[OoklaTheMok]

I just tested the Ars Technica bench unit in 3.0 under OS 9 on my iMac 500, and it's about 2 hours faster than the 2.04 version! 2.04 took 9 hours 40 minutes, and the 3.0 took 7 hours 24 minutes! I'll be testing the OS X one next; the 2.4 version took about 9 hours 20 minutes for the bench unit.

[cwasko]

I noticing some weird actions by SETI 3.0 as far as time is concerned. There seems to be a dicrepancy between the time that SETI reports and the time that it actualy takes to process a unit.

For example, last night, I was running SETI in the background on my G3 laptop while I was working in Adobe GoLive. Well, after 4 consecutive hours of work, SETI 3.0 only showed 1:30 of CPU time! The older client would have said 4 hours...

it seems as though SETI is only counting the *actual* CPU time spent on the unit, rather than the elapsed time. At least, this is my theory. Since this was a new version, I wrote down my start times. The 'seti' time showed my laptop doing 9.5% an hour, however, my actual time was 6% hour. My average %/hour for version 2.0.4 was 6% an hour. So, in conclusion, with the limited data sample I've had, SETI 3.0 is no slower or faster. Just different.

[Danometer]

Sorry, I don't see an speed increase. It looks to me that SETI 3 is about 25% SLOWER than 2.04. Current work unit is at 8 hr 50 min CPU time and 43.202% done, which means that it should be done after 20 hr 26 min CPU time. My average is 16 hr 48 min with 2.04. This is on an iMac Rev B with 160 MB (no VM or RAMDoubler) and a 1MB RAM Disk using OS9.04.

We'll see if it speeds up before the end of the unit.

[…Later…]

The work unit took 20 hr 15 min to process.

There is one advantage to 3 over 2.04: I can run it hidden in the background and my mate doesn't complain about it. Hell, she didn't even know it was running this evening.

I'm open to ideas as to how to crunch faster, outside of a processor upgrade. I'm already doing the RAM disk thing, which was good for a 10% boost.

[This message has been edited by Danometer (edited 10-05-2000).]

[iPad]

It is definently faster, one thing i've noticed though is that every time that graphic thing goes through, it adds on 0.005 percent onto your time.

------------------
People aren't as dumb as you think. Wait, what was I thinking, 90% of the world uses WINDOWS!
You don't like Macs? Good, more for me.

[djohnson]

3.0 is only faster on machines that have less than 1Mb cache. This would include iMacs iBooks and older macs. Only 466G3 w/ 1Mb cache, it takes 4 hours longer!!! This is with all speed enhancements. My average time with 2.0.4 was 5:45 and now it is 9:45... I didn't think that the new version was supposed to take this much longer, but hey its all for science...

[Todd Madson]

I also am noticing this - it's pretty horrible. On my G4/400 where I was previously experiencing five to six hour block times, I'm not expereincing them in the ten hour range. In OS X, these times are stretched to around fifteen hours. And this is with accelerated level two cache! I boosted the cache to 267.5 mhz, am running LibMotoSh, the works and it's that slow. Now, what I could experiment with using the Cache Profiler 1.3 is shutting my backside cache off altogether and try it then. That way I'd only have 64k of L1 and that might boost things even though logically it should not. My wife's iMac DV SE might also have some interesting results, but I'm reluctant to move to 3.0 there just yet.

[Todd Madson]

Arrgh. Change my crap typing to "I am NOW experiencing ten hour block times."

It's enough to make someone go back to distributed.net.

I'm trying this on my wife's iMac, I'm also going to experiment shutting off the L2 cache on the G4.

[Misha]

Speeds on my iMac DV are about the same as 2.04... speeds on my Pismo 400 are about 4 hours slower.

Hmm...

