My new laptop: Linux 64-bit vs Windows 32-bit using 2.2B SSSE3 app

[michael37]

I just got a new laptop, and, of course, I had to try to find better OS for it.  My currently installed OS versions are Microsoft Vista Home Basic 32-bit and Ubuntu Linux 7.04 64-bit.

To my disappointment, Linux 64-bit is not faster than Windows 32-bit.

Since SETI gets rid of solved workunits too quickly, here are the results rather than links to the workunits.

================ LINUX =============
CPU time   7240.77652

<core_client_version>5.8.16</core_client_version>
Optimized SETI@Home Enhanced application
Optimizers: Ben Herndon, Josef Segur, Alex Kan, Simon Zadra
Linux port: Crunch3r, Hans Dorn, Simon Zadra
   Version: Linux 64-bit based on S@H V5.15  'Noo? No - Ni!'
  Revision: R-2.2B|xT|FFT:IPP_SSSE3|Ben-Joe
     CPUID: Intel(R) Core(TM)2 CPU         T5300  @ 1.73GHz
      CPUs: 1, cores: 1,  cache: L1=32K,  L2=2048K, L3=0K
  Features: MMX  SSE  SSE2  SSE3  SSSE3 x86_64 
     Speed: 1729 MHz

Work Unit Info
True angle range:  0.425850

Spikes Pulses Triplets Gaussians Flops
   0      1       6        1     16109414470855



================ WINDOWS =============
CPU time   7213.595441

<core_client_version>5.10.12</core_client_version>
Optimized SETI@Home Enhanced application
Optimizers: Ben Herndon, Josef Segur, Alex Kan, Simon Zadra
   Version: Windows SSE3-Core 2 32-bit based on S@H V5.15  'Noo? No - Ni!'
  Revision: R-2.2B|xT|FFT:IPP_SSE3|Ben-Joe
     CPUID: Intel Core 2 Duo 'Conroe 2M'
      CPUs: 1, cores: 2, threads: 1   cache: L1=32K,  L2=2048K, L3=0K
  Features: MMX, SSE, SSE2, SSE3, SSSE3 
     speed: 1729 MHz  -- read MB/s: L1=6452, L2=5583, RAM=3286

Work Unit Info
True angle range:  0.426468
Restarted at 21.57 percent.

Spikes Pulses Triplets Gaussians Flops
   0      0       0        1     16094383483838

[Metod, S56RKO]

To my disappointment, Linux 64-bit is not faster than Windows 32-bit.

There are a couple of things one should keep in mind when making comparisons:

• The WUs you were comparing were similar but not identical. It is expected to see run-time difference of a couple of percent when running similar WUs by same executable and on same hardware. The time difference you observed is negligible from this point of view.
• All applications by Chicken soup are highly optimized for a given platform.
• Let's compare 32-bit and 64-bit capabilities of your computer:
• If running 64-bit application, it will use 64-bit pointers and 64-bit integers instead of 32-bit counterparts. If application only needs less than 2GB memory and does integer arithmetics on numbers that are less than approximately 2*10^9 (two billion), then using 64 bits will slow down the app.
• If an application uses many registers, it'll benefit from 64-bit environment as there are more general-use registers available in that mode. Gain will be negligible to considerable, depending on how heavy registers are in use.
• If an application uses floating-point arithmetics, then it really doesn't matter as FP part of processor is not changed in any way.
• If an application uses any of SS(S)E instructions, it'll be roughly the same. Most of SS(S) extensions are bitness-agnostic.
• The real benefit from using 64-bit platform in Linux comes from the fact, that default GCC configuration has been changed from using 387 (FPU) floating-point arithmetics to using SSE arithmetics. This could have been done earlier, but it really is safe to do so on 64-bit platform since there aren't any x86_64 processors (neither Intel nor AMD) that don't support at least SSE2 instructions.

As I wrote before: all Chicken apps are highly optimized (that is, they use instructions as advanced as possible for any given CPU) and you can't really expect to have 64-bit apps faster just because they are 64-bit.

What is most remarkable about Chicken apps is the fact, that Linux apps are generally (at least) as fast as Windows counterparts. This didn't use to be the fact until Chicken and the flock started to do their marvelous work. Kudos to all of them.