Nasty new SMP WU

Ive had more than my fair share of bad wus recently, my output is on its arse, but my 3macs, collectively 14cores are all folding SMP wus. Its not fair.
 
Yeah I got one of these. They absolutley suck. I was getting 990ppd when with some WUs I can get 2000ppd.
 
If the points are based on a quad at stock maybe it worth only running one client if you get one of these? - or has the datum machine just got and upgrade?

Snapshot - due you mind seeing it there's a major ppd boost with just the one smp client running. (more so that normal)
or are your ppd figures for a duel core?
 
My numbers are from a dual core PC (E4600 @ 3ghz). I don't remember any mention of Snapshot getting a quad so his are probably dual core too.
 
The 26** units are pretty cache-heavy, so maybe this one's even worse. A single 260* unit on the 1mb E21*0 chips can be as fast as a pair of them.

It'll certainly be worth trying SS's suggestion.
 
Last edited:
I'm only fold one WU at a time. That 2665 was the last SMP WU I folded under winSMP. I'm using ubuntu now and I've got a nice 1760 pointer to start me off. :)
 
I fold on 2 VMs which show 2 cores and only allocate a minimum amount of RAM to each. Turned off the advanced options too. I pretty much get 1760 pointers 100% of the time this way.
 
I must admit i like the consistency of the linux smp - since i've started using them every single wu has been a 1760 pointer.

Every wu counts though :-) remember its science that wins in the end.
 
shadowscotland said:
If the points are based on a quad at stock maybe it worth only running one client if you get one of these? - or has the datum machine just got and upgrade?

Snapshot - due you mind seeing it there's a major ppd boost with just the one smp client running. (more so that normal)
or are your ppd figures for a duel core?
Click to expand...
Will do. Got another WU due to finish on that PC soon so will shut down the affinity changer and see what happens. I've noticed that the 306x WUs don't respond as well as 2653s, etc. to being run two-up on a quad under the affinity changer so it could well be a cache issue. Sounds like a good argument for buying a Yorkfield....

@verbal - had a quad since last August and added a second in September although that one's not on full-time.
 
Last edited:
Absolutely pathetic - it has the undivided attention of a stock Q6600 and has gone from 828 PPD to 1044. I'm starting up the affinity changer and the second client again. Can't afford too much down time or 'Zerker will stomp me before I get to three million and that's not part of my plan....
 
Last edited:
That's not good at all :( and also sounds wrong some how...

Have you done a download new project lately - the first time i got some of the 305* they were showing ppd's of 400ish - was about to post but checked fahinfo first to check out what other were getting - only then did I noticed fahmon was using the wrong points total.

Didn't question it at first as Fahmon usually does an update when it see a new WU for the first time, all the info was showing to the right name type etc all correct, just points was wrong.

my workstation ([email protected] + affinity) out proforms that with a EVERY unit with two less core, and a lower clock :confused:

2651 (1760) - 1067ppd avg (1130 max)
2652 (1148) - 798ppd avg (885 max)
2653 (1760) - 1074ppd avg (1221 max)
3050 (1440) - 944ppd avg (968 max)
3051 (1440) - 929ppd avg (929 max - crunch over a weekend i guess)
3062 (1732) - 986ppd avg (986 max - as above)
3064 (1753) - 953ppd avg (971 max)
3065 (2144) - 841ppd avg (888 max)

Off to check Fahinfo now...
What the time of a single step - use fahlog.txt to check
 
Last edited:
I assure you it's correct as I checked it against the project list (which had actually been updated, for a change). The deadlines are longer which also indicates a slow WU.
 
Snapshot said:
I assure you it's correct.
Click to expand...

I guessed it would be, but thought I'd ask while off collecting the data below.

Fahinfo data.
2665 – AMD - Phenom X4 – Agena – 128.61 ppd/Ghz
2665 – Intel – C2 Quad – Yorkfield – 117.09 ppd/Ghz
2665 – intel – Pent D – Presler – 77.84 ppd/Ghz

Comparison - A nice 1760pt wu

2653 - AMD - Phenom X4 – Agena – 250.24 ppd/Ghz
2653 - Intel – C2 Quad – Yorkfield – 182.43 ppd/Ghz
2653 – intel – Pent D – Presler – 150.08 ppd/Ghz

2653 - Intel – C2 Quad – Yorkfield6M – 274.29 ppd/Ghz
2653 - Intel – C2 Duo – wolfdale – 322.13 ppd/Ghz
2653 - Intel – C2 Quad – penryn3M – 338.13 ppd/Ghz

Looks like your right Snapshot - rubish wu :(
At least it's not the 50% ppd drop (over a 2653wu) that the presler and Agena core's get :eek:
 
Last edited:
shadowscotland said:
Have you done a download new project lately - the first time i got some of the 305* they were showing ppd's of 400ish - was about to post but checked fahinfo first to check out what other were getting - only then did I noticed fahmon was using the wrong points total.

Didn't question it at first as Fahmon usually does an update when it see a new WU for the first time, all the info was showing to the right name type etc all correct, just points was wrong.
Click to expand...

Stanford have adjusted the points value for these WUs, which FahMon can't detect. It looks like you downloaded the project list when the points were at a previous value.
 
You must log in or register to reply here.
Back
Top Bottom