I have no idea of how many operations are included in each of those
reports. 113 milliseconds for an integer operation would be completely
unacceptable as this is only 9 integer operations per second. I expect
that this is the time to run some number of integer operations. I have no
idea if the count of floating operations for the test is the same as that
for the integer test.
jm7
Petr Hájek
<hajek.p...@gmail
.com> To
[email protected],
10/28/2009 12:41 "[email protected]"
PM <[email protected]>
cc
Subject
Re: [boinc_dev] BOINC for Mobile
Phones - please test on your Java
phone
You guys I tested how long takes +,-,*, / of 1.000.000 INT and 1.000.000 of
DOUBLES on JME with this program - test on your phone too *please* !
http://java.wmhelp.cz/Downloads/SpeedTester.zip
My times for Nokia e60 (200+ MHz) are:
INTs: 113, 158, 162, 416 ms
DOUBLEs: 969, 1012, 823, 876 ms
As we can ALL see, working with DOUBLEs is NOT 100x - 1000x times slower
damn!
2009/10/28, Lynn W. Taylor <[email protected]>:
Significant computing power is increasingly becoming a throw-away
commodity.
... and with video being pushed to phones, I'd expect next year's phones
to have significant CPU power.
I don't think time matters, except that in order for a phone to complete
a (2 or 3 year??) CPDN work unit, it has to survive 2 to 3 years.
-- Lynn
[email protected] wrote:
Only a very few projects will be able to create smaller tasks.
CPDN tasks cannot be reduced without shovelling about 1GB of data from
the
device back to the server.
s...@h tasks are already reduced to the minimum.
...
With no FPU, the increase in time is on the order of times 1000. Which
would mean that the crunch times would have to be reduced by 1000, or
the
deadline would have to be increased by a factor of 1000. �...@h for
example
would have to increase the deadlines from one month to 100 years. Or
the
data span would have to be reduced from 115 seconds of data to 0.1
seconds
of data (the overlap is currently 15 seconds of data).
Integer only projects such as (possibly) prime grid do not suffer from
this
problem.
Non-CPU intensive projects also do not suffer from the problem.
You should look to those types of projects for possibilities.
jm7
Petr Hájek
<hajek.p...@gmail
.com> To
Sent by: [email protected]
<boinc_dev-bounce cc
[email protected]
u> Subject
Re: [boinc_dev] BOINC for Mobile
Phones - please test on your Java 10/28/2009 09:57
phone AM
OK, for the 3rd time:
"2. There will be absolutely need for different and smaller units so it
may
be counted in few hours / days on typical phone / PDA"
2009/10/28, [email protected] <[email protected]>:
CPDN has long deadlines because it has correspondingly long crunch
times.
An 800 MHz computer with an FPU (and CPDN uses the FPU) takes well in
excess of 9 months to crunch the data running 24/7. A 600 MHz device
with
no FPU will not finish within the lifetime of the phone - even running
24/7.
Will this always be true? I cannot be certain - ever is an awfully
long
time.
Deadlines vaguely track crunch times on most projects. Long deadlines
usually have correspondingly long
jm7
Petr Hájek
<hajek.p...@gmail
.com>
To
Sent by: [email protected]
<boinc_dev-bounce
cc
[email protected]
u>
Subject
Re: [boinc_dev] BOINC for Mobile
Phones - please test on your
Java
10/28/2009 09:48 phone
AM
1. Some projects has LONG deadlines - Climate for example.
2. There will be absolutely need for different and smaller units so it
may
be counted in few hours / days.
2009/10/28, [email protected] <[email protected]>:
When you are not using the keypad and the phone is not active, the
processor is probably running at about 6 MHz. With no FPU.
jm7
"Lynn W. Taylor"
<[email protected]>
Sent by:
To
<boinc_dev-bounce Carl Christensen
[email protected] <[email protected]>
u>
cc
[email protected]
Subject
10/27/2009 02:54 Re: [boinc_dev] BOINC for
Mobile
PM Phones - please test on your
Java
phone
I keep thinking that there are a lot of cell phones out there, and a
lot
of untapped potential.
The one in my pocket (Palm Pre) is running some variant of the ARM
processor at something like 600 MHz, which is a nontrivial amount of
CPU.
Palm goofed on the battery (I can go two days, tops), but the rest of
the phone, including WebOS, is pretty cool.
Cell phones as a group are probably second only to smart cards in the
total number of available clock cycles.
-- Lynn
Carl Christensen wrote:
I don't quite understand the bashing of this guy's mobile project;
there
was that "boincoid" a year or two ago which was in vogue, and IMHO
the
same
ones bashing the "usefulness" of mobiles are the ones crowing about
how
great GPU's & CUDA & Sony Playstations etc are (completely ignoring
the
fact that 99.99999% of real-world science apps won't run on it). Not
to
mention that there's all sorts of dubious-benefit computer sciencey
stuff
out there trying to turn boinc into some god-awful grid mess. so I'm
willing to keep an open mind about it (and GPU's & grids ;-).
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
--
S Pozdravem
Petr Hájek
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
--
S Pozdravem
Petr Hájek
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
--
S Pozdravem
Petr Hájek
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
