Sweet Processor: Intel Core i7 3930K
by Mark W. Hibben
The New Workstation Standard
Reviewers and analysts continue to be somewhat ambivalent about Intel’s new X79 platform and its accompanying LGA 2011 processors. Many regard them as niche processors appealing only to a small number of PC enthusiasts and gamers. But Intel’s X79 announcement also targeted “power users” such as digital content creators who can always use more processing power. X79 is all about getting serious work done, and if you have serious computing work to do, Intel currently offers no better single processor solution than the Core i7 3930K. Computing professionals of all stripes, from digital video editors to scientists and engineers will find their productivity and workflow enhanced by computers based on the X79 platform, and using the Core i7 3930K, these platforms will be economical to own and operate due to modest initial cost, low power consumption and high reliability.
These first Sandy Bridge E (LGA 2011) processors are but the leading edge of a new line of high performance Xeon processors known as Sandy Bridge EP, to be called the Xeon E5 family. Many of these will feature 8 hyperthreaded cores. In fact, early die photos release by Intel of the Core i7 3960X clearly show the eight cores, with two of them grayed out.
Apparently, the 6 core 3960X is just an early version of the Xeon processor with two of the cores disabled, and this may be true of the 3930K as well, which also features 6 hyperthreaded cores. But the Xeon E5 series won’t be out ‘till later this year, so the Core i7 39XX processors are the best performing single processor options available from Intel for the time being. Most likely there will be a processor upgrade path for early adopters of X79 processors to even higher core-count processors (eight or more) as the Xeon E5 family becomes established and future Core i7 processors introduced. The Core i7 39XX processors only occupy a niche because they provide so much more capability than most people currently need. But what is leading edge today is tomorrow’s mainstream. Adding processor cores has been the processor industry’s main way of translating Moore’s law into tangible performance gains for about ten years now, and this trend can be expected to continue for some time. Thus we can expect six and eight core processors to become mainstream in the next year or so, and early adopters of the X79 platform are at least protecting themselves from obsolescence for the near future.
Does my emphasis on professional use of the X79 platform mean that I think gamers and PC enthusiasts should avoid X79? Not at all. But it may take a while for apps and games to grow into the processing power of the 3930K, so be patient if it seems that the 3930K doesn’t immediately provide the expected performance boost.
As I pointed out in X79 System Design, PC gaming performance, as measured in frames/sec (fps) capability is largely a function of the choice of graphics card for action-oriented games that make heavy use of DirectX. The processor platform itself plays only a minor supporting role in these games, so many games may not benefit much from the 3930K. This is illustrated in the following chart that compares Crysis 2 frame rate performance between two different computers, our X79 system using the 3930K processor and an older X38 system using a Core 2 Q9550 processor, both running Windows 7 Ultimate x64. The comparison used two different nVidia graphics cards, the GTX 580 and 470, both running identical drivers. The chart shows that performance correlated mostly with the choice of graphic card.
Forward Looking Features
The Core i7 3930K processor has a number of notable features including:
1) Intel’s most advanced multi-core architecture featuring six 64 bit processing cores sharing a huge 12 MB Level 3 instruction/data cache, which actually takes up the lion’s share of the processor real estate. Each core is hyperthreaded, meaning it can run two separate process threads simultaneously. From the standpoint of the OS, the processor has 12 cores, which you can see running in Task Manager in Windows.
2) Simplified chipset: like Sandy Bridge (Z68 et al.) X79 reduces the main components of the external chipset to a single Platform Controller Hub (PCH). The processor itself caries interfaces for memory and PCIE to be used for graphics cards and add-ons.
3) PCIE 3.0 support: The 3930K supports huge PCIE bandwidth with 40 lanes typically available as two x16 slots and an x8 slot and which support cards down to x1 in width. Most importantly, the processor supports the new PCIE 3.0 specification, which increases raw data throughput from 500 MB/s per lane for PCIE 2.0 to 800 MB/s for 3.0. As of this writing, the first graphics cards supporting version 3.0 are just making their appearance. Support for PCIE 3.0 is just one of the ways the 3930K provides protection against obsolescence.
