Advanced search

Message boards : Graphics cards (GPUs) : GPUGRID vs SETI@home beta

Author Message
Profile Krunchin-Keith [USA]
Avatar
Send message
Joined: 17 May 07
Posts: 512
Credit: 111,288,061
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 4377 - Posted: 16 Dec 2008 | 1:30:38 UTC

I started a new thread for this.

What applies for GPUGRID should apply for Seti.
So, 280 is the current fastest card.

gdf

Maybe, but I do not think their application is as intensive as GPUGRID. I tried to follow along in their forum but there is so much to read. They do not really need the fastest.

I think they have a shorter run time 2-3 hours for GPU's and probably can make use of slower cards not able to participate here.

It seems their deadline is about three weeks, 1/8/2009 for one downloaded a few hours ago, my first. But I still have to wait 8 hours for my current GPUGRID task to finish. Can't tell run length since it is the first it has an estimate of 00:00:00. I'm going to test a very few, but ultimately I will stay with GPUGRID and only use SETI@home as a backup should this project run dry of work, which seems to happen once and a while. Same for other projects as they come online.

I'll let you know how one compares after it completes.

I see they have windows only so far, and it appears their windows app only uses 4% or less of the CPU on windows. Currently the GPUGRID I'm running is showing 37 to 44.


OK here are some comparisons. On my one system, that is and under WindowsXP

All times approx. These are run with 6.4.5

GPUGRID runs 59000s/16.5h elapsed time with 48000s/13.33h cpu time, which is about 40% cpu usage, when observed with windows task manager.

SETI@home beta runs 26 minutes wall clock time and only uses 7min cpu time. It was hard to rate the cpu usage. It runs near full 44-50% for the first 30 seconds, then it drops to 3-12%, average 10%. Later in the task though it runs more like 12-22%. I saw spikes, both ways with the cpu anywhere from 2-23%. Mostly though you could say the average was 10%. A lot less than GPUGRID.

I do not know how credits compare yet.
____________
Alpha Tester ~~ BOINCin since 10-Apr-2004 (2.28) ~~~ Join team USA

Profile Punky260
Send message
Joined: 23 Jan 08
Posts: 3
Credit: 86,585
RAC: 0
Level

Scientific publications
wat
Message 4380 - Posted: 16 Dec 2008 | 9:34:20 UTC

On my System is very more CPU usage @GPU-Grid
The usage @seti is similar.

But I think, it's not necessary except for people with more than one GPU.

GPUGRID needs 0.90 CPUs, and SETI 0.03 ... so if you have to cards, GPUGRID takes 1.80 CPUs and SETI just 0.06. That means, that GPUGRID takes 2 CPU cores, same time as seti takes 1 core.


The Credits at SETI are still in Pendings, and won't get granted so fast.
Don't forget that SETI does this GPU-Computing till this WE and not as long als GPUGRID.

So I would prefer GPUGRID so far for people who wants to help or get Credits.
I decided to myself that I crunsh the two projects at 50:50. So I get enough Credits from GPUGRID and can help the SETI-People with theyre BETA-thing.

Profile Krunchin-Keith [USA]
Avatar
Send message
Joined: 17 May 07
Posts: 512
Credit: 111,288,061
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 4383 - Posted: 16 Dec 2008 | 12:57:39 UTC

I got a couple returned results.

interesting parts:
Flopcounter: 28798913900607.105000

They do not record GPUtime, so it is hard to tell exactly.

Best guess would be 70-73 credits claimed (none granted yet) for previous approximate 26 minutes work. This estimates out to about 2665 credits for 16.5 hours, whereas GPUGRID here will grant 3232 for same time (based on my system). Of course credit is granted per flops computed, but you get more from here per time used.

@GDF
How many Flops are in your tasks ?
I know there are 850,000 steps, but how many flops in each step.

This way we could better compare the applications performance.

STE\/E
Send message
Joined: 18 Sep 08
Posts: 368
Credit: 4,173,502,885
RAC: 27,150,888
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 4385 - Posted: 16 Dec 2008 | 13:14:23 UTC

I ran some over there too & had it figured about 420 Per Hour where I get about 600 Per Hour here on the same Box with a Overclocked GTX 260 (216 Version) ...

Thamir Ghaslan
Send message
Joined: 26 Aug 08
Posts: 55
Credit: 1,475,857
RAC: 0
Level
Ala
Scientific publications
watwatwat
Message 4394 - Posted: 16 Dec 2008 | 15:17:57 UTC - in response to Message 4383.

I got a couple returned results.

interesting parts:
Flopcounter: 28798913900607.105000

They do not record GPUtime, so it is hard to tell exactly.

Best guess would be 70-73 credits claimed (none granted yet) for previous approximate 26 minutes work. This estimates out to about 2665 credits for 16.5 hours, whereas GPUGRID here will grant 3232 for same time (based on my system). Of course credit is granted per flops computed, but you get more from here per time used.

@GDF
How many Flops are in your tasks ?
I know there are 850,000 steps, but how many flops in each step.

This way we could better compare the applications performance.


I'm running both, but more slanted towards gpugrid because its more stable, offers better credits, the science is better, and the user base are tech savvy.

Based on the current deadlines I guess I'll be crunching more gpugrids than seti: 1 month seti vs 4 days gpugrid.

Profile Edboard
Avatar
Send message
Joined: 24 Sep 08
Posts: 72
Credit: 12,410,275
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 4396 - Posted: 16 Dec 2008 | 16:27:34 UTC

A thing that I like from SETI is that you can have a lot of units in cache (which seems to contrdict the statement that it is BOINC, not GPUGRID that limit WU to 1 by each CPU)

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 4435 - Posted: 17 Dec 2008 | 19:36:43 UTC - in response to Message 4396.

(which seems to contrdict the statement that it is BOINC, not GPUGRID that limit WU to 1 by each CPU)


GPU-Grid limits the amount of concurrent WUs to 1 per CPU because BOINC oes not yet allow one to tie the amount of WUs to the number of GPUs. GPU-Grid needs this limitation because they need results back fast, whereas for SETI only the long term throughput is important.

MrS
____________
Scanning for our furry friends since Jan 2002

Post to thread

Message boards : Graphics cards (GPUs) : GPUGRID vs SETI@home beta

//