Intel Core i5 750 Review - Overclocking Phase-Change vs LN2

CPU by massman @ 2009-09-13

Intel´s latest CPU is put to the test against their own Core 2/Quad and Core i7 series, as well as AMD´s Phenom II. We wrap up the testing with an overclocking session using phase-change and LN2 Cooling. Does the Core i5 and S1156 i7 impress? Let us find out

  • prev
  • next

Intel\'s Turbo Mode Investigated

Intel's Turbo Boost in short

Turbo boost technology increases the core frequency whenever possible without breaking the TDP limitations set by Intel. Intel has limited the maximum TDP, or power consumption, to 95W divided over all four cores. So, each core can in theory produce a maximum of 23.75W; that is if we forget about the integrated memory controller other peripherals integrated on the uncore. Each processor is set at a certain stock frequency at which it will not exceed this TDP limitation, even when all four cores are working at full speed.

The problem here is that situations in which all four cores are working at full power at the same time are very rare; in other words, the processor will almost never use all the power it has available. Next to that, there's the fact that any CPU has room to be run at frequencies beyond those set at the Intel labs, as you can see from the many overclocking results. In other words: a lot of room to increase frequency and a lot of room to consume more power. So, what Intel came up with is a dynamic overclocking feature which they call Turbo Mode. It increases the clock frequency of one or two cores if the maximum TDP hasn't been reached in order to make applications that don't fully use the multi-core technology faster anyway.

We used our retail Core i5 750 to check out the performance gain from enabling turbo mode.

  • No turbo = 20x
  • 4 cores = 21x
  • 2 cores = 23x
  • 1 core = 24x

    We used the same batch of tests we used for the normal performance measurements.

    Madshrimps (c)

    Madshrimps (c)

    In SuperPi, turbo mode actually makes the Core i5 750 faster than the Core i7 965, whereas it was behind without the turbo mode. In Wprime, it doesn't, but then again ... turbo mode doesn't add cores *wink*.

    Madshrimps (c)

    Madshrimps (c)

    In initial test sessions, it seemed that Turbo mode didn't affect the memory bandwidth at all, but after re-running the tests properly we now see the real performance boost from Turbo mode.

    Madshrimps (c)

    In Cinebench, the extra core frequency is definitly noticeable! Especially the increase in single tread benchmark mode is significant.

    Madshrimps (c)

    Madshrimps (c)

    The PCMark05 benchmark is something different as the system score, which is used in overclocking environments, decreases quite a lot by enabling Turbo mode. Reason of that decrease seems to be the weaker scores in the graphics and hdd subtests, but why they are slower is an unanswered question at this moment. We suspect it has something to do with the way Windows manages the C-states (power saving); as C-states have to be enabled for Turbo mode to work, it may be in conflict with Windows settings. For more info about this oddity check our upcoming motherboard review.

    Madshrimps (c)

    In Entry mode, where the CPU frequency plays a big role, the Turbo mode is indeed faster, but more interesting are the performance and high preset results. Similar to what we saw in the PCMark05 benchmark, the more important the graphics card is, the more likely that Turbo Mode becomes slower.

    The question is if this issue is also present in extreme overclocking situations where, quite often, people use these turbo mode multipliers for higher clock frequencies. Something Madshrimps will definitely look into.

    Madshrimps (c)

    Turbo mode gives quite a nice improvement.

    Madshrimps (c)

    Although not spectacular, with Turbo mode enabled the Core i5 gets closer and closer to the Core i7.
    • prev
    • next
    Comment from Massman @ 2009/09/13
    As said: a lot more to digg into. Expect more overclocking action, more motherboards, more turbo mode testing, more memory testing ...

    Much more
    Comment from Diegis0n @ 2009/09/14
    Great work PJ!

    It makes me count my money and try to get a i5 platform

    UD3 + 750: 277€, surprisingly affordable since i can sell C2 combo out for ~200.

    I'll be ready for some action in a few months btw, lots of hardware will become "available"
    Comment from Kougar @ 2009/09/14
    Nice article... Looks like Turbo Mode will make subzero OC'ing rather interesting...

    just think about the secrecy with which AMD developed its new HD58xx series: so close to the launch and publicly no one has a real clue about its performance.
    Spoke too soon. Still, is a very valid point. The entire NDA situation with Lynnfield is not just absurd, it is plain stupid.
    Comment from thorgal @ 2009/09/15
    Seems you woke up someone at Anandtech LOL. One day after your article they present a clock comparison themselves
    Comment from Diegis0n @ 2009/09/15
    I well prefer PJ's writing skills; at least there's some nice pictures to look at
    Comment from bartx @ 2009/09/16
    Nice work. I like reviews from [M]