post #1 of 1
Thread Starter 
I'm thinking about a high-end Linux workstation, likely water-cooled unless the Sandia cooler hits the market before my budget allows me to build the machine. Some questions come up:

A hypothetical 3 GHz 8-core machine and 4 GHz 6-core have roughly the same overall power since 3*8 = 4*6, but I think you'd want the extra cores in a server and the faster clock in a workstation. Does that apply here? All these Intel CPUs have 140 Watt TDP, 6900 is 8-core with 3.2 GHz base clock & 4.0 turbo, 6850 6-core 3.6/4.0 and 6800 6-core, 3.4/3.8. For the same cooling solution, will they differ much in tolerance of overclocking?

If it were just linear, heat output = cores * clock * constant, the calculation would be easy & I'd obviously go for the cheaper 6-core since it would run at 8/6 the speed of an 8-core, but I suspect it is nowhere near linear. For one thing, power consumption & therefore heat go up as the square of the voltage and higher clocks often need higher voltage, & there may for all I know be other factors in play as well. If, say, you OC by 20% and need 20% more voltage, you get 1.2 cubed times the heat, more than a 70% increase. Ouch! This might mean the 6-core CPUs cannot be OCd far enough to be better than 8-core, Has anyone got info or links that might help work out the details for this?

Reports I see on the net generally disable Intel's turbo mode and just clock the beast as fast as they can without making it unstable. e.g.:
They report results from 4.7 GHhz on air to 5.3 with expensive exotic cooling, but that is for a 4-core CPU with 65W TDP at stock clock. I assume the chips with more cores & higher TDP will not go that fast without amazing cooling, but what are realistic expectations?

It seem to me the turbo mode is basically a good idea, extra oomph & extra heat only when needed. I'd like to aim at something like 4.0 base clock & 4.8 turbo. Is that likely to be feasible?