PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Wieferich and Wall-Sun-Sun Prime Search : No GPU tasks? (Message 151401)
Posted 21 days ago by niktak11
Each one of these work units take about 4min 45 sec to process on my RTX 3090 FE GPU. Was wondering why I can only get about 4 or 5 workunits at a time when requesting more workunits. The faster GPUs should be allowed to get more of them at one time. I have my prefs set to get 10 days worth with 10 days in reserve, yet I only get usually 4 or 5 at a time. This is something that the devs need to look into.

Most people only want 1-2 tasks at a time to increase the chances that they are a WW discover.
2) Message boards : Number crunching : PrimeGrid's 16th Birthday Challenge (Message 150642)
Posted 103 days ago by niktak11
My 3900X is set up to run 4x 3 threads. Getting 14 to 14-and-a-half hours per WU. Ran a couple ahead of time at 1x 12 threads - just over 5 hours on those. Not nearly as much throughput that way. What times are you seeing running 2x 6 threads?

I do have each 3 thread task tied to a CCX using taskset in linux. All cores run at about 3870 MHz fully loaded like this.

And is it just me, or does the birthday cake banner only have 15 candles?

I'm getting around 37k seconds per task if running with 6 cores each. About 100k seconds per task at 3 cores each. Windows 10.
3) Message boards : Number crunching : PrimeGrid's 16th Birthday Challenge (Message 150619)
Posted 105 days ago by niktak11
On a 3900x, 2x 6 core tasks seems to be quite a bit better with ESP than 4x 3 core tasks. On most projects it seems to be the opposite.
4) Message boards : Number crunching : √Čvariste Galois's Birthday Challenge (Message 144551)
Posted 340 days ago by niktak11
Do you just allow them to run on only even cores?
5) Message boards : Number crunching : √Čvariste Galois's Birthday Challenge (Message 144543)
Posted 340 days ago by niktak11
it seems like just using the Prim95 benchmark isn't sufficient? Do the results deviate from real world applications?

It is a starting point, and there can be differences.

For example, I just did a fresh install on a 3700X system. I know it should be best with 1 task per core (although I have seen elsewhere that using SMT might improve it further, I haven't tested this). When I set it up with 1 task per core, I saw Windows being stupid, putting two tasks on the 1st core, with the last core idle. I had to manually set affinity to correct this.

From experience, Windows needs a nudge at times when running 1 thread per task on HT/SMT enabled CPUs. Something changes when running more than one thread per task and that requires less manual intervention.

With my 3900x, it seems that I get higher throughput with this project by running 3 cores/task (4 at once) than a single task per core (12 at once).
6) Message boards : Number crunching : Alan Turing's Birthday Challenge (Message 141026)
Posted 460 days ago by niktak11
I can't seem to get multiple tasks per GPU working. I have this in my app_config.xml file. Everything else in that file is working fine.

<app>
<name>pps_sr2sieve</name>
</app>
<app_version>
<app_name>pps_sr2sieve</app_name>
<plan_class>cudaPPSsieve</plan_class>
<avg_ncpus>0.1</avg_ncpus>
<ngpus>0.5</ngpus>
</app_version>


try this:

<app_config>
<app>
<name>pps_sr2sieve</name>
<max_concurrent>2</max_concurrent>
</app>
<app_version>
<app_name>pps_sr2sieve</app_name>
<plan_class>cudaPPSsieve</plan_class>
<avg_ncpus>0.01</avg_ncpus>
<ngpus>.5</ngpus>
</app_version>
</app_config>

It turns out Windows changed my existing app_config.xml file to app_config.xml.txt lol
7) Message boards : Number crunching : Alan Turing's Birthday Challenge (Message 140997)
Posted 460 days ago by niktak11
I can't seem to get multiple tasks per GPU working. I have this in my app_config.xml file. Everything else in that file is working fine.

<app>
<name>pps_sr2sieve</name>
</app>
<app_version>
<app_name>pps_sr2sieve</app_name>
<plan_class>cudaPPSsieve</plan_class>
<avg_ncpus>0.1</avg_ncpus>
<ngpus>0.5</ngpus>
</app_version>
8) Message boards : Number crunching : Tour de Primes 2020 (Message 137734)
Posted 593 days ago by niktak11
I finally found a mega prime
9) Message boards : Generalized Fermat Prime Search : GFN-16 MEGA Prime Search starts March, 14, 2019 (Message 136979)
Posted 604 days ago by niktak11
[...]If this follows other LLR work, I'd expect optimal running configuration to be 2 cores per task for pretty much all recent CPUs.[...]

I turned off HT on my 3970X, so it's running 36 threads on 36 cores. Then I ran an experiment, running just a single task:

8 cores: 1:41
4 cores: 1:29
2 cores: 1:55

Those time are really good, and it seems that 2 cores is the best choice on this machine for maximum work. But, once I cranked up the number of tasks to use all the cores, 2 per task, times dropped back to the 4-6 hour range. Not really sure why this would be, considering it has 128mb of L3?



I noticed the same thing happening to me with my 3900x. Running 1 task with 3 threads is much faster than running 4 tasks with 3 threads each. I'm guessing windows isn't smart enough to keep each task confined to a single CCX and it's wasting a lot of time getting data from cache in other CCXs. The core clock is not much different for me in either scenario.
10) Message boards : AP26 - AP27 Search : AP27 FOUND!!! (Message 133219)
Posted 732 days ago by niktak11
Time to start looking for an AP28


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 4.17, 4.60, 4.12
Generated 27 Sep 2021 | 3:58:45 UTC