4) High Memory Bandwidth: The four separate DDR 3 memory channels (the most available on any Intel processor) can control up to 8 DIMMs with stock clock rates up to 1600 MHz. Intel has built in support for Extreme Memory Profile (XMP) DIMMs with clock rates up to 2133 MHz. Our test system used XMP DIMMs running at 2133 MHz and we found the setup very easy. XMP really helped memory throughput when overclocking the processor.
5) Vector Processing: The ability to perform Single Instruction – Multiple Data (SIMD) processing has been a part of Intel processors for some time. While maintaining compatibility with previous SIMD standards such as Intel SSE 4.1 and 4.2, the 3930K implements Advanced Vector Extensions (AVX), as do other Sandy Bridge processors. AVX supports wider 256 bit vectors (each vector element is a 256 bit floating point number) for higher data throughput while maintaining high power efficiency.
6) Enhanced Intel SpeedStep Technology: EIST continuous varies operating frequency of the processor in order to conserve energy while maintain high performance and responsiveness. All active cores are clocked at the same frequency. EIST also varies core voltage with voltage being increased as clock frequency is increased. These voltage-frequency operating points are known as P-states. Frequency changes are software (OS) controlled, while the processor optimizes core voltage based on temperature, power dissipation and other factors. Transition latency is very low, so multiple P-state transitions per second are possible in the processor. For non-turbo mode operation (where all cores are active) clock multiplication of the base clock frequency varies between a minimum of 12 to a maximum of 32 in the 3930K, and these min and max values are locked.
7) Turbo Boost Technology: If not all the processor cores are active, Turbo Boost can allow all the remaining cores to be clocked at a higher multiple of the base clock than the maximum for the case when all cores are active. For the 3930K, the Turbo multiplier maximum is set to 38 by default. In stock operating mode, whether this maximum is reached is contingent on other factors including the number of active cores and the power consumption and temperature of the processor. If the processor hits a thermal design limit such as power dissipation, Turbo Boost logic automatically throttles back the multiplier. In overclocking, the maximum multiplier and the thermal design limits can be over-ridden by the user via BIOS settings.
8) Virtualization: The ability run multiple concurrent operating systems is probably of more interest in a server context, but I will note that Intel Virtualization Technology for Directed I/O (Intel VT-d) was disabled in the first batch of Sandy Bridge E processors due to a design error. The second stepping, now available, has corrected the error and VT-d is now enabled.
As discussed in X79 System Design, we selected the 3930K based on results posted by users of Performance Test 7 on Passmark’s web site. Passmark averages the results of many users in compiling their ranking of all Intel and AMD processors, and the 3960X and 3930K were number 1 and 2. We chose the 3930K based on cost effectiveness. In our own testing of the 3930K, we weren’t disappointed. In overclocking we easily exceeded the posted benchmark results. Furthermore, the 3930K is a sturdy little beast able to take far greater than the Intel spec TDP of 130 Watts, as we’ll show in the overclocking discussion in this article.
The Performance Test 7 results for the CPU Mark category are shown in the graph on the next panel. This is the most processor specific of all the tests in the PT7 test suite and covers many typical computer operations: tests integer and floating point arithmetic using 32 and 64 bit operations; performs vector (SIMD) processing using 128 bit Intel SSE instructions; executes a prime number finding algorithm; compresses data using an adaptive encoding algorithm; performs data encryption; executes a physics simulation using the Tokamak Physics Engine; sorts an array of 100,000 random strings. PT7 can run up to 2 threads in each core, so it tends to reflect the advantage of the 3930K’s 6 hyperthreaded cores. Of course, not all apps may be able to achieve equivalent utilization of the processor.
The stock configuration of the 3930K scored slightly lower in our testing at 13739 than the average posted Passmark result of 14514 but far higher than the next highest scoring processor, the Intel Core i7 995X at 10945. In terms of raw CPU performance, the 3960X and 3930K are really in a class by themselves. Along with the stock non-overclocked results, the graphs show results for the processor overclocked to 4.375 GHz and 4.625 GHz. The exact method of overclocking is discussed in the next section.
Finding the Overclocked Sweet Spot
The new LGA 2011 processors are ready for overclocking, and overclocking the 3930K even mildly will make up any performance disadvantage it has compared to its big brother, the 3960X, in stock configuration. As I discussed in more detail in X79 System Design, the useful clock adjustments that the user can make are the Fixed Clock Multiplier and the Turbo Mode Maximum Multiplier, as shown in the Logic Flow Diagram.
