SoB: Did we start on a new K value?

[RemyCanad]

It looks like all my machines are producing less now. I am figuring we started on a new K value that is slower to process than the one we are on....

 

[deerslayer]

My rigs are still about where they were before

 

[Stormgiant]

yes, i think so...

my p4 was in the area of 310-320.000, now is burning at 250-270.000....

 

[RemyCanad]

Originally posted by: Stormgiant
yes, i think so...

my p4 was in the area of 310-320.000, now is burning at 250-270.000....

Ok good I thought it was just me.

 

[Slatzman]

what is the k and n that you are crunching?

As the n's go up, the cems/sec go up.

I have noticed that 3 of my last 5 k/n pairs have been recycled ones from much lower in the n range than the current maxes...that slows the cems/sec down noticably

Slatz

 

[Stormgiant]

the P4 2000 is crushing a 2620303 at 265.000 and this 3000 is doing a 2613286 at 250.000 :Q

 

[RemyCanad]

Yep it looks like we are getting the recycles.

 

[SinfulWeeper]

Current k=5359
Current n=2441430
cEMs/sec=159163+/-

P4 1.6@2.13

 

[Slatzman]

I am noticing a slowdown also with the k/n pair I got today...was running in the mid 90's kcem/sec now down in the low 80's

Slatz

 

[J.Zorg]

My production also dropped from ~310K to ~270K.
First thought on of my nodes was down, then i was worrying something was going wrong with my nodes. But the i found this thread.
Now i´m reliefed that no cpu cycles are wasted.

 

[Confused]

All mine are running as services...so i can't see the speed lol

Hopefully we'll be moving across to using FLOPS for the speed, which will eliminate the need to worry about k and n values for the speed That should also help people when judging the best overclock


Confused

 

[RemyCanad]

Originally posted by: Confused
All mine are running as services...so i can't see the speed lol

Hopefully we'll be moving across to using FLOPS for the speed, which will eliminate the need to worry about k and n values for the speed That should also help people when judging the best overclock


Confused

Yeah I really need for them to get that figured out. I can't belive I was trying to overclock my CPU and at the same time got a really slow combo of numbers. So when I started it up at a higher speed the thing ran really slow.

 

[MADn3ss]

At a similar n-value, switching k values has little impact on the total computation time. The n-value is definitely the most important factor in test length.

That said, the guess that lower n values that were issues a while back are either being re-tested or re-issued is a good one.

BTW, SoB is testing all of the remaining k values, for those who didn't know.

Keep crunching,

MAD-ness
Team Prime Rib

 

[RemyCanad]

Thanks for the info MADn3ss.

Also do you know what n value range we are currenty testing? (Not the one we a retesting but the others.)

I have a 2644340 n right now with a 21181 k. And my athlon XP 2000+ at 1.75GHz is only doing 177081 cEMs/sec. At one point this machine was doing 210000 cEMs/sec at only 1.7GHz. Is my n value that low?

Also nice to see you around the forums.

 

[MADn3ss]

I am an idiot.

Even after writing up this FAQ entry I didn't realize that the performance issues we started experiencing in the last day or two were most likely due to the FFT size increasing.

A while back, Louie speculated that the cross-over point for the FFT size was ~ n =2.6 million

Because SoB uses the FFT routines from Prime95 that George Woltman wrote, it (obviously) has many of the same performance characteristics. FFT cross-overs are not all that frequent (in GIMPS they are years apart at times, but I don't know what the interval is at the smaller numbers we are testing).

So, I figured I would drop by and let y'all know, because this will likely be a permanent thing once the remaining tests under the cross-over (~n=2.6million) are completed.

MAD-ness

Team Prime Rib

 

[RemyCanad]

Ok that makes more sense now. Once again thanks for the info.

 

[Slatzman]

here is the link to the message I posted on the SoB forum asking about the slowdown.

linky

like madness said...it is a fft crossover thing

Slatz

 

[deerslayer]

My main rig has gone from 210k to 165k :Q

 

[deerslayer]

EDIT: Nevermind, if you client is on all night with nothing to do, your rate is calculated over the entire period of time the client is running. Restarted the client and it's at about 200 again