Which projects use double precision on GPUs?

Message boards : GPUs : Which projects use double precision on GPUs?
Message board moderation

To post messages, you must log in.

AuthorMessage
Peter Hucker
Avatar

Send message
Joined: 6 Oct 06
Posts: 609
United Kingdom
Message 94702 - Posted: 5 Jan 2020, 11:54:01 UTC

I know Milkyway is double precision for GPUs. Any other projects? I'm looking for a secondary project to run when Milkyway doesn't have tasks available, and would prefer one that's double precision, as I have two new cards coming that are very good at it.
ID: 94702 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 546
United States
Message 94767 - Posted: 8 Jan 2020, 3:53:30 UTC - in response to Message 94702.  

I know Milkyway is double precision for GPUs. Any other projects? I'm looking for a secondary project to run when Milkyway doesn't have tasks available, and would prefer one that's double precision, as I have two new cards coming that are very good at it.

I don't know, but Milkyway, Einstein, Seti, and moo! always have jobs available, and are most widely supported.
Perhaps checking each project, if they support a variety of hardware, and asking on those forums, will get you closer to your answer?
Then you can update us all here, because I also would like to know the answer.
ID: 94767 · Report as offensive
Peter Hucker
Avatar

Send message
Joined: 6 Oct 06
Posts: 609
United Kingdom
Message 94774 - Posted: 8 Jan 2020, 20:09:18 UTC - in response to Message 94767.  
Last modified: 8 Jan 2020, 20:11:58 UTC

I know Milkyway is double precision for GPUs. Any other projects? I'm looking for a secondary project to run when Milkyway doesn't have tasks available, and would prefer one that's double precision, as I have two new cards coming that are very good at it.

I don't know, but Milkyway, Einstein, Seti, and moo! always have jobs available, and are most widely supported.
Perhaps checking each project, if they support a variety of hardware, and asking on those forums, will get you closer to your answer?
Then you can update us all here, because I also would like to know the answer.


Tomorrow (time permitting), I will install my two new cards. I will then have cards with different 32:64 ratios, so I can compare the speed each project runs on the different cards and work out which must be using double precision. I'll respond back here once I've done the test.

I already know Milkyway is double and Einstein is single (from when I did a comparison with two cards before - but one of them expired....)
ID: 94774 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14157
Netherlands
Message 94775 - Posted: 8 Jan 2020, 20:38:49 UTC - in response to Message 94774.  

As far as I know, the only project requiring double precision floating point is Milkyway. All of the others run single precision float only.
Please do not private message me for tech support, these will be ignored!
ID: 94775 · Report as offensive
Peter Hucker
Avatar

Send message
Joined: 6 Oct 06
Posts: 609
United Kingdom
Message 94776 - Posted: 8 Jan 2020, 20:52:14 UTC - in response to Message 94775.  

As far as I know, the only project requiring double precision floating point is Milkyway. All of the others run single precision float only.


I'll find out for sure tomorrow. I'll try the test anyway, since I'm sure Einstein used to use double and Milkyway used to use single.
ID: 94776 · Report as offensive
Peter Hucker
Avatar

Send message
Joined: 6 Oct 06
Posts: 609
United Kingdom
Message 94799 - Posted: 9 Jan 2020, 20:26:52 UTC

Testing complete:

List created from https://boinc.berkeley.edu/projects.php
Testing all projects with AMD logo, comparing speed difference between RX 560 and R9 280X, expecting factor of 1.57 for single precision or 6.28 for double precision:

Milkyway = double
Numberfields = double (but uses quite a bit of memory - not wise for 8GB machines)

Prime = mixed? (3.4 times faster on R9 280X) (large tasks, unsuitable as 0 priority backup project)

Einstein = single
Collatz conjecture = single
Seti = single

Amicable numbers = unable to test - 8GB RAM on test machine insufficient to run tasks
Moo wrapper = unable to test - won't run on RX 560 (only uses the CPU and leaves GPU idle)

Enigma = no tasks available
LHC = no tasks available
ID: 94799 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 546
United States
Message 94831 - Posted: 10 Jan 2020, 22:40:40 UTC

Some projects, like the collatz, might show high numbers, but might still be single precision.
ID: 94831 · Report as offensive
Peter Hucker
Avatar

Send message
Joined: 6 Oct 06
Posts: 609
United Kingdom
Message 94837 - Posted: 10 Jan 2020, 22:55:11 UTC - in response to Message 94831.  

Some projects, like the collatz, might show high numbers, but might still be single precision.


I ran similar tasks on two GPUs at once. The better GPU is several times faster at double precision, and only twice as fast at single. So if the task goes twice as quick, it's single, if it goes several times faster, it's double. Must be an easier way to check though....
ID: 94837 · Report as offensive

Message boards : GPUs : Which projects use double precision on GPUs?

Copyright © 2020 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.