Slashdot Mirror
Linux Hardware Looks at Core 2
Penguin Lover writes "Linux Hardware has just posted a new story on how Intel's new Conroe performs under Linux. From the article: 'Now is a great time to be CPU shopping because no matter which side of the isle you look on, you have great choice for both CPUs and motherboards. Along with Intel's chipset offerings, keep in mind that NVIDIA has the nForce series for Intel CPUs which would give you SLI support for all your Quake Wars and UT2007 gaming needs.'"
The Core2 E6600 seems to be a nice bang for the money as it is right around the middle of the currently available speed grades and is the cheapest iteration with 4MB of L2 cache. It also seems to be competitive with dual core AMD products (my usual choice) that are substantially more expensive. It's definitely a buyer's market these days...let the price war begin! :)
I'm going to buy an e6600-based desktop in the winter, and I'm wondering if GCC will add tuning for the Core 2 processors. My guess is that '-Os -march=pentium-m -msse3 -mfpmath=sse' is the way to do it now in 32-bit mode, but there have been enough changes in the Core 2 to make for a new DFA profile, no? I thought they radically revamped the Pentium3 core, adding execution units and such.
I just rebuilt my Core (1) Duo laptop with the aforementioned options and it seems to get me the most bang for my buck.
From what I've read, compiling and running in 32-bit mode is still the best choice for now, the Core 2 is a lot better than the EM64T pentium 4 at running 64-bit code, but still not as fast as just using the 32-bit code (stuff like video encoding is happening in the 128-bit vector unit anyway, and I don't need more than 4GB addressable space).
"Sometimes, I think Trent just needs a cup of hot chocolate and a blankie." -Tori Amos on Nine Inch Nails
A good point, but keep in mind that the performance and power differences between Prescott (90nm) and Cedarmill (65nm) were decidely underwhelming. The greater leakage current in Intel's 65nm process ate into most of the gains from the smaller process. It did make the die smaller and thus cheaper, but not much better in either performance or power consumption.
Perhaps the AMD/IBM SOI process will do better at 65nm in controlling leakage current and provide the needed performance boost. Intel plans to release the 45nm Penryn in 2H07, and claims to have greatly improved the leakage current situation. AMD needs to leapfrom Conroe's performance with their 65nm part to remain competitive. Schedule is critical here: if Intel is late, AMD will regain momemtum; if AMD is late, Penryn will make the 65nm AMD part unattractive.