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 : WW project is live (Message 146766)
Posted 279 days ago by Profile mfl0pProject donor
I'll give my developer input on the CPU credit discussion.

The WW program was designed for GPUs. The CPU program is an adaption of the GPU algorithm so people can crunch on their CPUs if they don't have access to a GPU. At PrimeGrid, this a has been the tradition to allow as many people to participate as possible.

The CPU app can be made much faster with a different algorithm, however, in order for the checksums to match for workunit validation, it must process the workunit like the GPU program. This results in what seems like "low" CPU credits for WW.
2) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Slow speed on RX 5700XT (Message 146571)
Posted 285 days ago by Profile mfl0pProject donor

The only difference I could spot so far is WWocl version 1.00/1.0.2 vs 1.02/1.0.3 - does this explain the vastly shorter runtime?


Yes, gpu load should be higher with the new version.
3) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Adjusting GPU load / screen refresh (Message 146512)
Posted 286 days ago by Profile mfl0pProject donor
The new version 1.03 for Windows/Linux has an optional command line parameter that can be used to tune screen refresh and GPU load.

This option isn't needed or supported for NVIDIA Pascal or newer GPUs on Windows 10.
Windows 10 uses different GPU scheduling on Pascal+ GPUs making this feature unnecessary.

The -ws # command line option is for the user to adjust GPU load higher for faster computation or lower for better screen refresh rate. Usage: -ws 10 for default, -ws 1 for lowest GPU load, -ws 100 for highest GPU load. Worksize can be any number from 1 to 100.

You will need to make an app_config.xml in the BOINC\projects\www.primegrid.com directory with the following for Nvidia:
<app_config> <app_version> <app_name>ww</app_name> <plan_class>OCL_cuda_WW</plan_class> <cmdline>-ws 10</cmdline> </app_version> </app_config>


Or for AMD:
<app_config> <app_version> <app_name>ww</app_name> <plan_class>opencl_ati_WW</plan_class> <cmdline>-ws 10</cmdline> </app_version> </app_config>


edited, thanks stream
4) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Slow speed on RX 5700XT (Message 145934)
Posted 296 days ago by Profile mfl0pProject donor
A future update to the program should see increased GPU utilization percentage.

The 20 compute units info comes directly from AMD's driver. You can see this in GFN's stderr.txt also, where the compute units are listed at 20. The programs simply report what AMD's video driver is telling it. In this case, incorrect information.
5) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Error's running GPU WU's !!! (Message 145918)
Posted 296 days ago by Profile mfl0pProject donor
The linux 960 looks like it needs opencl installed. Try installing ocl-icd-opencl-dev
6) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Any idea on WU times CPU vs. GPU? (Message 145754)
Posted 298 days ago by Profile mfl0pProject donor
Will the CPU app use any of those "fancy" instruction sets like AVX or SSE, or will it just use regular, plain x86?


No vector operations at this time.
7) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Any idea on WU times CPU vs. GPU? (Message 145657)
Posted 301 days ago by Profile mfl0pProject donor
Mike is correct, 64 bit integer math is used on the gpu. The standard double checking is used because each WW workunit is testing millions of primes. A LLR2 workunit is testing just one number for primality, allowing for a fast dc test.
8) Message boards : Number crunching : WW testing (Message 144887)
Posted 327 days ago by Profile mfl0pProject donor
you need to run the program 3 times, with the search range parameters like this:

./WW_opencl_macintel64 461168595e10 461168602e10
expected checksum: 67029FBFBC1AEFB0

./WW_opencl_macintel64 1 6e10
expected checksum: 2F32CAD23BA5B1CA

./WW_opencl_macintel64 227630e10 227636e10
expected checksum: B8B040FA3813FEB2

to run on the AMD gpu and not Intel, try making a file called init_data.xml containing
<app_init_data> <gpu_type>ATI</gpu_type> <gpu_device_num>0</gpu_device_num> </app_init_data>

and put it in the same directory.
9) Message boards : Number crunching : WW testing (Message 144504)
Posted 337 days ago by Profile mfl0pProject donor
Mac OpenCL app is available for testing.
10) Message boards : Number crunching : WW testing (Message 144433)
Posted 338 days ago by Profile mfl0pProject donor

I found that it's fine with a parameter up to 6e8:

Checkpoint range size 599999873 processed at 600.00 Million numbers per second.
Estimated time remaining: 0s
Workunit complete. Average speed: 31.33 Million primes per second.
Checksum: 3A693971FB152B22

It's with 6e9 up it fails.


The error should be fixed now, if you could test the new build that would be appreciated.
I'm waiting on a new build of the Mac OpenCL app and will post to this thread when available.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.10, 0.03, 0.01
Generated 23 Sep 2021 | 9:25:13 UTC