How is GIMPS' sustained throughput defined?

NBtarheel_33 · 2012-12-10, 09:57

http://www.mersenne.org/primenet/ is reporting that we are running at

99.127 TFLOP/sec over the last 24 hours,
107.404 TFLOP/sec over the last 7 days, and
98.339 TFLOP/sec over the last 30 days.

Which of these metrics would be reported as GIMPS' sustained throughput (if we were comparing to a supercomputer, for example)? Certainly the monthly throughput would be a fair candidate, possibly even the weekly throughput. I am wondering because, as far as I know, GIMPS has never broken the 100 TFLOP/sec barrier for sustained computation until just recently. In fact, I believe that when (it's getting closer) the monthly throughput breaks 100 TFLOP/sec, that will be the first time that's ever happened. So it seems as though this is certainly a milestone for GIMPS, and would seem to indicate that our ranks are increasing, which is a good thing. (Note another recent milestone: there are now over 700,000 registered CPUs crunching for GIMPS, many of them likely multi-core systems!)

Quote:

Originally Posted by Peter Nelson, November 2005, when GIMPS' sustained throughput was struggling to reach 20 TFLOPS

...I think we can go well beyond 100TFLOPS one day.

Personally, I think we will go petascale by 2020, with 100-million-digit tests becoming 8- to 12-week, mainstream assignments by 2017. All hail GPUs!

Now if only George could port Prime95 to smartphones and tablets. The Samsung Galaxy IV is slated to have a quad-core processor and over 1GB of RAM. P-1 on a *phone* is becoming theoretically feasible!

c10ck3r · 2012-12-10, 14:24

P-1 on a phone would kill the battery. It'd overheat, and if it didn't, it wouldn't last more than an hour.

science_man_88 · 2012-12-10, 14:27

Quote:

Originally Posted by c10ck3r

P-1 on a phone would kill the battery. It'd overheat, and if it didn't, it wouldn't last more than an hour.

plug it into it's charger ? if the battery goes low, as to the overheat scenario I'm unsure

firejuggler · 2012-12-10, 15:31

the easiest solution would be P-1 that take less than one hour.

henryzz · 2012-12-10, 16:37

I did a LL test on a phone that took over 24 hours once.

kracker · 2012-12-10, 19:32

I have a 1.3 GHz quadcore on a tablet..... wish it could be used somehow.

NBtarheel_33 · 2012-12-11, 20:29

The interesting thing is that, IIRC, Android is very similar to (a derivative of?) Linux. So it makes me wonder how hard it would actually be to port something like mprime to the Android platform. Obviously, cooling would be a major issue, as there are no heatsinks or fans in a phone, let alone how hot the battery would be getting. Perhaps a dry ice cooling solution?

Back on topic: still haven't breached 100 TFLOP/s sustained for a month. Numbers right now are 112.127 today/105.327 week/98.674 month.

Dubslow · 2012-12-11, 21:48

Quote:

Originally Posted by NBtarheel_33

The interesting thing is that, IIRC, Android is very similar to (a derivative of?) Linux. So it makes me wonder how hard it would actually be to port something like mprime to the Android platform. Obviously, cooling would be a major issue, as there are no heatsinks or fans in a phone, let alone how hot the battery would be getting. Perhaps a dry ice cooling solution?

Android is a derivative of Linux; for a while the changes weren't being back ported to Linux, but IIRC recently Torvalds told everyone to STFU about Google and just accept their code. So now Android uses the "vanilla" Linux kernel as its code base.

NBtarheel_33 · 2012-12-13, 05:04

2012-12-13 0400 GMT: ~99.5 TFLOP/s sustained for the last 30 days.
2012-12-13 0500 GMT: ~99.7 TFLOP/s sustained for the last 30 days.

Will probably break 100 in the next few hours. For the last 24 hours, we are at 130+ TFLOP/s.

LaurV · 2012-12-13, 05:20

Quote:

Originally Posted by NBtarheel_33

The interesting thing is that, IIRC, Android is very similar to (a derivative of?) Linux. So it makes me wonder how hard it would actually be to port something like mprime to the Android platform. Obviously, cooling would be a major issue, as there are no heatsinks or fans in a phone, let alone how hot the battery would be getting. Perhaps a dry ice cooling solution?

Quote:

Originally Posted by Dubslow

Android is a derivative of Linux; for a while the changes weren't being back ported to Linux, but IIRC recently Torvalds told everyone to STFU about Google and just accept their code. So now Android uses the "vanilla" Linux kernel as its code base.

The most difficult would be porting the high-optimized-for-x86 assembler part to ARM or whatever CPU's those mobiles have (TI's Neon GPU's or even Tegra3?

)

NBtarheel_33 · 2012-12-13, 10:51

As of 2012-12-13 1000 GMT, PrimeNet is reporting that we have sustained 100.147 TFLOP/s for the last 30 days!!

More interestingly, in the last 24 hours, we are outperforming our theoretical maximum capacity: 145.277 TFLOP/s actual with only 130.277 TFLOP/s max capacity!

Onward and upward...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
another defined sequence	MattcAnderson	Miscellaneous Math	4	2017-07-05 08:05
Budget PC Throughput	Rodrigo	Hardware	14	2011-09-26 10:16
SZPTR - whare is it defined	tichy	Software	4	2010-12-22 09:02
how is the throughput calculated?	ixfd64	PrimeNet	5	2008-05-21 13:39
My throughput does not compute...	petrw1	Hardware	9	2007-08-13 14:38

2012-12-10, 14:24	#2
c10ck3r Aug 2010 Kansas 547 Posts	P-1 on a phone would kill the battery. It'd overheat, and if it didn't, it wouldn't last more than an hour.

2012-12-10, 15:31	#4
firejuggler Apr 2010 Over the rainbow 2·1,259 Posts	the easiest solution would be P-1 that take less than one hour.

2012-12-10, 16:37	#5
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2×2,897 Posts	I did a LL test on a phone that took over 24 hours once.

2012-12-10, 19:32	#6
kracker ἀβουλία "Mr. Meeseeks" Jan 2012 California, USA 3²×241 Posts	I have a 1.3 GHz quadcore on a tablet..... wish it could be used somehow.

2012-12-11, 20:29	#7
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 10001011011₂ Posts	The interesting thing is that, IIRC, Android is very similar to (a derivative of?) Linux. So it makes me wonder how hard it would actually be to port something like mprime to the Android platform. Obviously, cooling would be a major issue, as there are no heatsinks or fans in a phone, let alone how hot the battery would be getting. Perhaps a dry ice cooling solution? Back on topic: still haven't breached 100 TFLOP/s sustained for a month. Numbers right now are 112.127 today/105.327 week/98.674 month.

2012-12-13, 05:04	#9
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5×223 Posts	2012-12-13 0400 GMT: ~99.5 TFLOP/s sustained for the last 30 days. 2012-12-13 0500 GMT: ~99.7 TFLOP/s sustained for the last 30 days. Will probably break 100 in the next few hours. For the last 24 hours, we are at 130+ TFLOP/s.

2012-12-13, 10:51	#11
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 1115₁₀ Posts	As of 2012-12-13 1000 GMT, PrimeNet is reporting that we have sustained 100.147 TFLOP/s for the last 30 days!! More interestingly, in the last 24 hours, we are outperforming our theoretical maximum capacity: 145.277 TFLOP/s actual with only 130.277 TFLOP/s max capacity! Onward and upward...