Descided to install BOINC on 3 of my linux machines, and old pentium 4 and 2 laptops with decent CPU's. The benchmark results and the time it takes for them to complete tasks, does not make sense, at all. 'Cause it looks like the Core 2 Duo is twice as fast as the Core i7, using the same kernel, the same preferences (through a manager).

The results:

Processor: 4 GenuineIntel Intel(R) Core(TM) i7 CPU M 620 @ 2.67GHz [Family 6 Model 37 Stepping 5]
Processor: 4.00 MB cache
Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good xtopology nonstop_tsc aperfmperf pni pclmulqdq dte

• 3105 floating point MIPS (Whetstone) per CPU
  
• 8237 integer MIPS (Dhrystone) per CPU
  
  MIPS per CPU change a lot, even with performance governor. These were the top values i got, once, couldn't repeat (average was around 2600/6700).
  
• 3105 * 4 = 12420.
  
• 8237 * 4 = 32948.
  
• 4 tasks, 3 to 4 hours completion time per task.

Processor: 2 GenuineIntel Intel(R) Core(TM)2 Duo CPU T9300 @ 2.50GHz [Family 6 Model 23 Stepping 6]
Processor: 6.00 MB cache
Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall lm constant_tsc arch_perfmon pebs bts rep_good aperfmperf pni dtes64 monitor ds_cpl vmx est tm2 ssse3 cx16

• 2782 floating point MIPS (Whetstone) per CPU
  
• 10778 integer MIPS (Dhrystone) per CPU
  
  Dhrystone MIPS per CPU is quite higher as the i7.
  
• 2782 * 2 = 5564.
  
• 10778 * 2 = 21556.
  * 2 tasks, 1 to 2 hours completion time per task.

Which boils down to a mobile Core 2 Duo actually performing twice as well as a mobile Core i7, even though the complete MIPS are lower.

Again. Does not make sense.

Don't pin yourself on benchmarks. They don't show the actual reality of what your CPUs are capable of. You can even influence the outcome of the benchmarks, by loading (specific) programs in memory.

I've always seen on all Windows version that when I load any version of Winamp and then allow BOINC to do benchmarks, that the integer values are off by 500 - 1,500 MIPs, while the FPs are off by ~200 MIPs as opposed to when running no Winamp.

Even running 10 benchmarks in a row will give 10 different values.

And when it comes to comparing tasks, then I really have to point you back to the project forums of the project you're trying to compare. No task is exactly alike, and if you're trying to compare Seti tasks here, then you have to take into account that different angle rates give different run times on the same CPUs. If you're trying to compare Einstein tasks, same thing...

---

Don't pin yourself on benchmarks. They don't show the actual reality of what your CPUs are capable of. You can even influence the outcome of the benchmarks, by loading (specific) programs in memory.

I've always seen on all Windows version that when I load any version of Winamp and then allow BOINC to do benchmarks, that the integer values are off by 500 - 1,500 MIPs, while the FPs are off by ~200 MIPs as opposed to when running no Winamp.

This was the Core i7 beeing completely idle (constant 100% BIONC usage on all cores) the whole night, nothing else running, and the Core 2 Duo with around 5-10% constant CPU usage since it's running a media centre as well as several IRC connections.. and i even watched a 45 minute video on it.

If that (applications stealing idle time) would have been the problem, the Core i7 laptop would "have won".

And when it comes to comparing tasks, then I really have to point you back to the project forums of the project you're trying to compare. No task is exactly alike, and if you're trying to compare Seti tasks here, then you have to take into account that different angle rates give different run times on the same CPUs. If you're trying to compare Einstein tasks, same thing...

This would only be valid, if certain CPU's (or amounts of) request bigger, and other smaller tasks. I've been monitoring the completion time of all tasks on both hosts, and they vary, but only for 5 to 15 minutes on one host, not up to double the time.

This is just the AndrOINC (follow up to the MilestoneRSA) project on all machines.

I've found a trick in the FAQ, to stop BOINC, change the correction time to 1.000000 (it was set to 3.XXXXXX), and start BOINC again.. but that didn't change anything.

The reason i'm digging into this, is 'cause i wonder why the i7 isn't performing as expected, and as i build my own kernels and tune system params, i wonder if it's something to my own blame. So if there's anything to check, or make really sure it's nothing to worry about, i'd love to hear it.

Thanks in advance.

After searching about the i7 performance, i quickly hit a few topics about the "Turbo mode". Turns out that once all CPU's are beeing used, it can't go into that mode. Even though i don't believe this should result in more tasks beeing completed, i thought it was worth a try.

Now, the AndrOINC project has files of 258 byte big, all the same size, with all the same contents: 2 strings that need to be cracked, of exact same size, all the time. So, the fact that the task time was different constantly on this i7 (and not on the others, like the Core 2 Duo), was weird imho.. but that might be the fact that my knowledge lacks in that area.

So i set BOINC to 3 threads instead of 4 (1 physical package, 2 cores, 2 HT threads per core = 4 CPU's showing up). Starting BOINC again was awesome:

Dhrystone: ~6600 > ~8600,
3 tasks at the same time, 55 minutes per task.

Unfortunately, and this puzzles me, it didn't take long. After the first 55-minutes tasks, the time increased and increased.. and in the end now it's taking around 3.5 to even 7 hours per task, to the level the old Pentium 4 needs for a task, and the Core 2 Duo continueing with 1 hour 15 minutes steady for days for each task now.

Maybe i fail to understand how it works.. but this is weird. The two Core 2 Duo's that are crunching are pumping out tasks as if it's their favorite thing to do.. and the Core i7 doesn't perform at all and falls back to a sad sluggish thing. So far haven't found what could cause this performance degradation and problem, the i7 Turbo does kick in (from 2.67 to 2.87 MHz with 3 threads). I'm gonna ask on the AndrOINC forums as well for some insight in how it works, but my bets are that there are some issues here.