum or signature error for llrSR5_299573732
13-Jun-2018 14:40:35 [PrimeGrid] Incomplete read of 49.000000 < 5KB for llrSR5_299573931 - truncating
13-Jun-2018 14:40:35 [PrimeGrid] Incomplete read of 49.000000 < 5KB for llrSR5_299569064 - truncating
13-Jun-2018 14:40:35 [PrimeGrid] Finished download of llrSR5_299573931
13-Jun-2018 14:40:35 [PrimeGrid] Finished download of llrSR5_299569064
13-Jun-2018 14:40:35 [PrimeGrid] Started download of llrSR5_299573932
13-Jun-2018 14:40:35 [PrimeGrid] Started download of llrSR5_299574018
13-Jun-2018 14:40:35 [PrimeGrid] File llrSR5_299573931 has wrong size: expected 36, got 0
13-Jun-2018 14:40:35 [PrimeGrid] Checksum or signature error for llrSR5_299573931
13-Jun-2018 14:40:35 [PrimeGrid] File llrSR5_299569064 has wrong size: expected 36, got 0
13-Jun-2018 14:40:35 [PrimeGrid] Checksum or signature error for llrSR5_299569064
13-Jun-2018 14:40:36 [PrimeGrid] Incomplete read of 48.000000 < 5KB for llrSR5_299573932 - truncating
13-Jun-2018 14:40:36 [PrimeGrid] Incomplete read of 49.000000 < 5KB for llrSR5_299574018 - truncating
13-Jun-2018 14:40:36 [PrimeGrid] Finished download of llrSR5_299573932
13-Jun-2018 14:40:36 [PrimeGrid] Finished download of llrSR5_299574018
13-Jun-2018 14:40:36 [PrimeGrid] Started download of llrSR5_299573850
13-Jun-2018 14:40:36 [PrimeGrid] Started download of llrSR5_299573935
13-Jun-2018 14:40:36 [PrimeGrid] File llrSR5_299573932 has wrong size: expected 35, got 0
13-Jun-2018 14:40:36 [PrimeGrid] Checksum or signature error for llrSR5_299573932
13-Jun-2018 14:40:36 [PrimeGrid] File llrSR5_299574018 has wrong size: expected 36, got 0
13-Jun-2018 14:40:36 [PrimeGrid] Checksum or signature error for llrSR5_299574018
13-Jun-2018 14:40:37 [PrimeGrid] Incomplete read of 48.000000 < 5KB for llrSR5_299573850 - truncating
13-Jun-2018 14:40:37 [PrimeGrid] Incomplete read of 48.000000 < 5KB for llrSR5_299573935 - truncating
13-Jun-2018 14:40:37 [PrimeGrid] Finished download of llrSR5_299573850
13-Jun-2018 14:40:37 [PrimeGrid] Finished download of llrSR5_299573935
13-Jun-2018 14:40:37 [PrimeGrid] File llrSR5_299573850 has wrong size: expected 35, got 0
13-Jun-2018 14:40:37 [PrimeGrid] Checksum or signature error for llrSR5_299573850
13-Jun-2018 14:40:37 [PrimeGrid] File llrSR5_299573935 has wrong size: expected 35, got 0
13-Jun-2018 14:40:37 [PrimeGrid] Checksum or signature error for llrSR5_299573935

I running my 2700x with SMT off with 8 cores per task. If performance is poor compared to my Ivy bridge EP CPUs, I will abort the tasks and move the ryzen to TN-Grid.

 

[Howdy]

Having issues again with 1 machine, "validation inconclusive" on 5 results returned. I am letting it run a few more to see what happens. If it pops a few more results like that, I am shutting it down. It is running everything at stock clocks, CPU, RAM and even the GPU........I just can't figure this one out.

 

[Ken g6]

Day 1 stats:

Rank___Credits____Username
1______1054195____xii5ku
7______450085_____emoga
10_____361395_____crashtech
38_____125926___10esseeTony
56_____89537______biodoc
60_____87160______Ken_g6
64_____74782______Tejas
78_____62721______Howdy2u2
91_____50536______Orange Kid
121____32020______zzuupp
173____14565______indydude, Son of crashtech
216____4423_______markfw

Rank__Credits____Team
1_____3081257____Czech National Team
2_____2958538____SETI.Germany
3_____2597659____Aggie The Pew
4_____2407351____TeAm AnandTech
5_____1694937____Sicituradastra.
6_____687824_____BOINC@Poland
7_____608600_____BOINC@MIXI

Wow, we have a full TeAm here! And a really nice #4 ranking! Not to mention the three individuals in the top 10! Wow!

 

[crashtech]

Did you take these run times from your account at the PrimeGrid web site? If so, then those are wrong. The bookkeeping is totally buggy as soon as you run the PrimeGrid applications multithreaded...

Is the best way to determine run times to disable networking and view the finished WUs, then?