[wlonh]

revert to 2.04? i suppose that's an option

too early for me to report any results using 3.0, with any exactitude... i won't extrapolate a finished time from a percentage of WU done as a WU is not the same from end to end (and i'm lazy)

but it seems that 3.0 is slower on my 466 G3, and perhaps just a bit faster on my 300 iBook

This is from the Seti@Home site:

"The combined effect of FFT optimization, pulse detection, and the extended doppler drift range is that a typical workunit will take about 40% longer to complete with version 3.0 on any given platform. On balance, this gives the project the best science for the CPU cycles used."

Version 3 will be a mandatory upgrade (there will be a 2 week grace period to finish your current work units).

[This message has been edited by Fuse (edited 10-05-2000).]

[Todd Madson]

After some experimentation, I've reverted to 2.04 on the G4 until I'm forced to upgrade.

I've decided to leave the iMac DV SE at 3.0 to see what, if any, performance enhancements I receive. It will need to be at that eventually anyway and since my wife will be away for a day or so on business that will be a good, uninterrupted test.

As far as my total block count goes, I might make 1000 blocks if I'm very lucky (I'm at 886 now) before the switchover is forced. Then I don't mind so much.

Frankly, I'd love it if there was a text only 3.0 current client for OS X. The graphic nonsense can be hidden, but it's still dog slow.

-t

[elzinat]

It seems like not many people have completely read the release notes with the new version. It is to be expected that some WU times are half what they were, while others are much higher. The new version scans the data for interesting Gaussians, and if the data is interesting only then does it perform the large numbers of FFTs. So a boring WU will be completed more quickly, and an interesting one will take longer. In other words, the WU times with the new version 3.0 will be much more inconsistent, and depend far more on the interestingness of the particular WU than on the speed of the computer.


These are my SUM stats for my first 3.0 WU.
_____________________________________________
Date: Thursday, October 5, 2000 10:10:20 AM

<SUM 260>: UNiT "10jn00aa.18214.26337.861064.226" finished. total time: 0:11:44
crunch time: 0:11:42
CPU time: 0:11:32
CPU usage total: 98.2%
CPU usage while crunching: 98.5%
Gaussian Score: 0.283784
Peak Score: 0.648367
_____________________________________________
My G4 350's average over it's first 99 WUs was 7:33. This clearly was a rather interesting WU that I happened to get. So 3.0 took 4 more hours than my average with 2. On the other hand, many WUs will see much lower times.

In order to evalute the actual overall change in processing time with the new version 3.0 of the SETI software, we will need to analyze a Large number of WU's, and look at the new averages balancing long interesting and brief uninteresting datasets.

[crazycylon]

I was getting pathetic performance with Seti 3.0. It was processing 2 minutes every 2 hours my system was on. I have since increased my disk cache from the default 4096 K to 5120 K. I have done 5.7% after 30 minutes on a G3 400Mhz chipset every since I have made this change.

[This message has been edited by crazycylon (edited 10-09-2000).]

[wlonh]

ah ha! capital idea, i am going to boost my disk cache right now

thanks

oops, i just tried to... can't! my default is 8160 and it will go no higher... hmm... drat.

[elzinat]

I've now done 5 WU's since installing 3.0.
I'm running a G4/350.
Average with 2.04: 7.5 hours/WU over 99 WUs
Average with 3.0: 10.75 hours/WU over 5 WUs

I think i'm going to have to downgrade...

[druber]

Also, I'm pretty sure 2.0.4 ran in the background just fine, though slower, natch. 3.00 doesn't seem to run in the background at all. I mean not a bit. Is my memory faulty, or is there something I can do to share clock cycles here?

oh, I'm also using a ramdisk, which I hadn't used with 2.0.4. Could that make the difference?

[This message has been edited by druber (edited 10-10-2000).]

[crazycylon]

My unit completion is now slightly faster on 3.0 than it was on 2.04. It took me approximately 11.5 to 12 hours per unit. I am now doing a unit in about 10.5 to 11.25 hours per unit. I am only using a 400Mhz iMac DV SE. It appears that 3.0 skips all the boring parts of the data.

