Hello,
I have been cruching Milkyway WU's for over a year on most of my main PC with a few other projects on a lower resource shares.

The main cruchinch PC was getting an RAC 5000'ish when running 24hrs and when I cut the time the PC was switched on for down to 12hrs/day, the RAC settled at 2200'ish and has been that way for 6 months now.

After processing 2 million credits I have put Milkyway into the background and give SETI a very high resource share. I did this at the begining of the this year, however after 3 weeks this particular PC is no longer seeming to perform and has only reached a RAC of 700 and now seems to be flatting out.

I am using optimised apps on all the PC's (Lunatics latest) and if you look at my computers you can see that much slower processors/PC's are technically doing better with their RAC's yet the main crunching PC which is running AK_v8b_win_x64_SSE41.exe and is a Intel(R) Core(TM)2 Quad CPU Q8300 @ 2.50GHz is doing poorly.


I can see all cores processing on this PC's and everything is the same, all I did was install the latest lunatic installer and chose SSSE4.1 which is what CPUz states.

My PC's are viewable at http://setiathome.berkeley.edu/hosts_user.php?userid=295004 and the PC which seems to me as really underperforming is is Tom-PC and in number http://setiathome.berkeley.edu/show_host_detail.php?hostid=5243910

Any ideas why the RAC is low on this Quad core and how can I improve the RAC back up to 2200'ish as it was on Milkyway

Thanks

---

I'd try the SSSE3x app instead of the SSE41 app, the SSE41 app tends to only be faster on fast Dual cores with fast memory,
on quads there's more memory contention, and the SSE41 app tends to be slower, as always, your mileage may vary,

Claggy

Do you have HT turned on? Given the smallish amount of memory Tom-PC has (< 3GB), that might also cause a memory contention problem.

---

Core 2 Duo's and Quads don't have HT,

Claggy

Also consider adding CUDA-capable cards.

MW is known to give more Credits than Seti and many other projects, on my machines it's about 2x more than Seti (both projects with opt. apps).

---

Core 2 Duo's and Quads don't have HT,

Claggy

Whoops - my bad.

That's because I use AMD Barcelonas, so I'm ignorant of Intel details.

---

If you're in it for sheer abs quantity of numbers, this is not the place. Other projects process different specific types of data, thus may or may not be able to create a greater credit 'score' per unit of time.

Might also check and see if something else is taking CPU time as your Q8300 is taking almost 3 times as long as my Q8200.

I am running the SSSE3 app. Along with the OpenCL MB app for ATI GPU's.

---

Sean Arrowsmith wrote:

...
My PC's are viewable at http://setiathome.berkeley.edu/hosts_user.php?userid=295004 and the PC which seems to me as really underperforming is is Tom-PC and in number http://setiathome.berkeley.edu/show_host_detail.php?hostid=5243910

Any ideas why the RAC is low on this Quad core and how can I improve the RAC back up to 2200'ish as it was on Milkyway

Thanks

RAC isn't going to be suitable to judge whether the computer is underperforming, because credits are not exactly comparable between projects as others have pointed out, because you're not crunching 24/7, and because RAC takes about 5 weeks to stabilize.

The "Average processing rate" (APR) shown on the Application details is a suitable measure. It's calculated the same for all computers doing SETI@home Enhanced work from the run time and splitter specified fpops, and is nominally in GFLOPS. That computer achieved about 5.31 when it was using the stock 6.03 application, and is nearly at 6.55 with the optimized SSE4.1 build. The ~23% improvement is at the low end of what I'd have guessed and I agree that the SSSE3 build is likely to give a small improvement.

But what caught my attention immediately is that 6.55 figure very nearly matches my Pentium-M laptop at 1.4 GHz and running the 32 bit SSE2 optimized application. Your Q8300 at 2.5 GHz and with the Core 2 architectural improvements ought to be at least twice as fast as my P-M.

