Quote Originally Posted by villa1n View Post
toms link.

It is not temp downclocking. You can see these options in the bios screenshots as well, on the EE edition you can override them.
What is hardwired into the chip, is it is limited to powerdraw as well, not only temp. So if you buy a i920, and you luck out, like i did with my q6600 and your tdp is way down lets say 72 watts for arguements sake, you ve got massive headroom and the volts would probably get to high before you reach that wall. However, there is a spread , like in all processors, so if you get an i920 that has 110W tdp at default volts you wont get very far before that wall comes into play. Since the i920 i would assume is the lower bins, the luck of getting a low TDP chip is not on your side, and this will otherwise limit oc'ing for a chip, that might have to be pushed harder than another.
The benefit or protection element doesnt make sense to me. If you have temp based down clocking.. i dont see the necessity for this, as you void your rma abilities if you go beyond spec anyways... so either they found a high inciedence in fried chips.. meaning this can't handle volts well (which i doubt) or they wanted to stop budget chips from getting up to 4 - 4.5 on air... it makes perfect business sense, just hoops us budget oc'rs who cant drop 1k on an EE chip.
I'm not 100%, but I'm pretty sure it's not power based downclocking, it's merely the basis for when turbo mode is allowed to kick. The override on the 965 allows you to set that limit so that turbo mode functions nearly all the time when the cpu is in use. On the locked chips, turbo mode only kicks in when this headroom is available. There is no forced thermal throttling when these chips reach this point, just turbo mode won't kick in.

OC'ing will still work, just if you get close to the TDP, you probably won't get that extra oomph to kick in.


Quote Originally Posted by Jacky View Post
......


Most reviews conclude that tri-channel provides close to zero benefits, is there a reason why people still think they need tri-channel?
Right now, the cpu probably doesn't have quite enough execution throughput to fully utilize the full memory bandwidth that is available to it when all 8 virtual cores are running full bore. There's also the issue that most programs can't even get close to running these chips full bore, so there's plenty there to be used to fulfill the cpu's every whim. Prefetching probably also puts some extra use on the memory, but not much. As you can see from some benchmarks, there is a boost, just only when the cpu is pushing all 8 threads to the max.

Currently the real reason for the extra bandwidth comes from the server space. Since gainstown introduces a NUMA (non-uniform memory access) type platform on intel's line, there will be calls from one cpu to the other to get stuff stored in either chips local memory. This is the case where the bandwidth comes into play. Say CPU0 is running full speed working on some verry threaded server workload, sucking up as much memory bandwidth as it can take. Then CPU1 comes along after not finding what it needs in the memory attached to it, and requests it from the memory attached to CPU0. The bandwidth is still available to fulfill this request without causing any slowdown in what CPU0 is doing.

The real boom for tripple channel probably won't happen till westmere comes out with it's 12 virtual cores that will be even more hungry for bandwidth. Right now, the most people here with bloomfield won't really see much gain unless you're using very threaded programs, but I'm pretty sure that all the crunchers here who are thinking about a gainstown platform will see big improvements when filling all 3 channels to both sockets.