Sorry for three posts in a row, but I don't want to post in the same post to avoid confusion. I have done some more extensive testing now and I am convinced that the TDP limiter of some sort is kicking in causing the processor do downclock. How did I determine that? Read on and keep in mind all tests were done with the chip clocked to 4.4GHz, all throttling and power control set to disabled in the BIOS, all C-states disabled in the BIOS as well.
First I ran Cinebench, and the clock does not flinch at all. It stays solid at 4.4GHz for the whole run and the results correspond to it (~12.5pts at 4.4GHz). This is consistent with Cinebench load which is not that high even though usage is at 100%.
Next I ran LinX and set the memory to use to approximately ~5GB in one test or at full 32GB in the second test and then monitor the load. While the memory is filling up the clock is solid at 4.4GHz. As soon as the memory is all loaded and high computational load kicks in, the throttling starts. The time before the throttling kicks in exactly corresponds to the second when the LinX process changes from loading data into memory to performing calculations.
Ok, at this point things are really pointing towards a TDP limitation of some sort, but I decided to test further so I decided to do an even more controlled test.
I loaded up Prime95 and limited the number of threads to 4. No throttling as you can see below.
Then I decided to set the number of threads to 8. Throttling starts, but it is not quite so aggressive yet as you can see below.
Finally, I decided to go all out with 12 threads. Not only is the throttling kicking in majorly, but it is pretty much constantly throttling the chip as you can see below and very aggressively too.
For the record, I have seen this behavior before...a long time ago when the first Nehalem chips came out. Asus botched their implementation of the Turbo feature and TDP limits. That is why this tool came out http://www.xtremesystems.org/forums/archive/index.php/t-225450.html
after I and a few others worked with Asus to get the issue resolved (initially in that case I was also convinced that there was no problem and that everything worked "as designed"). In that case things were a bit better because even though the chip would throttle when using Turbo and exceeding TDP threshold at least fixed multipliers worked with Turbo disabled and were unaffected by the TDP limit. In this case we do not have any recourse. In fact, fixed multipliers without Turbo enabled do not even work on this board at all. The board always just defaults to running the chip at 3.2GHz if Turbo is disabled which is so plain-as-day majorly broken it is almost comical (@Sin0822 - This is something very specific that you can bring forward to Gigabyte as it may be quicker to fix than the other issue
For some more reading see the following thread on XS Asus P6T / Rampage II Extreme: no "CPU VR Current Override" ???
Eventually, in that case Asus added a fix for the TDP limit issue in BIOS 1104 and everything was resolved. No more throttling at all.
I really hope that this is useful information to bring forward. This should be an easy problem to fix, and for that matter even reproduce. Fire up latest CPU-Z then just grab a Version 1.0 GA-X79S-UP5-WIFI board, plop in an i7 3930K SROKY
, crank the multiplier to 44x, then boot Windows 7 (or Windows 8) and run 64-bit version of Prime95 Blend with all cores/threads used (in this case 12), then watch the board throttle. It will do it like clockwork. The other one is even easier. Disable all turbo and power management features, then set the multiplier to 44x and watch the CPU load up at measly 3.2GHz as though no adjustments to the clock speed were made.Edit:
I just finished testing UEFI BIOS F4b as well since that one was reported as "working" in this thread, and the results were exactly the same. Throttling all the way.Edited by dejanh - 10/31/12 at 11:01pm