Nice and steady 120+ TFlop estimate.

Message boards : Number crunching : Nice and steady 120+ TFlop estimate.

To post messages, you must log in.

AuthorMessage
Profile Chilean
Avatar

Send message
Joined: 16 Oct 05
Posts: 711
Credit: 26,694,507
RAC: 0
Message 68342 - Posted: 2 Nov 2010, 15:03:29 UTC
Last modified: 2 Nov 2010, 15:03:57 UTC

It's been there for a while now :D! A little bit more and we'll reach the 130 TFlop mark. And I don't even have my strongest PC on!
ID: 68342 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile rochester new york
Avatar

Send message
Joined: 2 Jul 06
Posts: 2842
Credit: 2,020,043
RAC: 0
Message 69607 - Posted: 5 Feb 2011, 23:33:32 UTC - in response to Message 68342.  

It's been there for a while now :D! A little bit more and we'll reach the 130 TFlop mark. And I don't even have my strongest PC on!


boinc shows 47 t flops rosetta site says well over 100
http://boincstats.com/stats/project_graph.php?pr=rosetta&view=countries
ID: 69607 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 69608 - Posted: 6 Feb 2011, 0:58:06 UTC - in response to Message 69607.  
Last modified: 6 Feb 2011, 0:58:23 UTC

boinc shows 47 t flops rosetta site says well over 100


I would guess that the BOINC Stats estimate is based on granted credit. If so, Rosetta's tendency to grant less credit per CPU time than other projects could be skewing the estimate.
ID: 69608 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Chilean
Avatar

Send message
Joined: 16 Oct 05
Posts: 711
Credit: 26,694,507
RAC: 0
Message 69609 - Posted: 6 Feb 2011, 1:43:58 UTC

The estimate on this site is apparently derived from the "Credits last 24h". Does this mean that 1 million R@H credits in one day is the equivalent to 1 TFLOP? Which means... that 1 TFLOP is 1 million credits divided by the seconds in one day...

Am I going in the right track?
ID: 69609 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 69611 - Posted: 6 Feb 2011, 2:40:27 UTC - in response to Message 69609.  

The estimate on this site is apparently derived from the "Credits last 24h".


I never noticed that before.

I would guess then that the BOINC stats figure is probably an average of a longer period of time. Maybe 30 or 60 days. If so, 30 days ago was the big crash and the speed of Rosetta was 0 TFlops. Once that clears out of the BOINC stats calculation then the two figures should arrive at a similar level.
ID: 69611 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mad_Max

Send message
Joined: 31 Dec 09
Posts: 207
Credit: 23,103,332
RAC: 13,199
Message 69618 - Posted: 9 Feb 2011, 1:07:30 UTC - in response to Message 69609.  

The estimate on this site is apparently derived from the "Credits last 24h". Does this mean that 1 million R@H credits in one day is the equivalent to 1 TFLOP? Which means... that 1 TFLOP is 1 million credits divided by the seconds in one day...

Am I going in the right track?

Rosetta main page (Server Status) calculate FLOPS estimate from "Credits last 24h" (1 million credits per day = 10 TFLOPS @ 24/7)
Boinc stat (http://boincstats.com/stats/project_graph.php?pr=rosetta&view=credit) calculate FLOPS estimate from Recent average credit (RAC - sliding average from last ~30 days), but shows only half of it (50%) cuz 1 gobblestone (credit) composed from 50% of integer speed and 50% floating point speed.
So major(2х) difference is (TFLOPS only) vs (TFLOPS + TIOPS).
Minor difference is RAC (last 30 days) vs (last 24h) and affected by last big downtime of project in January.
Then RAC stabilise BOINC will show ~ 60-65 TFLOPS, and R@H ~ 120-130 TFLOPS.
ID: 69618 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile rochester new york
Avatar

Send message
Joined: 2 Jul 06
Posts: 2842
Credit: 2,020,043
RAC: 0
Message 69620 - Posted: 9 Feb 2011, 20:03:53 UTC - in response to Message 69618.  
Last modified: 9 Feb 2011, 20:53:17 UTC

The estimate on this site is apparently derived from the "Credits last 24h". Does this mean that 1 million R@H credits in one day is the equivalent to 1 TFLOP? Which means... that 1 TFLOP is 1 million credits divided by the seconds in one day...

