Flops

log in

Advanced search

Message boards : Science : Flops

Author Message
74861359908
 
Send message
Joined: 11 Mar 18
Posts: 2
Credit: 60,031
RAC: 0
Total hours: 183,338
Message 6456 - Posted: 18 Feb 2019, 22:17:27 UTC

What about an idea for a statistic, which would be in concern to worths [values] of "Flops [GigaFlops, GFLOPS]"? Would this be possible to do?

Dirk Broer
       
Avatar
Send message
Joined: 3 Nov 10
Posts: 109
Credit: 2,973,513
RAC: 823
Total hours: 8,223,614
Message 6728 - Posted: 14 Jun 2019, 7:51:16 UTC - in response to Message 6456.
Last modified: 14 Jun 2019, 7:52:31 UTC

What about an idea for a statistic, which would be in concern to worths [values] of "Flops [GigaFlops, GFLOPS]"? Would this be possible to do?


What did you have in mind? The power of a CPU or GPU can be expressed in FLOPS, and the WU are handed out containing a FLOP value too.

The CPU/GPU power is not 'hard' value though, as it is dependent on OS and installed software. Likewise the value given for the WUs most likely suffers the same problem.

You might ultimately arrive at a table not unlike given under 'results' at WUProp.
____________

74861359908
 
Send message
Joined: 11 Mar 18
Posts: 2
Credit: 60,031
RAC: 0
Total hours: 183,338
Message 6731 - Posted: 17 Jun 2019, 9:40:33 UTC - in response to Message 6728.

FLOPS [Floating Point Operations Per Second]



The idea is: in a statistic table which is concerning to FLOPS per time, different computers [with different hardware and/or different operating systems] could be compared.

An example: there is a computer with a CPU with 1 GHz, which is doing workunits for one special project with one special application [the same amount of FLOPS needed to be calculated] and it is running 24 hours per day. That computer could be compared with another computer with a CPU with [for example] 4 GHz, which is doing the same workunits and it is running 24 hours per day, also.

In a table which is concerning only to the time [used by the workunits], there would be listed that 24 hours after 1 day were used for both computers for this project [to multiply by the number of cores of the CPU which were used]. But the computer with the CPU with 4 GHz, which can do more FLOPS, should have done more workunits per day [if the workunits need, for example, about 1 hour on the 4 GHz CPU]. In a table which is concerning only to the time, the differences in FLOPS [for example "GFlops"] which were done can not be seen.

There is different hardware, there are different operating systems, there is "VirtualBox".

The idea is, whether a statistic table concerning to FLOPS per time would be possible to do.

Profile Conan
       
Avatar
Send message
Joined: 28 Mar 10
Posts: 565
Credit: 1,173,845
RAC: 186
Total hours: 3,242,375
Message 6732 - Posted: 17 Jun 2019, 13:06:54 UTC

Your idea may not be fully possible with BOINC.

Not every project uses FLOPS, some are only Integer based, which is one reason that BOINC's scoring system is, and always has been, flawed.

It is one reason that fixed credits were introduced by a lot of projects to get round the Benchmark problems.

Using the BOINC flop scoring benchmark does not always give what you think it should.

WUProps results tables show most of what you ask already, maybe just not in the format that you want.

Conan
____________


Post to thread

Message boards : Science : Flops


Home | My Account | Message Boards | Results


Copyright © 2024 Sebastien