To get a better idea of what it might do I looked for some other Q8300 systems within the top computers list. Two running 64 bit Windows and the stock 6.03 are 5306191 and 5525417, both have much higher APR than yours did for stock. Then there's 5259251 running 64 bit optimized with APR 16.63 but it's the SSE3 rather than SSSE3 version. Finally, two running optimized 32 bit with APR more than twice yours, 4869855 and 5752798.

It's serious underperformance, and I don't see any good clues to what's causing it. Run time isn't too much more than CPU time as happens when some other process is stealing a lot of CPU, your benchmarks are in line with the other computers I've referenced, you have plenty of RAM to support four instances each using ~50 MiB, etc. But I have no experience with Win 7 64 bit nor multicore CPUs, I hope those with better knowledge will have some helpful ideas.

                                                                 Joe

If you're in it for sheer abs quantity of numbers, this is not the place.

Been doing SETI for 10 years now, so it's not about the numbers. Just trying to make sure that if I do workunit then at least I am getting the best I can out of the equipment I am using after all it costs me money to run SETI and other projects in electricity.

---

Thanks.

I have now downgradge to Lunatics SSSE3 2 days ago and shall see what happens over the next week.

I have looked at the process priority and the apps run in low priority by default everytime the PC runs. I have altered the priority to high several times over the last few weeks but does not seem to make a difference and as soo as the PC shuts down then reset to low again on restart. When the projects apps run they all each hog 25% of the Q8300 (or 100% of each core as it depends how you look at it)

Perhaps running the 32bit version might work?

There are no other major applications running on the PC for most of thr 12hours apart from the last 4 hours when my Son uses the PC for video work (Aninmation) and that is just really taking snapshots via a webcam.

Perhaps alot was my expectation that all projects allocated WU credits evenly across all projects so SETI would give the same as Milkyway.

I shall look at the computer Id's you have quoted and see if or what I can do to make things work better.


Thanks

Sean Arrowsmith wrote:

...
My PC's are viewable at http://setiathome.berkeley.edu/hosts_user.php?userid=295004 and the PC which seems to me as really underperforming is is Tom-PC and in number http://setiathome.berkeley.edu/show_host_detail.php?hostid=5243910

Any ideas why the RAC is low on this Quad core and how can I improve the RAC back up to 2200'ish as it was on Milkyway

Thanks

RAC isn't going to be suitable to judge whether the computer is underperforming, because credits are not exactly comparable between projects as others have pointed out, because you're not crunching 24/7, and because RAC takes about 5 weeks to stabilize.

The "Average processing rate" (APR) shown on the Application details is a suitable measure. It's calculated the same for all computers doing SETI@home Enhanced work from the run time and splitter specified fpops, and is nominally in GFLOPS. That computer achieved about 5.31 when it was using the stock 6.03 application, and is nearly at 6.55 with the optimized SSE4.1 build. The ~23% improvement is at the low end of what I'd have guessed and I agree that the SSSE3 build is likely to give a small improvement.

But what caught my attention immediately is that 6.55 figure very nearly matches my Pentium-M laptop at 1.4 GHz and running the 32 bit SSE2 optimized application. Your Q8300 at 2.5 GHz and with the Core 2 architectural improvements ought to be at least twice as fast as my P-M.

To get a better idea of what it might do I looked for some other Q8300 systems within the top computers list. Two running 64 bit Windows and the stock 6.03 are 5306191 and 5525417, both have much higher APR than yours did for stock. Then there's 5259251 running 64 bit optimized with APR 16.63 but it's the SSE3 rather than SSSE3 version. Finally, two running optimized 32 bit with APR more than twice yours, 4869855 and 5752798.

It's serious underperformance, and I don't see any good clues to what's causing it. Run time isn't too much more than CPU time as happens when some other process is stealing a lot of CPU, your benchmarks are in line with the other computers I've referenced, you have plenty of RAM to support four instances each using ~50 MiB, etc. But I have no experience with Win 7 64 bit nor multicore CPUs, I hope those with better knowledge will have some helpful ideas.

                                                                 Joe

---