Am I going in the right track?

Rosetta main page (Server Status) calculate FLOPS estimate from "Credits last 24h" (1 million credits per day = 10 TFLOPS @ 24/7)
Boinc stat (http://boincstats.com/stats/project_graph.php?pr=rosetta&view=credit) calculate FLOPS estimate from Recent average credit (RAC - sliding average from last ~30 days), but shows only half of it (50%) cuz 1 gobblestone (credit) composed from 50% of integer speed and 50% floating point speed.
So major(2х) difference is (TFLOPS only) vs (TFLOPS + TIOPS).
Minor difference is RAC (last 30 days) vs (last 24h) and affected by last big downtime of project in January.
Then RAC stabilise BOINC will show ~ 60-65 TFLOPS, and R@H ~ 120-130 TFLOPS.



they cant both be right..... boinc shows active computers
how long do you think before it stabilizes??
ID: 69620 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Aegis Maelstrom

Send message
Joined: 29 Oct 08
Posts: 61
Credit: 2,137,555
RAC: 0
Message 69621 - Posted: 10 Feb 2011, 12:50:52 UTC - in response to Message 69620.  
Last modified: 10 Feb 2011, 12:53:26 UTC

The estimate on this site is apparently derived from the "Credits last 24h". Does this mean that 1 million R@H credits in one day is the equivalent to 1 TFLOP? Which means... that 1 TFLOP is 1 million credits divided by the seconds in one day...

Am I going in the right track?

Rosetta main page (Server Status) calculate FLOPS estimate from "Credits last 24h" (1 million credits per day = 10 TFLOPS @ 24/7)
Boinc stat (http://boincstats.com/stats/project_graph.php?pr=rosetta&view=credit) calculate FLOPS estimate from Recent average credit (RAC - sliding average from last ~30 days), but shows only half of it (50%) cuz 1 gobblestone (credit) composed from 50% of integer speed and 50% floating point speed.
So major(2х) difference is (TFLOPS only) vs (TFLOPS + TIOPS).
Minor difference is RAC (last 30 days) vs (last 24h) and affected by last big downtime of project in January.
Then RAC stabilise BOINC will show ~ 60-65 TFLOPS, and R@H ~ 120-130 TFLOPS.



they cant both be right..... boinc shows active computers
how long do you think before it stabilizes??


As Mad Max nicely explained, there are two major differences between methodologies on BoincStats and Rosetta sites.

Firstly, boincstats site uses RAC, which takes an average from the last thirty days. Thus, it takes 30 days since the come back after outage to "stabilize".

On the contrary, Rosetta site uses only last 24 hours.

Secondly, boincstats uses a different translation between cobblestones (BOINC points) and FLOPSes. It seems to assume that 1 cobblestone = 5 MegaFLOPSes.
On the other side, Rosetta estimates 1 cobblestone = 10 MegaFLOPSes.

Regarding this second difference - which approach is more reliable?
My guess is that Rosetta team knows better the nature of their calculations and can adjust the multiplier (cobblestones / FLOPSes).
On the other hand, boincstats makes one general assumption on all the BOINCprojects, which obviously are not uniform.

Summing up, I would prefer the number provided by the Rosetta Team (however taking it with a grain of salt).

I hope I helped.

Best from Warsaw, BOINC@Poland. :)

P.S. Sidenote: FLOPS (FLoating Operations Per Second) is singular. FLOPSes is plural; thus 130 TFLOPS is O.K. (like 130 m or 130 kg) but a TFLOP is a big flop. ;P
ID: 69621 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile rochester new york
Avatar

Send message
Joined: 2 Jul 06
Posts: 2842
Credit: 2,020,043
RAC: 0
Message 69622 - Posted: 10 Feb 2011, 14:38:35 UTC - in response to Message 69621.  
Last modified: 10 Feb 2011, 14:43:29 UTC

The estimate on this site is apparently derived from the "Credits last 24h". Does this mean that 1 million R@H credits in one day is the equivalent to 1 TFLOP? Which means... that 1 TFLOP is 1 million credits divided by the seconds in one day...

Am I going in the right track?

