Enhancing Turbo Boost
The Enhanced Turbo Boost
What Intel did was to bring its Turbo Boost idea further along. In Bloomfield, one typically can only ramp up the clock speed by one multiplier and at most you'll find it going up by two grades if it's a lightly or singly threaded application. These values for the Turbo Mode multiplier were fixed and one has to get the Extreme Edition Core i7 models, which are unlocked, to set more aggressive multipliers.
The same 'lock' applies for our very 'non-extreme' Lynnfield processors, as we were unable to change the Turbo Boost ratios in the BIOS. What has changed however is the extent of the default multipliers. For instance, the Core i5-750 is clocked at 2.66GHz, rising up to 3.2GHz when only a few cores were taxed, which is an increase of 4 multipliers (every 133MHz is a multiplier given the base clock of 133MHz).
This is quite the jump, especially for older, single threaded applications which hardly benefit from having more processor cores. It is also perhaps, an implicit admission from Intel that the pace of multi-core development has not been reflected in the software arena. Many applications remain constricted to dual-cores, much less than the HyperThreaded future envisioned by Intel.
Hence, this move to bolster Turbo Boost, not only bodes well in the performance stakes, but also for the sake of power efficiency. Obviously, having your processor vary its clock speed aggressively according to the load is the path to optimal power efficiency, since it's more likely to spend the idle moments at lower clock speeds and voltages.
In fact, this whole dynamic clock scaling Turbo Boost feature is limited by the thermal envelope of the processor. For these Lynnfield processors, it is 95 watts and whether one core or four cores are being utilized, the processors will not exceed the TDP threshold. The Lynnfield's new Power Control Unit (PCU) is responsible for tackling this aspect and we've detailed about the PCU's operation during our intimate discussions with Intel back in our Computex trip.
Turbo Boost in Action
We decided to explore how Turbo Boost worked by running various applications and checking the clock speed. The first thing we noticed was that even when the system appeared to be idling, the clock speeds can fluctuate very quickly in a short time span. For the Core i7-870, the idling multiplier seemed to be between 13 and 14x (1.7 ~ 1.8GHz), but it can also hover around the 10x multiplier level if there aren't any background tasks operating as well as your settings utilized in the BIOS to define the processor's operational levels.
To test if Turbo Boost worked as advertised, we first ran AquaMark3, an aging benchmark that's mostly single threaded. We found that our Core i7-870 scaled up to 3.6GHz, its rated maximum in this application. Next, we tried Cinebench 10 with all the cores active. At this full load, the maximum was only around 3.06GHz, (133 x 23), with slight upward fluctuations from time to time. It looked like the Core i7-870 was hitting its TDP limits here in Cinebench and a glance at our power meter confirmed the much higher power draw.
In a game that is supposed to be multi-threaded, like Crysis, we found a similar behavior, with the CPU topping out at 3.06GHz, with occasional moments when it went up to 25x and higher. Testing other applications yielded similar results, with multi-threaded ones hitting a limit at the 23x to 24x multiplier mark.
Looking at our results (which will be shown in our benchmarks), we can roughly generalize that having Turbo Boost enabled will always mean higher performance, though the amount of scaling and performance gain varied from application to application. For sure, we expect this feature to be very useful for power efficiency and likely a key reason to get Lynnfield over Bloomfield.