SETI on a G5
I used to try to run SETI @ Home on my old 550 MHZ TiBook, but since it processed data so slowly I never really let it run for that long.
So now that I have a G5, I thought I'd see how fast it could analyse a SETI work unit (WU). All I can say is...WOW! I have no clue how well this compares to any other machine, Mac or PC, out there, but it definitely blows away my old TiBook!
On a stock G5, with processor set to highest, I was able to get through 1 WU in:
5h 45 min 19.4 sec
Moreover, large chunks of this were run as I was doing other work: I could see SETI going much much faster when it was the only thing running, taking up close to 100% of my processor. Interestingly, the G5's fans never got all that lound, which was nice to see!
So, if other people could post some SETI times I'd be curious to see them
So now that I have a G5, I thought I'd see how fast it could analyse a SETI work unit (WU). All I can say is...WOW! I have no clue how well this compares to any other machine, Mac or PC, out there, but it definitely blows away my old TiBook!
On a stock G5, with processor set to highest, I was able to get through 1 WU in:
5h 45 min 19.4 sec
Moreover, large chunks of this were run as I was doing other work: I could see SETI going much much faster when it was the only thing running, taking up close to 100% of my processor. Interestingly, the G5's fans never got all that lound, which was nice to see!
So, if other people could post some SETI times I'd be curious to see them
Comments
also check out some of the other totals for users and groups
cpus used in last 5 years.... 486dx100, AMD K6, PowerPC G4/G3
only run it occasionally these days, but almost finished my 158th WU
almost 7500 hours (most on antique machines= skewed avg of 47 hrs per WU)
The T.H.G. test used the comand line version and isn't that a little faster? They do have a comand line verson for OS X. You may try that to see if you get a little extra speed.
Originally posted by Ebby
5 Hours + That stinks. My Bro's 1.8Ghz G5 blows through them in little over an 2 hours. 512MB ram too.
The T.H.G. test used the comand line version and isn't that a little faster? They do have a comand line verson for OS X. You may try that to see if you get a little extra speed.
Yeah, I think I'll try the command line version at some point. I suspect that I should be able to get the same speeds as your brother.
Remember, in getting that 5 hours I was playing with Safari, Excel, Word and iTunes. Essentially, I was curious to see how well SETI would run in the background. Mind you, when I realised the answer was "not all that well" I starting letting it hog the CPU
GUI version.. ~5hrs.
never figured out how to use command line version.
Of course, SETI hasn't been been optimized for the G5 yet. I would hope to see maybe a 30% improvement from just recompiling with the latest version of gcc.
1. Using the graphical client on my 800 MHz iBook normally took 20 (!) hours per work unit.
2. Switching to the command line client cut that time IN HALF! Yes, just 10 hours per work unit, on an 800 MHz G3.
3. I had an AMD Athlon XP 1700+ for a few weeks, and that computer only took about four hours per work unit. This is a 1.46 GHz processor, using the command line client.
I'm anxious to see how my 1 GHz eMac does when using the CLI client. You can even keep it running as long as you're not doing anything demanding - SETI automatically sets itself to the lowest priority so it rarely slows you down unless you have a slow computer or you're doing something demanding like a game.
Aren't those things just check boxes/flags in the compiler?
Originally posted by bunge
I used SETI for a long time, but got annoyed at the fact that they infrequently updated and didn't compile the Mac versions with much thought. I don't know if things have changed, but no Velocity Engine or dual processor support.
Aren't those things just check boxes/flags in the compiler?
No, they aren't. There are flags to the compiler to allow the generation of VE code but the compiler doesn't do it for you (at least not yet -- auto-vectorizing compilers will arrive eventually but they aren't as effective as hand-coded vector code).
I thought SETI was multi-threaded, but I haven't checked myself. You could probably just run two instances from the command line.
Originally posted by Programmer
I thought SETI was multi-threaded, but I haven't checked myself. You could probably just run two instances from the command line.
Seti@Home is multithreaded! When I run it on my Dual 1GHz both processors run at almost 100% and process-monitor reports 155% CPU load. No need to run two clients. Avg. Time: 5.5-7 hours depending on the work-unit (not all units are equally processor intensive).
I also found out, that after every update the first unit takes appr. twice as long as the following ones - no idea why this happens.
Originally posted by Gulliver
Seti@Home is multithreaded! When I run it on my Dual 1GHz both processors run at almost 100% and process-monitor reports 155% CPU load. No need to run two clients. Avg. Time: 5.5-7 hours depending on the work-unit (not all units are equally processor intensive).
I also found out, that after every update the first unit takes appr. twice as long as the following ones - no idea why this happens.
From the SETI FAQ:
Question: SETI@home uses 90% of the time on two CPUs. Why does it do this if it's not multithreaded?
Answer: One CPU is doing graphics and the other is doing data analysis (so actually it is multithreaded, in a limited way).
Question: Are there versions of SETI@home for parallel systems such as Beowulf or multiprocessors?
Answer: No. You can parallelize SETI@home by running multiple instances of it, either on a multiprocessor or on the nodes of a cluster. Just make sure that each instance runs in a different directory.
Originally posted by Tidris
From the SETI FAQ:
Question: SETI@home uses 90% of the time on two CPUs. Why does it do this if it's not multithreaded?
Answer: One CPU is doing graphics and the other is doing data analysis (so actually it is multithreaded, in a limited way).
Question: Are there versions of SETI@home for parallel systems such as Beowulf or multiprocessors?
Answer: No. You can parallelize SETI@home by running multiple instances of it, either on a multiprocessor or on the nodes of a cluster. Just make sure that each instance runs in a different directory.
Thanx for the clarification!
I run 2 instances of the CLI client on my dual gig and manage them with Seti Dockling, which works great, and provides feedback. You can even re-nice the clients with it, and you never have to touch the terminal once you've got it set up.
It looks like with 2 clients running concurrently, a dual 2.0 shoud turn out 19-20 W/U's a day. can you imagine what it would do if SETI were Altivec-enhanced??
CV
SetiDocking is great.
Originally posted by chris v
I was looking at the SETI message boards just a few minutes ago, and two or three people there were reporting 2.5 hours per W/U with a dual 2.0. This was using the CLI version of the client, which is not multi-threaded.
That is faster than on my dual G5. I wonder if I am getting a batch of particularly tough data units or the other guy is getting a batch of particularly easy ones.
Originally posted by Tidris
That is faster than on my dual G5. I wonder if I am getting a batch of particularly tough data units or the other guy is getting a batch of particularly easy ones.
They do vary substantially in size. I've had some on my dual gig take 4.5 and some take 7.5.
CV
So I'd expect the G5 2x2GHz to do the same in about 4hours.
So, assuming there will ever be a G5 optimized version, this should be possible to reduce to 3hrs.
G-News