The downside of overclocking is often greatly increased power consumption, both at the processor and board level. Even mild overclocking can undo some of the energy saving features Intel has brought over from Sandy Bridge and turn the processor into a power hog. In the chart on the next panel I graph the normalized power consumption of the 3930K processor and the total system as a function of increasing only the Turbo Mode maximum multiplier, the typical overclocking approach. Power consumption is the peak power observed during the course of the PT7 CPU Mark test and is normalized to the power consumption under the same test conditions for the stock processor configuration.
The stock configuration power consumption was 113 Watts for the processor and 277 Watts for the complete system (wall plug power consumption). CPU power consumption reflects whatever Core voltage change was required to make the system stable for a given overclock frequency. System power consumption is a function of board voltage regulation efficiency and power supply efficiency, so may vary from system to system, but I regard these numbers as fairly representative. Also shown is the change in CPU mark score as a function of overclock frequency where the stock score was 13739.4.
From the standpoint of power consumption, overclocking seems like a lousy deal. For a 15% increase in performance, the processor consumes 30% more power at the Turbo Max frequency of 4.0 GHz. At the extreme overclock frequency of 4.6 GHz, the increase in processor power consumption approaches 50% and dominates the system power consumption increase as well. Given that the processor peak power is now over 170 Watts (40 more than the official Intel TDP of 130 Watts) and the processor peak temperature is 80 C (~7 degrees over the TDP of 72.6 C) one would have to question the wisdom of operating in this regime for an extended period. Is there an overclocking regime that actually makes sense?
Well, there is, and we stumbled into it in the course of exploring the clock options of the 3930K. I call it Hybrid Mode because it involves a combination of increasing the Fixed Multiplier to 1.25 as well as changing the maximum Turbo Mode multiplier. Increasing the fixed multiplier sets the maximum, non-Turbo mode, frequency to 4.0GHz. Then we varied the Turbo Mode maximum multiplier over a small range (34-37) and tabulated peak power consumption during the PT7 CPU Mark test as before. The results are shown on the next panel.
In this chart, all quantities are normalized to the stock (Turbo Max = 3.8 GHz) baseline as before. It’s not at all clear why this Hybrid Mode is so much more efficient from a power consumption standpoint than Turbo Mode for a given Turbo Maximum clock frequency. This mode doesn’t give up any performance gain, as the CPU Mark scores are approximately the same for a given frequency. The difference in efficiency may be an artifact of non-optimal EIST voltage selection in the Pure Turbo Mode.
I consider the 4.375 GHz operating frequency to be the Sweet Spot in the overclocked space. CPU performance has risen by 25% while CPU peak power consumption has risen by only 24% to 140 W, a modest 10 W over the Intel spec TDP. Peak processor temperature was well below the TDP of 72.6 C at 64 C, well within safe limits. Furthermore, operating at this frequency does not require over-riding the EIST Core voltage control in order to achieve stable operation, so this is set to Auto in the BIOS. Only the CPU Thermal Monitor was disabled, since it can be tripped by the somewhat over-spec power consumption.
A summary of the BIOS settings for the Sweet Spot and our other overclock settings is shown below:
|Turbo Maximum Clock Freq. (GHz)||Fixed Multiplier||Turbo Max Multiplier||Vcore||CPU Thermal Monitor|
|4.5 (Turbo Mode)||1.0||45||1.45||Disabled|
|4.5 (Hybrid Mode)||1.25||36||1.4||Disabled|
Setting the fixed multiplier to 1.25, setting the Turbo Mode maximum multiplier to 35 and setting the CPU Thermal Monitor to Disabled are the only settings changes required for the Sweet Spot. All overclock settings were tested for stability in extended game play, which is pretty much the acid test for app and system stability and we found no problems. However, be advised that the reader uses these settings at the reader’s risk, and the author accepts no responsibility for damage or other unintended consequences of using these settings (our usual disclaimer). Keep in mind that peak operating temperature of the processor is going to be a function of processor cooler type, so the reader may not exactly reproduce the results we got. For more details of the X79 system used for testing see my X79 System Design article.
The Sweet Spot makes the 3930K an even sweeter deal. In the Sweet Spot, you get 95% of the CPU Mark performance that you would by overclocking to 4.6 GHz, without the huge power drain and stability headaches.