Running SETI on a Xeon

[Spacehead]

It turns out that the computer i got SETI running on at work is a 2ghz Xeon, not a P4 as i was first told.
Just wondering though, how many instances of SETI would you run at a time on that?
I've heard that you can run 2 instances on a Xeon without much of a performance hit, SETI wise, as it has the hyperthreading ability, i guess???
Currently running 1 instance & over 170 WU's the average time is about 3:20 using SETI Driver.

 

[Baldy18]

I would run 1 instance per cpu. I suppose that if hyperthreading is enabled then it would probably be appropriate to run 2 instances.

 

[Migroo]

Nice WU time!

 

[Spacehead]

OK, i'll leave it as is. Just thought i'd ask.
Luckily for me/us, the operator of that machine likes to leave it on all the time & no sleep mode overnight.

 

[Swanny]

From what I understand, hyperthreading offers no performance increase if the 2 operations it is trying to perform are the same. Basically it only works well if the 2 apps are needing to perform 2 different operations on the CPU. At least that's what I understand from Anand's article.

 

[zeruty]

maybe some seti+eccp is in order



edit: woot! 900th post!

 

[PieDerro]

I remember a couple of months back someone around here (a pretty old-time member... I just can't remember who it was... Soni? Engineer? hmm ) had access to a dual P4 Xeon 2.2GHz with Hyperthreading enabled.

Running SETI with Hyperthreading enabled scored that member around 18-20WUs/day from what one machine!!! :Q:Q

My advice is, run SETI and hyperthread. But make sure you run 2 instances per CPU. SETI Driver mucks up the times, so don't be surprised if it says 5:30 mins estimated time for one WU. It's just that each CPU would do 2 WU's in that amount of time!!!

Good luck
-PieDerro

 

[Poof]

It was Barbary who did some excellent work benching his Xeon (Prestonia) last January. Here were his findings:

Subscriber linky
Non-subscriber linky

His work with the Ars TLC bench unit prompted a good discussion with TLC benchmark data monitor, good old Roelof Engelbrecht (of SetiSpy fame) to try to determine CpF and how one might report the results in order to compare apples with apples. Basically, the Prestonia run with hyper-threading on, would give you more "production" out of a machine over a unit time, although actual SETI-reported WU time would not change from that obtained with hyper-threading off.

So if I'm interpreting correctly - and as an example...say it normally did reported 3 hour WUs and that 1 CPU could produce 8 per day w/o hyper-threading. Turning that feature on resulted in what would calculate as 2.5 hour WUs because rather than 8 WUs uploaded per day, you'd get something like 10 per day, although the WU times would still be reported as 3 hours.

Capeche?