On Sun, Jun 14, 2009 at 12:31 PM, Christian Packmann<Christian.Packmann@xxxxxx> wrote: > After looking through the results, I think that a SSE2 codepath is not > interesting, as the MMX/SSE code always performs better. A SSSE3 version may > make sense for modern Core2 and i7/Nehalem systems, and maybe AMDs Bulldozer > (due 2011). So I'll try to optimize the MMX/SSE routine further, as it is > the most useful for general use. I have a Via C3 800mhz machine somewhere that I might dredge up and test... it's been a long time since I used it with Haiku (or anything for that matter). It's such a dog... might be fun though. > Urias McCullough - 2009-06-14 19:56 : >> >> On Sun, Jun 14, 2009 at 10:52 AM, Stephan Assmus<superstippi@xxxxxx> >> wrote: >>> >>> Wow, so my lowly Core 2 at 1.8 GHz really is 2.4 times faster at this! I >>> didn't think the efficiency increase between Pentium 4 and Core 2 was >>> that >>> awesome. > >> >> >> What's really strange is that the PIII 450mhz seems to be twice as >> fast as the P4 at C code...something ain't right. > > Uhm, the results are in clock cycles. :-) This allows for easy comparison of > efficiency between CPU architectures without having to normalize for clock > speed of a given system. Aha, that clears things up. > P-III: > The Pentium-III performs very well on integer; on par with the K8 > architecture per clock cycle. I wouldn't have expected this. But of course > this is offset by the much lower clock speed of the PIII. Yes, I was under the impression that the P4 architecture was quite slow compared to PIII, which is why Pentium M and Core arches were ultimately based on the PIII core design rather than extending P4 further. P4 allowed higher frequencies and more power utilization (I see this clearly by comparing the power used at the wall outlet between a P4 system and others) I believe I have read that the P4 design pretty much died with the Pentium D line due to the horrible inefficiencies ;) > It's also interesting to see that the Atom is basically as efficient per > clock as the P4 on the integer version. Of course the P4 runs at much higher > clocks, so the 2.8GHz P4 will be 2.8/1.6 = 1.75x faster than an 1.6GHz Atom > in absolute terms, but it's still interesting that a low-power architecture > like Atom can perform so well in comparison. Imagine how many Atoms you can > run for the power budget of one P4. :-) Yes, while the Acer Aspire One is quite pleasant to use Haiku on, the Atom is extremely slow for CPU-intensive tasks. This demonstrates for me that the raw CPU speed is less important for usability and perceived speed at least ;) > And GCC2 vs. GCC4 system really doesn't matter, the benchmark code itself > doesn't call the OS. The measured minimum times should be pretty consistent > regardless of Haiku version, and this is what matters. I'm glad to hear that, although I'd be happy to compile a Haiku/gcc4 version of your benchmark if you want :D - Urias