For cross-ref. here's one of my Quad's:

8/17/2009 3:13:36 PM Starting BOINC client version 6.6.36 for windows_intelx86
8/17/2009 3:13:36 PM log flags: task, file_xfer, sched_ops
8/17/2009 3:13:36 PM Libraries: libcurl/7.19.4 OpenSSL/0.9.8j zlib/1.2.3
8/17/2009 3:13:36 PM Data directory: C:\Documents and Settings\All Users\Application Data\BOINC
8/17/2009 3:13:36 PM Running under account Boss
8/17/2009 3:13:36 PM Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9650 @ 3.00GHz [x86 Family 6 Model 23 Stepping 10]
8/17/2009 3:13:36 PM Processor features: fpu tsc pae nx sse sse2 mmx
8/17/2009 3:13:36 PM OS: Microsoft Windows XP: Professional x86 Edition, Service Pack 3, (05.01.2600.00)
8/17/2009 3:13:36 PM Memory: 3.00 GB physical, 4.84 GB virtual
8/17/2009 3:13:36 PM Disk: 455.82 GB total, 425.58 GB free
8/17/2009 3:13:36 PM Local time is UTC -5 hours
8/17/2009 3:13:37 PM CUDA device: GeForce 9800 GT (driver version 18618, compute capability 1.1, 512MB, est. 60GFLOPS)
8/17/2009 3:13:37 PM CUDA device: GeForce 9800 GT (driver version 18618, compute capability 1.1, 512MB, est. 60GFLOPS)
8/17/2009 3:13:38 PM Not using a proxy
8/17/2009 3:13:38 PM World Community Grid URL: http://www.worldcommunitygrid.org/; Computer ID: XXXXXX; location: (none); project prefs: default
8/17/2009 3:13:38 PM World Community Grid General prefs: from World Community Grid (last modified 15-Aug-2009 23:02:43)
8/17/2009 3:13:38 PM World Community Grid Host location: none
8/17/2009 3:13:38 PM World Community Grid General prefs: using your defaults
8/17/2009 3:13:38 PM Preferences limit memory usage when active to 3069.40MB
8/17/2009 3:13:38 PM Preferences limit memory usage when idle to 3069.40MB
8/17/2009 3:13:39 PM Preferences limit disk usage to 200.00GB


8/20/2009 4:24:06 PM Benchmark results:
8/20/2009 4:24:06 PM Number of CPUs: 4
8/20/2009 4:24:06 PM 3113 floating point MIPS (Whetstone) per CPU
8/20/2009 4:24:06 PM 6284 integer MIPS (Dhrystone) per CPU

Beyond me why the BOINC benchmark is 1/5th of regular, also because the benchmark runs at normal priority by default and not at idle level priority which the sciences run at. I'd be interested to hear how the benchmark performs when boinc.exe priority is elevated to 1 below realtime. Also interested how the machine does running OCCT, linpack test which sure does my quad fans kick in max gear.

There is a prior discussion on this in a thread. Fire up 6.6.38, open up process manager Explorer, threads tab, hit benchmark and watch boinc open a new thread. Click this new thread opening, showing I/O Normal taking a bunch of cycles, then disappearing the second the benchmark finishes. Other test. Set the BOINC to normal and you get benchmarks all over the place depending on system load. Set boinc client to HP and the benchmark is steady as a rock. So where to see it's at idle level?

Edit: Manager to read Explorer.

Hi All,

We've been kind of busy these last few days so we apologize for not posting anything back regarding our status and progress.

We think we've finally figured this beast out. Nothing like the older processors so there was a steep learning curve for us, but we think we have it now.

We had the system OC'd to 4Ghz (yes we've had it higher just for testing) and could complete 8 FAH work units completing like this:

faah7810_ ZINC01381120_ xmdEq_ 1MSN_ 00_ 0--  	Valid 	8/21/09 23:16:25 	8/22/09 04:59:16 	2.39 	70.7 / 92.0
faah7810_ ZINC04887873_ xmdEq_ 1MSN_ 00_ 0--  	Valid 	8/21/09 23:07:57 	8/22/09 04:59:16 	2.21 	65.2 / 84.7
faah7809_ ZINC04944280_ xmdEq_ 1MSN_ 01_ 0--  	Valid 	8/21/09 22:27:14 	8/22/09 03:44:23 	2.18 	64.0 / 86.1
faah7809_ ZINC04944504_ xmdEq_ 1MSN_ 01_ 0--  	Valid 	8/21/09 21:56:30 	8/22/09 03:23:17 	2.22 	65.3 / 83.5
faah7809_ ZINC01342715_ xmdEq_ 1MSN_ 00_ 0--  	Valid 	8/21/09 21:30:47 	8/22/09 02:59:30 	2.35 	68.7 / 88.1
faah7809_ ZINC04952178_ xmdEq_ 1MSN_ 01_ 0--  	Valid 	8/21/09 21:09:12 	8/22/09 02:59:30 	2.20 	64.2 / 85.1

At that time we were seeing Floating Point calculations in the 4100 / cpu and integer calculations of about 9500 / cpu (according to the cpu benchmark took for bionc).

Our temps during that period time were acceptable (65 - 66) we decided to back the system off just a bit and are now running at a modest 3.4. with temps hovering between 58-61 which we believe is well within reason.

At our current settings we are seeing this for the cpu bench marks.

8/22/2009 1:03:09 PM||Benchmark results:
8/22/2009 1:03:09 PM||   Number of CPUs: 8
8/22/2009 1:03:09 PM||   3300 floating point MIPS (Whetstone) per CPU
8/22/2009 1:03:09 PM||   8232 integer MIPS (Dhrystone) per CPU
8/22/2009 1:03:10 PM||Resuming computation

We did make one interesting observation. When the system is first booted and after the applications are all available (right now it's just CPU-z; RealTemp and Bionc) when the cpu bench mark tool is used, it appears to be rather unreliable in all cases. This is even more evident when the database with our results was limited. Why? We don't have a clue.

Obviously, attempting to use something like the CMD2 wu's which have a fixed amount of time did little to us actually for instrumentation purposes.

The other thing we have observed and are really curious about is with the cpu benchmark process in and of itself. As we type out this post, the CPU is churning along at a nice 3.46 with the 58-62 temp range on the CPU. When the CPU benchmark runs, the temps on all cores drops to 38-44 and then when the work units take over again, it's off to the races again with the higher temps. This is certainly something we can't explain, but it is observable and causes us many questions.

Anyway, as we stated in the beginning of this post, we think we have finally dialed this thing in so we understand better where our ranges are for overclocking the machine.

We would like to thank everyone for their kind assistance in helping us figure this out and we really appreciated everyone's assistance, comments, direction and guidance.

Now we have a machine that can more closely compete with our Granddads's system and if we want, we can really kick his (_!_)

So what was the root cause, change you made that kicked the benchmarks into 'normal' levels of performance?

Ingleside, seems there's a field I missed and was focusing on I/O priority. At top of the thread view it says Base priority: 1. Remains odd that changing the Parent priority to higher always produces the near same benchmark, whereas at normal it's a yo yo... mostly the integer part, in the composite rather overweighted to me considering the prevailing floating point comps we do.