Announcement

**Guru** · 3 December 2000, 17:00

Originally posted by Rattledagger:
If we say 2 hours goes to graphics in 3.02

Well the "graphics" actualy takes more than 50%!

------------------
Join the MURC SETI team! | SETI @ MURC

Bad dog. BAD DOG! I SAID BAD DOG!!! Go HUMP SOMEONE ELSES LEG GODDAMNIT!!!

**-=Narcissus=-** · 3 December 2000, 17:03

Yuck!

Will we be able to keep our v3.0 clients???

I lost some good crunching machines in the past because people were saying that v3 was too slow...

I thought that WU times were supposed to get better with time, not worse!

Guess the old saying is still true :
"What Intel giveth, Microsoft taketh away"...

which should now be modernized a bit :
"What AMD gives, SETI takes away!"

Fred

**Paulr** · 3 December 2000, 19:02

I just want to keep my 2.70 client I've been running for ages.
I'm knocking these work units out nice and fast but I can see this changing once 3.02 is made a 'must have'

**Rattledagger** · 3 December 2000, 20:18

Guru, if you take a look on the Beta 3.01-result, this is 1 hour slower than the cmd-version. 1 hour of 6,5 hour isn't 50%.
Of course, it's just a guess for ver. 3.02, but if it's like 3.01, the graphics hit should be under 2 hours on this machine.
On another machine, the 3.01-hit is 1,5 hours, but the normal speed here is 11,7h.
All results for comparable mid-range-angle-ranges.

Narcissus, the bug in the old 1.x-clients is that they bombards the server with requests if the normal discontinuity of old clients is used. This wasn't found out before ver. 3 was out. That will happen is that the old server will be turned off, and for all 3.00- and previous clients seti@home is litterally removed from existence.

[This message has been edited by Rattledagger (edited 04 December 2000).]

**Butcher** · 3 December 2000, 20:23

personally I think not wanting seti to extend their calcs just so you get better crunch times a bit selfish, are you just in this for an ego boost or what?

**-=Narcissus=-** · 3 December 2000, 20:51

Butcher,

Please don't get offended by my (?) post. This is NOT an ego boost by my part. But on the other hand, I don't really believe in extra-terrestrial life (or al least in the form that we're looking for them through SETI). (Please don't start a debate with this )

I just love the idea of making a supercomputer with millions of PCs. It's THAT part of SETI@Home that first got me into this.

On the other hand, I don't really understand how SETI@Home has been optimised through it's many evolutions. For instance, it seems that the xeons with a lot of cache have such a humungus advantage just because of their cache. Why couldn't the program be optimized for the home PC with some smaller cache sizes? I'm not asking for Celeron small, but let's say Pentium small.

My personal computer (C466/128MB ram) isn't an hour slower with V3, it's 3-4 hours slower. No, I haven't optimized my PC for use with SETI, but then again I don't think that the majority of the PCs running S@H are...

Hey I'm no computer engineer or programmer. And of course if the guys at S@H believe that this new extended chirp range is useful, by all means, go on through with it. But don't control 10% of the machines crunching for Team Narc and I know that if our WUs diminish, production will slow down and people will loose interest (like it has happened in the past). My teammates will loose interest in this friendly competition, but probably some people will also in S@H in general.

Ok,ok, I'll stop nagging now...

Just got to get every little WU out before the new version becomes manditory.

Keep on crunching,
Fred

P.S. Sorry for the numerous spelling mistakes, my english is a little shaky...

**agallag** · 3 December 2000, 22:51

I have no problem with the processing time getting longer in order to do more science, as long as they force the upgrade for everyone. If everyone's WUs take longer, there's really no difference is there?

------------------
Andrew
Carpe Cerevisi

**SteveC** · 4 December 2000, 09:59

^^ Wot 'e sed ^^

------------------
Cheers,
Steve

"Life is what we make of it, yet most of us just fake"

**Butcher** · 4 December 2000, 10:17

Fred,
My post wasn't aimed at you in particular it was just general comment. As for optimizing for PCs not servers, v3 is. In v2 my xeons would have given me a massive advantage over anyone on a normal pentium (probably 2x speed), however with v3 the advantage is much smaller, more like 30% or so.

**Guru** · 4 December 2000, 10:23

More like ~15% sorry m8!

------------------
Join the MURC SETI team! | SETI @ MURC

Bad dog. BAD DOG! I SAID BAD DOG!!! Go HUMP SOMEONE ELSES LEG GODDAMNIT!!!

**klambert** · 4 December 2000, 14:59

Umm - If you want to know how much running the graphics affects the times, why not run the same unit on 3.02 with and without the graphics? That way we'd know if 3.02 run without graphics is actually slower than the 3.0 CLI.

**Rattledagger** · 4 December 2000, 19:25

Well, I haven't tested the same wu with and without graphics, but I've processed a couple of units on two different machines. I've used beta 3.02 on the fast machine, and beta 3.01 on the slow:
Unit 1, with graphics:
fast: 12h28m23s; slow: 13h13m32s
Unit 2, without graphics:
fast: 12h59m34s; slow: 12h47m29s
cmd 3.00: 6h48m3s

Interesting development.

The average in the fast machine for this kind of wu is actually 6,749h/wu, so the hit with beta 3.01 is lower than 15%. For the slow machine, 11,604h/wu. For both, this is under 15% hit compared to cmd 3.00.

If we use 15% hit for 3.02, the expected new results should be that cmd 3.x uses 67% longer time, or a 600MHz-machine using cmd 3.x will perform as a 350MHz using cmd 3.00.

The next version will at any rate be slower than the current 3.00-versions. The reason for extending the chirp range is this:
1: Because of so much more users than expected, "we could do even more extensive processing on each work unit than we had thought practical"
2: After adding the new pulse/triplet searching, "work units were being completed and returned to us more rapidly than before because we had also implemented new, faster methods for the earlier types of calculations."
3: And probably the biggest reason:
"As a result, the internet traffic began exceeding the limits of what the UC Berkeley infrastructure is prepared to handle."

Funny, why aren't any of the people that wants to use old (and broken) beta-versions using beta 2.66? Since the point with seti@home is the science, beta 2.66 is just as good as beta 2.70.

**Assimilator1** · 6 December 2000, 12:43

Narcissus

>>>Why couldn't the program be optimized for the home PC with some smaller cache sizes<<<<

They have with v3 (& v2.7x?) ,I've found that using v3 CLi with a Celeron PC is faster (on average) than the old CLi v2.4.
This didn't work for my old PII 233 @ 280 I had though ,it was slower for that.I guess that v3 prefers fast cache to big cache relative to v2.04/2.4

I have no problems with WU's being slower if it improves 'The search'.
If some people don't like the idea of slower WU's tell them its quality over quantity with v3.
Oh yeah ,of course if some MURCers drop out ,that just makes it easy for us to stay ahead

------------------
I guess I won't recruit for my team here

Team Anandtech

**Rattledagger** · 6 December 2000, 16:29

Assimilator, your team is currently 70.000 ahead. Then the new 3.x-clients is released, this should in crunch-time be the same as you're leading with 117.000 units.

As for users dropping out, this shouldn't be much different for any team.

Announcement

The new 3.x-clients will be slower.

The new 3.x-clients will be slower.

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment