Intel X79 System Design: Component Selection, Gaming Performance, and Overclocking
by Mark W. Hibben
Cost Effective Performance
In support of a planned series of reviews centered around Intel’s X79 high performance platform, we built a complete system rather than simply mounting a motherboard on a test fixture for a quick series of benchmarks. I felt this would provide a more realistic assessment of performance, especially in overclocked mode, where thermal management is critical. In this first Tech Chat of the new year, I’ll present an overview of the system design, including trades for processor, motherboard, and graphics card selections. We also delved into gaming frame rate performance for the X79 platform, and I’ll present results that shed light on the question of why there is little apparent performance improvement in Sandy Bridge E platforms. The details of overclocking the Intel LGA 2011 processors is an area many reviewers have glossed over. Intel’s documentation is sketchy at best, but I’ll try to demystify the subject and detail our approach.
When Intel announced the X79 chipset and LGA 2011 processor series back in November 2011, the flurry of reviews of X79 systems, many on the day of the announcement, all featured the top of the line Core i7 3960X Extreme Edition processor. Intel had pushed samples of the 3960X onto the reviewers who had accepted them rather than risk delaying their reviews. Intel may have thought they were putting their best foot forward with the 3960X. Both the 3960X and 3930K processors share the same basic Sandy Bridge E architecture described in my article Is X79 Really Worth It? Sandy Bridge E is an evolution from Sandy Bridge that provides more PCIE lanes (40 vs. Sandy Bridge’s 16) and more memory channels (4 vs. 2 for Sandy Bridge).
The 40 PCIE lanes are organized as two x16 slots and an x8 slot, and every motherboard I’ve seen offers at least this much PCIE expansion, as well as some PCIE slots derived from the Platform Controller Hub (PCH). The processor PCIE interfaces support PCIE 3.0, when devices become available, and X79 board manufacturers claim PCIE 3.0 compatibility.
The Sandy Bridge E processors use the new Land Grid Array (LGA) 2011 form factor to accommodate the extra PCIE lanes as well as the four independent DDR 3 memory channels that are rated by Intel at 1600 MHz. Both the 3960X and the 3930K are 6 core designs featuring Intel Hyperthreading for a total of 12 virtual cores (as far as the OS in concerned). Both processors lack the on board graphics processor of the Sandy Bridge series, on the reasonable assumption that the high performance applications for which the processors are intended will require an external graphics card. Other Intel technologies shared by the processors include Intel 64 Architecture, Enhanced Intel Speed Step (EIST), Turbo Boost, and Advanced Vector Extensions (AVX).
Enhanced Speed Step dynamically varies core clock rate and core voltage to minimize power consumption. At idle, the processor clock rate will be considerably lower than the rated clock frequency. For the 3930K, that minimum clock rate in stock configuration is a mere 1.2 GHz. Turbo Boost provides for operation of one or more cores at higher than the processor rated clock frequency for short periods of time. Generally, when LGA 2011 processors have been “overclocked”, it is the maximum Turbo Boost frequency that is increased. In the table on the next panel, I compare the specs of the 3930K and 3960X as well as a Sandy Bridge Core i7 2700K.
|Processor||Core i7 - 2700 K (Sandy Bridge)||Core i7 - 3930K (Sandy Bridge - E)||Core i7 - 3960X (Sandy Bridge - E)|
|Launch Date||Q2 2011||Q4 2011||Q4 2011|
|Retail Price in US$ on Newegg.com as of 1/26/12||369.00||599.00||1049.00|
|Passmark CPU Score (Stock Configuration)||10312||14464||14846|
|Number of Cores||4||6||6|
|Number of threads||8||12||12|
|Max Proc. Clock Speed in GHz||3.5||3.2||3.3|
|Max Turbo Clock Speed in GHz||3.9||3.8||3.9|
|Support Chip Interface||DMI||DMI||DMI|
|Direct Media Interface (DMI) Speed in GB/s each direction||2||2||2|
|Smart Cache Size in MB||8||12||15|
|Memory Type||DDR3 1333||DDR3 1600||DDR3 1600|
|Processor PCI Express 2.0 Lanes||16||40||40|
|Fabrication Process Type||32 nm||32 nm||32 nm|
|Processor Max Thermal Design Point in Watts||95||130||130|
|Processor Graphics||Intel HD 3000||NA||NA|
According to the specs, the differences between the 3960X and the 3930K are very slight: the 3960X has a slightly larger lowest level (Smart) cache, and a slightly higher clock rate. The 3960X also scores slightly higher (in stock, non-overclocked configuration) on Passmark Software’s Performance Test 7 CPU benchmark, averaged over the results of many PT7 users. One of my first stops in researching components for a PC build is Passmark’s web site. Not only do they engineer a great PC benchmarking utility, (available in 32 and 64 bit versions for Windows), but the software allows users to upload test results to the Passmark site, which are tabulated and displayed as average results. This is a great boon to the PC builder, since you can view the results of literally thousands of tests of a given processor. I regard this as far more reliable than an individual review, since it averages out the effect of various user motherboard configurations. And no, I have no financial interest in Passmark, nor do I receive any promotional consideration from them.
As I’ve stated in the past, the advantage for the reader of providing Performance Test 7 results is that you can compare these not only to results tabulated on the Passmark site, but to your own system by running PT7 at home. This is the best way for the reader to decide if an upgrade or new system is worthwhile. Passmark offers PT7 as a free trial for 30 days, so I encourage everyone to download it and give it a try.
The 3930K actually scored higher (on average) than the 3960X when overclocked, although the average reported overclocked rate for both processors was a modest 4.1 GHz. Most early reviewers of the 3960X were able to achieve stable operation at 4.7-4.8 GHz, and we were able to achieve stable operation of the 3930K at 4.6 GHz for a CPU Mark score of 18,101, far higher than the Passmark overclock average of 16,662.
The Passmark results made the choice of the 3930K an easy one. For roughly 60% of the money, you get roughly 97% of the performance (at stock clock rate), and if you have an extra US$ 500 burning a hole in your pocket, we’ll show you better ways to spend it to enhance the performance of your new X79 system than buying the 3960X. You can achieve much more tangible performance benefits by investing in a RAID system, solid state drive, or an extra graphics card for SLI. For the purposes of our planned reviews, we invested in both RAID and two way SLI. We’ll go into much more performance detail in our 3930K CPU review.
More than a year ago we reviewed the Gigabyte X58-UD7 motherboard and were impressed with the quality of the board, its sturdiness, and the smooth uneventful system build with the board. Since then it has been completely trouble free. We had such a good experience with the board that we naturally leaned towards an X79 Gigabyte board. Once again, let me just state that we have no financial interest in Gigabyte and receive no promotional consideration from them. However, the good experience we had with the previous board was not actually decisive in the choice of Gigabyte this time around. The X79 architecture doesn’t allow for much differentiation between motherboards.
All X79 boards have at least 2 PCIE x16 slots and 1 PCIE x8 slot, since these interface directly to the LGA 2011 processor. All X79 boards offer at least 2 SATA 3.0 ports, since these interface to the X79 Platform Controller Hub. So the main discriminators became the number of memory slots available and the availability of additional SATA 3.0 ports controlled by an external controller chip. The LGA 2011 processor can interface to as many as eight memory modules, but many manufacturers provide only 4 slots to save space on the motherboard. Since we like being able to upgrade memory without having to throw away existing modules, we ruled out any boards that offered only four slots, even though our initial build used only four modules. Motherboards that offer additional SATA 3.0 ports usually use one of the Marvell SATA 3.0 controllers such as the 88SE9172, which has 2 SATA 3.0 ports and is configurable as RAID 0 or RAID 1. Our experience with the Marvell 88SE9128 controller on the Gigabyte X58-UD7 system has been uniformly positive. Combined with the Western Digital Velociraptor 10000 RPM drives, the Marvell RAID system has been completely trouble free for over a year. Using RAID 0, (also called striping) where data is divided between the two drives to provide faster sustained read and write speeds, we obtained speeds comparable to single solid state drives, but at lower cost and with enormous 1.2 Terabyte capacity. We now take SATA 3.0 RAID for granted, and consider additional RAID controllers on the motherboard a big plus. The Gigabyte X79-UD5 comes equipped with not one, not two, but three Marvell 9172 controllers, providing both external and internal SATA 3.0 connections.
The table on the next panel lists just some of the many good X79 motherboards available today. Selection of a motherboard is largely a matter of personal preference in terms of motherboard features, PCIE expansion, and intended use. For our purposes, 4 way SLI/Crossfire capability was not important, since our intended use for the system, high definition video capture and editing, doesn’t require multi-GPU capability beyond possibly 2 way SLI, which we wanted to test for this series of reviews anyway. The basic X79 architecture provides all the PCIE expansion we needed for this system without resorting to PCIE bridge chips, so we stayed away from the pricey Asus Rampage IV Extreme.
|Model||Manufacturer||Memory Slots||Extra SATA 3.0 Controlllers||Form Factor||Other Noteworthy Features|
|P9X79 WS||Asus||8||1 Marvell 9128||SSI CEB 12” x 10.5”||Capable of 4 way SLI/Crossfire|
|X79 Extreme 9||ASRock||8||1 Marvell 9220, 2 Marvell 9172||ATX||No on board audio, SB PCIEx1 card bundled|
|Rampage IV Extreme||Asus||8||1 AsMedia with 4 SATA 3.0 ports||E-ATX||Capable of 4 way SLI/Crossfire|
|GA-X79-UD3||Gigabyte||4||3 Marvell 9172 (6 SATA 3.0 ports)||ATX|
|GA-X79-UD5||Gigabyte||8||3 Marvell 9172 (6 SATA 3.0 ports)||E-ATX||Dual UEFI BIOS|
Of the motherboards that were surveyed for this system, only the Gigabyte X79-UD5 offered eight memory slots and the three Marvell controllers without giving up some other functionality such as on-motherboard audio. The Asus P9X79 WS would be my second choice, as I’ve also had good experience with previous Asus WS boards, but having only one Marvell 9128 controller kept it from being the top choice. The price to be paid for all the functionality of the Gigabyte X79-UD5 is that the board ends up being larger than the standard ATX form factor. This is often the case with X79 boards with many being Extended ATX (as in the case of the Gigabyte X79-UD5) with dimensions of 30.5 x 26.4 cm. The board mounts on the same hole pattern as ATX, has the same back panel width, but is a little deeper. The E-ATX form factor was not a deterrent in our case, since we had a computer case available with more than ample room, but the reader should take this into consideration when compiling your shopping list for your X79 system.
Combined with our overall good experience with the previous Gigabyte board, the features of the X79-UD5 also made it an easy choice. In addition to what I listed in the table, the X79-UD5 also comes with a number of features the reader may find valuable, including one of the first “production” UEFI BIOS systems, that provides for a mouse driven user interface, and a switchable dual BIOS, which is effectively two separate independent BIOS modules that can be individually configured, not only in terms of BIOS settings, but also the version of the BIOS itself, since each BIOS can be flashed separately. Also bundled with the motherboard was a PCIE x1 combined 802.11n Wifi and Bluetooth card, which actually turned out to work quite well. We’ll go into the X79-UD5 features in much more detail in our motherboard review.
Graphics Card Selection
To select the graphics adapter, we once again were heavily influenced by the GPU test results Passmark publishes on their web site. PT7 also performs a battery of graphics card tests, and the 3D graphics test result is what’s tabulated on their site. The site has results for just about every nVidia and AMD (ATI) card currently available. At the top of the list is the nVidia GeForce GTX 580, even beating out its big brother, the GTX 590, a dual GPU, single card configuration.
Since we also had excellent experience with our previous X58 system using a GTX 470, this became a no-brainer. The only moderately difficult choice was deciding between the many good manufacturers of nVidia based cards. Letting past experience be our guide, we chose EVGA, since they also made the GTX 470 we had used.
As part of our review, we tested frame rate performance of our X79 system with the GTX 580 graphics card for four different games, Crysis 2, Call of Duty: Modern Warfare 3, Elder Scrolls V: Skyrim, and Batman: Arkham City. But we questioned whether the gaming results should be part of the motherboard review, or part of the separate graphics card review. This has been a nagging issue since the first X79 reviews came out, because virtually all of the game-based testing showed negligible difference in frame rate performance for X79 boards compared to previous generation Z68 boards.
This came as a disappointment to some reviewers, but it probably should not have. The simple fact is that for modern action-oriented games that rely on DX11, the processor platform makes almost no difference. To illustrate this, we performed a series of benchmarks using Crysis 2 at 1920x1080 screen pixel resolution, all graphics quality settings to maximum, and the DX11 and Tesselation packs installed. We mixed and matched between our X79 system in stock, non-overclocked, configuration, an older X38 system, and the GTX 580 and GTX 470 cards. The GTX 470 was a good choice for this comparison because it has Physx and DX11 capability just like the 580, and uses the same nVidia Win7 x64 driver (Version 285.62), but is not quite as fast as the 580. Literally, all we did was plug cards in and out between the systems. The X38 system uses the Asus P5E WS motherboard, a Core 2 Quad 9550, 4 core, non-hyperthreaded processor running at 3.0 GHz. The board has 4 GB of DDR 2 1066 memory. Both systems run Windows 7 Ultimate 64 bit editions.
Clearly, there is some interaction between the motherboard and graphics card that affects performance, but we think these results demonstrate that gaming performance mainly reflects on the choice of graphics card rather than the motherboard. We realize that most reviewers are accustomed to including gaming benchmarks in motherboard reviews, and strictly speaking, there’s nothing wrong with this, as long as all the test conditions, especially choice of graphics cards, are specified. However, the potential for confusion about what the test results really mean is very high. Therefore, we think it’s more appropriate to include our gaming benchmark results, both single card and dual card SLI, in our GTX 580 review, covering its performance and features.
On the next panel is a list of the major components of our X79 system.
Major Components List
|Component||Design Parameter(s)||Quantitative Rating|
|Windows 7 Operating System||Version, Build #||Ultimate x64 with SP1, 7601|
|Gigabyte GA-X79-UD5 Motherboard||Memory Bus Type, Max Supported Clock Rate||DDR3 2133 MHz|
|American Megatrends BIOS||Version No., Date||F7, 12/26/2011|
|Intel Core i7 3930K||Clock Multiplier Range, Internal Clock Range, Core Count, LL Cache||12-32, 1200-3200 MHz, 6 Cores (12 threads), 12 MB|
|Thermaltake FrioOCK Processor Cooler||Cooler Type, Number of Fans, Maximum Power Dissipation||Heat pipe linked radiator, 2 fans, 240 W|
|G.Skill DDR3 2133 MHz XMP Memory||Number x Size||4 x 4GB|
|EVGA NVidia GeForce GTX 580 (x2) P/N 015-P3-1580-AR||CUDA Cores, Graphics Card Memory, Driver Version||512, 1536 MB GDDR 5, 285.62|
|Western Digital VelociRaptor WD6000HLHX (x2)||Capacity, Spindle Speed, SATA Speed||600 GB, 10000 RPM, 6 Gb/s|
|Thermaltake Toughpower Grand Power Supply||Maximum Combined Power, Typical Energy Efficiency||1200 Watts, 87-93 %|
|ABS Black Pearl Aluminum Tower Case||Motherboard Form Factors, Internal 5.25”, 3.5” drive bays||ATX and E-ATX, 7, 8|
As I stated previously, the Enhanced Intel Speed Step (EIST) logic continuously varies the clock frequency of a given core to minimize power consumption, going as low as 1.2GHz for the 3930K in stock configuration. Under the circumstances, the reader may wonder what exactly overclocking means. To add to the potential confusion, Intel allows for three separate ways for the user to change the default clock settings of the CPU. The clock frequency that is applied to a given processor core is the product of three terms:
fcore = fbase * Mfixed * h(t) where
fbase = the internally generated base clock, which defaults to 100 MHz for the stock configuration but which can be varied from 80-133.33 MHz in 0.01 MHz increments by the user in the BIOS.
Mfixed = a small constant multiplier, which defaults to 1 for the stock configuration, but which can be changed to 1.25, 1.66, or 2.5 by the user in the BIOS.
h(t) = a large clock multiplier dynamically selected by the EIST logic and which can vary between 12-32 for non-Turbo mode operation for the Core i7 3930K processor. This means that 32 is the largest clock multiplier EIST will apply to all processor cores simultaneously in the 3930K, and this value is locked and not changeable by the user. For Turbo mode, the maximum clock multiplier can be larger than 32. For the stock 3930K configuration, this multiplier can be as high as 38, but will be applied only to some of the processor cores, and for a limited time. To illustrate the relationships between the various clock multipliers, I’ve prepared a logic flow diagram for the processor clock, shown next panel.
This diagram is intended to show logical relationships and not actual signal flows within the processor, since Intel hasn’t published sufficient information for a true signal flow diagram. The clock related knobs that the user can turn are highlighted in blue. Intel has pointed to the ability to change the base clock frequency as an indicator that Intel is serious about supporting overclocking. But this particular adjustment is difficult to use successfully. Our experience was that this was a sure fire way to induce BIOS startup failure. This probably has to do with the fact that the PCIE and memory controllers are timed off the Base Clock. The fixed multiplier, sometimes referred to as the clock gear ratio, can be used successfully, but we found it impractical to go above 1.25, since the next multiplier of 1.66 leads to a non-Turbo maximum clock rate of 5.312 GHz, because there’s no way to turn down the maximum clock multiplier of 32. Increasing the Turbo mode maximum clock multiplier has been the preferred way to “overclock” the processor. It’s also the simplest, since there’s only one clock setting to change: to overclock to 4.6 GHz you just increase the Turbo mode maximum multiplier to 46 and leave the other clock settings unchanged. We found that we were able to achieve the highest stable overclock frequency on the 3930K through a combination of increasing the fixed multiplier and adjusting the Turbo mode max multiplier. By “stable” I mean that we could run all benchmarks including the gaming benchmarks on our finished system without causing system or application crashes over a period of about two weeks of testing.
Although I don’t show it in the diagram, the memory clock is also derived from the Base Clock Frequency, with the default multiplier, as shown in the BIOS, of 16. Increasing the Fixed Multiplier automatically “overclocks” the memory modules, although selecting the memory XMP must be done by the user in the BIOS.
The table below shows the frequency and other BIOS settings for the stock and overclocked configurations for our X79 system. Although the overclock settings worked reliably in the Technomicon X79 system, you are cautioned to apply these setting at your own risk. Technomicon accepts no responsibility for any damage or other unintended consequences of using the overclock settings shown below.
|Parameter||Stock Configuration||Overclocked Configuration|
|Base Clock Frequency||100 MHz||100 MHz|
|Minimum Clock Frequency||1.2 GHz||1.5 GHz|
|Non-Turbo Maximum Frequency||3.2 GHz||4.0 GHz|
|Turbo Maximum Multiplier||38||37|
|Turbo Maximum Frequency||3.8 GHz||4.625 GHz|
|Memory Clock Multiplier||16||16|
|Memory Clock Frequency||1.6 GHz||2.0 GHz|
|Memory XMP||Disabled||Profile 1|
|Vcore||Auto (1.26 V)||1.45 V|
|Turbo Power Limit||Auto (130 W)||Auto (130 W)|
|Core Current Limit||Auto (135 Amps)||Auto (135 Amps)|
|CPU Thermal Monitor||Auto||Disabled|
In our upcoming series of hardware reviews covering the Intel 3930K CPU, the Gigabyte X79-UD5, and the EVGA GTX580 graphics card, I go into the performance details, user features, and issues for each of these key components as they performed in our X79 system.