Rosetta main page (Server Status) calculate FLOPS estimate from "Credits last 24h" (1 million credits per day = 10 TFLOPS @ 24/7)
Boinc stat (http://boincstats.com/stats/project_graph.php?pr=rosetta&view=credit) calculate FLOPS estimate from Recent average credit (RAC - sliding average from last ~30 days), but shows only half of it (50%) cuz 1 gobblestone (credit) composed from 50% of integer speed and 50% floating point speed.
So major(2х) difference is (TFLOPS only) vs (TFLOPS + TIOPS).
Minor difference is RAC (last 30 days) vs (last 24h) and affected by last big downtime of project in January.
Then RAC stabilise BOINC will show ~ 60-65 TFLOPS, and R@H ~ 120-130 TFLOPS.



they cant both be right..... boinc shows active computers
how long do you think before it stabilizes??


As Mad Max nicely explained, there are two major differences between methodologies on BoincStats and Rosetta sites.

Firstly, boincstats site uses RAC, which takes an average from the last thirty days. Thus, it takes 30 days since the come back after outage to "stabilize".

On the contrary, Rosetta site uses only last 24 hours.

Secondly, boincstats uses a different translation between cobblestones (BOINC points) and FLOPSes. It seems to assume that 1 cobblestone = 5 MegaFLOPSes.
On the other side, Rosetta estimates 1 cobblestone = 10 MegaFLOPSes.

Regarding this second difference - which approach is more reliable?
My guess is that Rosetta team knows better the nature of their calculations and can adjust the multiplier (cobblestones / FLOPSes).
On the other hand, boincstats makes one general assumption on all the BOINCprojects, which obviously are not uniform.

Summing up, I would prefer the number provided by the Rosetta Team (however taking it with a grain of salt).

I hope I helped.

Best from Warsaw, BOINC@Poland. :)

P.S. Sidenote: FLOPS (FLoating Operations Per Second) is singular. FLOPSes is plural; thus 130 TFLOPS is O.K. (like 130 m or 130 kg) but a TFLOP is a big flop. ;P


i like the boinc site because it shows current hosts and users.... rosetta shows total hosts supposedly since the project started it does not show current hosts and users
ID: 69622 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 69623 - Posted: 11 Feb 2011, 2:02:29 UTC - in response to Message 69621.  

P.S. Sidenote: FLOPS (FLoating Operations Per Second) is singular. FLOPSes is plural; thus 130 TFLOPS is O.K. (like 130 m or 130 kg) but a TFLOP is a big flop. ;P


It may vary between languages, but in English FLOPS is both singular and plural. When singular it means "Floating Operation Per Second" and when plural it means "Floating Operations Per Second". You are correct that FLOP is definitely wrong though as that would mean "Floating Operation Per", which doesn't make sense.

I have never seen FLOPSes used in English before.
ID: 69623 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Tony DeBari

Send message
Joined: 13 Apr 06
Posts: 12
Credit: 2,944,090
RAC: 0
Message 69625 - Posted: 11 Feb 2011, 20:16:04 UTC - in response to Message 69623.  

P.S. Sidenote: FLOPS (FLoating Operations Per Second) is singular. FLOPSes is plural; thus 130 TFLOPS is O.K. (like 130 m or 130 kg) but a TFLOP is a big flop. ;P


It may vary between languages, but in English FLOPS is both singular and plural. When singular it means "Floating Operation Per Second" and when plural it means "Floating Operations Per Second". You are correct that FLOP is definitely wrong though as that would mean "Floating Operation Per", which doesn't make sense.


Note quite...

A FLOP is a FLoating-point OPeration. The plural is FLOPs (lower-case "s"), and it is perfectly correct to talk about megaFLOPs, gigaFLOPs, teraFLOPs, etc. as scalar quantities: "This calculation will requite 42 megaFLOPs to complete." or "1,000 MFLOPs equals 1 GFLOP."

FLOPS (upper-case "S") - FLoating-point Operations Per Second - is a rate of speed (like Miles Per Hour) and is neither singular nor plural: "Rosetta@Home is currently running at over 130 teraFLOPS." or "My old 1 GHz Pentium III has a computation speed of just under 1 GFLOPS."



-- Tony D.

ID: 69625 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : Nice and steady 120+ TFlop estimate.



©2024 University of Washington
https://www.bakerlab.org