Forum > Windows

Core 2 QX6700 SSSE3 crunching time

(1/6) > >>

jezek:
Hello I am using optimalized app on several PC's and it runs great. But on my Kentsfield I run version 1.41 with SSSE3 and I have worst results than others with P4 @ 2.6GHz as I see. They doing it in 2 000s but my QX6700 in 4500s. Can somebody help me please?

BenHer:
Not sure exactly what you are referring to...
Pehaps it could be each WU has an angle range that was used when it was collected at arecibo.

To compare how long a given WU takes for different optimized crunchers you need to make sure you are comparing WUs on each machine that have as close an angle range as possible.

Angle ranges are not reported in normal web areas when you are looking at your results pages, instead from a results page you must click on any link, under the Result ID column (for completed WUs) and look in the "stderr out" area for "True angle range:" or "ar=".

jezek:
I'm sorry. I'm talking about same WU on 2 different PC with different application. 1.PC is QX6700 with SSE3 optimalized client 2.P4 @ 2.6GHz with un-optimalized client. They doing same WU (with same true angle rate).
My optimalized QX6700 had 4700second for this WU and really much slowest P4 with unoptimalized client needs only 2000s.I just want to know why? Is there problem in client?

Jason G:
A couple of ideas there that may or may not apply:

first remember a Test WU is run on one core only: so it alone probably shouldn't run all that much faster than the p4  running one WU but:
1- In real use The kenstfield would run 4 workunits at once making the average WU time 4500/4 = 1125 secs ... approx twice as fast as the P4 ( @ 2000secs ) many people are eagerly awaiting a new appp for these cpus, you aren't alone.

2- The p4 would be running v2 stock apps? these are a lot newer ( better ? ) than 1.41 on most all other cpus except the core2's....

3 - people have been discussing on the seti@home forums about memory bandwidth coming into the problem,  4 processors competing on the same memory bus is a challenge and even people running Xeons with ram in quad channel mode are finding a bottleneck.

4 - let them run and compare individual RACS when stabilsed... that may be a more reliable judge.

Just some ideas, Wish I owned a kentsfield. Good luck finding out what's going on :D

jason

msattler:
Yes, memory bandwidth is a MAJOR concern with a quad.  4 cores competing for memory access is a definate bottlneck.  If you are OC'ing, back off on the RAM timings and go for more bandwidth, it helps...I am here to tell ya.
Also, if you are just looking at crunching times in the last week or so....SETI has been parsing out a lot of crappy little WUs lately (trying to keep it polite, I could use other terms).  Cleaning the dusty tapes out of the vault (Matt's bookshelf).  Much crunch, few points.  The 1.41 app for the core 2 is the best thing going at this time overall, but it sure has a weak point, and they are it.
So, hang in there, check the results you are looking at to see if the rig you are ocmparing to is OCd or at stock speed.  Check your rig to see if you have thermal limiting downclocking your cpu.
Good luck and good crunching.

Navigation

[0] Message Index

[#] Next page