S@H v3.0 client speeds?

[Hellburner]

Maybe they were tinkering with the % completion stuff or something but right now some of my work units are showing twice as fast as the 2.76 beta ????

Right now I'm showing about 2:00 hrs on a P3 650@866, the best time was 3:40 hrs under 2.76.

Update:

Those numbers are showing a slowdown but still looking much faster, I'll keep you posted

More:

The numbers have settled down to about the same as the 2.76 beta, maybe I just never paid enough attention to the progress. Those numbers were freaking me out

 

[TuffGuy]

i'm benchmarking it (v3.0) with the ars benchmark unit. it should be done in a couple of hours. i'll post the time then. the time with v2.70 was 3h4min. we'll see how v3.0 compares.

 

[TuffGuy]

update:

v2.70: 3h4min
v3.00: 3h22min

 

[Assimilator1]

Anyone else got more times?

 

[cory]

i am currently trying it on a few pent III and celery's under
linux.

so far it looks much faster than 2.04

 

[nexus9]

Anyone have solaris benchmarks?

-Nexus9

 

[Ken g6]

On my PII-400, the first WU took about 10.5 hrs, and the next looks like it will take as long. V2.0 CLi gives me 8-hour WUs on average.

Edit: OS is Win98. Maybe Linux has regained its supremacy?

 

[ReDSkuLL]

On my Athlon 700-Classic I am getting around 7-8 hours with 3.0, while with 2.04 I got about 8:45+. It did however knock off a hour for my celeron 466. I'll try it on my p233MMX once it finishes the packet its on.

 

[cory]

one thing about the new 3.0 is that the % comp
does not seem to go at a steady rate.
it jumps around alot. but
on one pent III machine it went from about 5 hours
to 3.6 hours.

it does seems to be faster but the programs that i had written to
look at the processes running on my cluster have very inacurate
compl times until it gets above 90%

it seems that for the first 20% or so it is fast then the next 40%
is slow and than at %60 it starts speeding up again.

 

[networkman]

My own findings concur with what Cory has said - that the percentage completed does jump around quite a bit. Also, I do not yet have an average time of completion for my machines; my Athlon 750 has so far been my best performer with the 3.0 client, but has jumped around considerably over the last 10 work units, with reported times anywhere between 3.9 hours(awesome!) up to as high as 8.4 hours(bummer!).

 

[TuffGuy]

i think that the time jumps due to triplet calculations. those are very cpu intensive. if setispy does an "instantaneous" estimate while the client is analyzing a triplet the estimate will be high. when the client finishes the triplet times go down. they might also go back up if another triplet is encountered.

 

[cory]

well i have a monitor program that checks each of my nodes
every 2 minutes and updates a log file with the current % complete
, the amount of time spent on that w/u, the amount of time
it will take to finish.

the log file is over 25 meg right now.

that way i can go back and look at progress and i do a tail -f
that can tell me if any of the nodes are not running.

there does seem to be a much greater variation in times
for the new version 3.0 but the best times are much better
than the best times of the old 2.04.

 

[TuffGuy]

25megs of text... that is quite a bit of data to sift through. let us know what you find out.

 

[Hellburner]

I was just reviewing my results log, checking for best completion times for my best machines.

P3 650@866 3:40
C 366@605 5:15

That is an exact clock speed ratio between the two, from this I would guess that memory bandwidth and L2 cache dependencies may be completely gone.

 

[JWMiddleton]

This is a copy of what I put into another thead...fits better in this one.

-------

Times for WU's on a PIII-700/933 w/CLI 3.0 are as follows:

3.351 Hrs 99.31%
4.274 Hrs 99.49%
4.374 Hrs 99.23% Played game of Hero of Might & Magic III while this one was working
4.446 Hrs 99.99% Did a bit of web surfing

My last 4 WU's with CLI2.4 averaged 4.765 hours with very little variation in completion times.

 

[PhotoLenny]

My first two 3.0cli wu's clocked in 20% faster than my average, where the 3.0gui was clocking in 20% slower - I seem, in the early going, to be finishing wu's in aout 7.5 hrs, vs 9.5 for the 3.0gui. The command line texy only version seems to be really fast, and does not suffer from the memory bottleneck on dually's that double cli's used to.

 

[TuffGuy]

Current average over at SetiQ using a tbird-700@927. times will improve since i bumped it up to 1133