[Todd Madson]

I think was the Seti @ home people did is in doing the extra science, they
also made it more "friendly" to people with cheaper, slower computers. In doing so, they made it more difficult for individuals with more expensive, faster machines. Since the larger more expensive boxes typically have larger L2 cache arrays, it appears that the new client favors smaller cache configurations and will actually run faster on the machines with smaller L2 cache. It's all dependent on FSB, L2 cache, and speed. I imagine that if, say, there was a 1 ghz G4 in a lab somewhere it would do wonders with the v3.0 client. But right now, it's just sort of annoying that it takes longer on the more capable machines due to the way it was programmed.

Even more annoying, the OS X client is even slower than the 3.0 OS 9 client and after processing for 18 hours crashed just as it was going to
send a block. Ouch! When I relaunched it, it has lost that data too.

Well 3 is definately slower.

AVG time on G4/400 with 2.04: 6hrs

AVG time wih 3: 10hrs

[OoklaTheMok]

Yeah, on machines with 1 meg cache it will be a lot slower, but on ones with 512k cache it will be faster or not much slower. And I'm sure it runs much better on the new iBooks than 2.04 does.

[gsxrboy]

iMac500DVSE 256mb ram
2.04 average 10 hrs
3.00 average 10 hrs

[proux]

Similar times on an iMac SE 500Mhz, 10h (from 9h with 2.04, cache at 250Mhz).

iBook SE 366Mhz, 13.5h (from 12.5, cache at 183Mhz).

There doesn't seem to be any speed improvements left from there.

Anyone knows how long it takes on equivalent Wintel machines?

[zac4mac]

3.0 was so slow on my new G4(7-8 hrs/wu) I dropped back to 2.04. Now I'm seeing times of 4:45 - 5:00 per. I understand the why, and will happily switch to 3.0, but only when forced to. That being said, does anyone know for sure how much longer before we HAVE to switch to 3.0?

[Joe Cool]

I even tried using a program called G4 Strip to set my cache on my G4 450 to 512k. It did nothing. Where can I get 2.0.4. now?

[Todd Madson]

Setting your cache faster usually shaves 10-15 minutes off of your block time. No other mod is as dramatic as that. I'm afraid G4s will suck under the new client simply because our L2 cache modules are large. The
client has more or less been optimized to be friendlier to slower CPUs it
appears since they typically have smaller L2 caches. It appears that if
you have a 1 meg or larger L2 cache, you will go slower than the rest.

[monte]

weird... It seems to me that Seti has 'leveled' the playing field across the board.. While the one's with the most go slower and the 'lesser' machines are quicker, the times appear to me, to be more or less the same. While this is just an observation, I don't mean to accuse or represent a problem, it's just really weird. crunch crunch crunch...

[gsxrboy]

iMac500DVSE, now using Libmoto and Plogix cache 1.3 (at 250mhz)
Looks like I am heading for a 8hrs even with current block, gotta love that v3 speed+ tricks... : )

[proux]

The machines with 1MB cache will not be slower than the ones with 512KB,
but the lower cache machines will not be at a disadvantage anymore.

Still G4 machines should be a bit faster because the motherboard is more
advanced and the G4 chip has higher FPU at same clock speed.

Ciao

[gsxrboy]

Excuse me for wohooo-ing, but woohoo!!! forget 8hrs, it made it in 7.5hrs (iMac500DVSE), not anything special to g4 owners but it does make me feel much better than the 10 hrs from b4...

Apologies to those who couldn't care less.....

[proux]

Not all units are created equal!

My revA iMac just finished one in 10h flat...

[Fred CHOTTIN]

On an iMac rev.B with 64MB RAM and MacOS 9.04 average time was something around 23 hours on SETI 2.03. Since SETI 3, all my data blocks needed more than 30 hours work.They musty have been interesting ones as in SETI FAQ, they say the new version make deep analysis (thus longer) for interesting data and skip uninteresting ones. So far, it did not happen to me yet.

[proux]

You should get much faster times, my Rev A iMac with 3.0 9.0.4
has an average of 17.5h (from 14 with 2.x).

Do you use the screensaver feature?