> > Or - is there another alternative - which is that in order to achieve these > 'insane' > > Sandy Bridge overclocks you have to use either a very expensive cooling solution, > or > > a very loud one, or both, which thus prevents Intel from selling these things at > such > > high clock rates? > > I'm using a ~$50 air cooler to run 1 GHz over stock, and it's got 2 large slow fans > in a push-pull configuration, so it's no louder than the stock Intel cooler. > Remember, Intel was shipping the Prescott P4s about 6 years ago, and those things > idled very hot, let alone under load. If they brought back that cooler they'd be all > set for stock-rated 4+ GHz parts I think.
Let me preface this remark by saying that $50 additional for a better heat sink/fan probably does not alter the fact that the Sandy Bridge has superior price:performance when compared to most (all?) other x86 chips, at least for the MAME workload.
However, it's unknown how much Intel would have to charge for a Sandy Bridge with a stock cooling solution that could reach the overclocked speeds you have quoted without excessive noise or size; I think that stock coolers generally are low profile to reduce the potential for layout issues that wouldn't allow the processor + heatsink/fan to fit in the majority of cases. And you are not going to get a 120 mm push-pull configuration in a low profile cooler. Which means that they would have to have a fan that is low profile but with the same heat dissipation capacity as your 120 mm push/pull, and such a cooling solution might add significantly to the cost of the Sandy Bridge chip.
Of course, the Sandy Bridge as an OEM part without a fan would not carry this burden and could be lots cheaper even if rated to run at higher clock speeds.
Also there is the power consumption angle; I don't really know how much power a Sandby Bridge uses when overclocked (did you have to overvoltage to get to 4+ Ghz? If so that is troubling for alot of reasons, processor lifetime not the least of them), but that factors into how much the CPU can be sold for. Nobody wants 150 Watt P4s anymore (and for good reason, they sucked in so many ways).
Finally, would you be able to confirm your earlier claim that your overclocked Sandy Bridge can compile the entire MAME source tree from a clean state in 30 seconds? I find this claim incredible considering that my 3.0 Ghz 6 core Phenom II can only do it in 2.5 minutes. A 5x speedup with 2 fewer cores just sounds unbelievably incredible. Although, that being said, maybe there is something wrong with my setup; I think I calculated that the compiled lines of code per second for the MAME tree at 2.5 minutes is something like 2700 lines second per processor for my 6 processor system ((~ 2.5 million lines of MAME code) / (6 procs * 150 seconds = 900 seconds) = ~2700 lines of code per second per processor), and I just expected it to be much higher than that given that the old Sun Sparcstation that I used to use in university in 1992 could do 1000 lines per second (of course code was simpler back then - C vs. C++ - but still ...).
|