|Introduction and The Setup|
|It has been a while since I have personally setup a test bench, rolled up my sleeves and built a test system with top-shelf components, a fresh OS install and some overclocking mojo. However, Intel's recent release of Ivy Bridge-E gave us a hankering to do just that. We've been testing a lot of tablets, smartphones and ultrabooks, but there's a certain satisfaction that comes with building a high performance system from the ground up, that you just don't get from pre-built devices. So I decided to see if I still had the chops.
EVGA recently took the wraps off a new high-end X79 motherboard dubbed the X79 Dark and its black PCB with red accents, as it turned out, matched quite nicely with Corsair's Vengeance Pro series memory and H90 water cooler. But we needed stable clean power too, so we enlisted an EVGA SuperNOVA 1000 G2 power supply to offer up the juice. All told, with an OCZ Vertex 3 Max IOPs SSD for storage, we ended up with quite a test platform to explore the out limits of clock speed with Intel's Core i7-4960X Ivy Bridge-E six-core processor.
Pretty, isn't it? Beauty is in the eye of the beholder, as they say; and this is the computer geek's version of Elisha Cuthbert or Candice Swanepoel. Okay, maybe not quite Candice...
A HotHardware Test Bench - Well endowed for blinding speed.
Intel Core i7-4960X gently nestled in its LGA 2011 socket and ready to sweat.
On the following pages, we aim to give you a glimpse of the upper-end potential for overclocking with Intel's Ivy Bridge-E processor architecture. We'll be running with a $999 Core i7-4960X that Turbos up to 4GHz but you could just as easily get by with the much more affordable Core i7-4930K as well. For our purposes, we wanted to see the fastest clock speeds we could muster from Intel's current flagship desktop chip, but that doesn't mean you couldn't coax the same performance from its more affordable sibling. And as always, your results may and probably will vary. Onward then. Let's take a closer look at the gear and our setup.
|Hardware Configuration: EVGA, Corsair and More|
|The cornerstone of our overclocking efforts is EVGA's X79 Dark motherboard. Simply named, this Intel X79 chipset based board has all the right features in all the right places. The board has a five full-length x16 PCIe slots and one x4 slot, all of which can be arranged in 1x16, 2x16 , 3x8 or 4x8 configurations with the X4 slot still available. With six fan headers, support for quad-channel memory up to 2400MHz and beyond, solid state caps, dual 8-pin CPU power, on-board power, reset and clear CMOS buttons and even a triple BIOS switch for three separate profiles, the X79 Dark is a full-featured enthusiast's overclocking hot rod.
EVGA X79 Dark - Intel X79 Chipset Motherboard - An Overclocking Powerhouse
The EVGA X79 Dark also sports 4-Way SLI capability, six SATA 6G ports, four SATA 3G, six USB 3.0, ten USB 2.0 ports, two Intel Gigabit NICs and a 12 layer PCB. Also, if you look closely, the X79 Dark has onboard Bluetooth situated on the top of one of its USB IO blocks. As you can see, connector placement is excellent as well, with right angle USB and various power connectors right at the edge of the PCB, for easy, clean cable management. One word of caution is that this is an EATX board that is an inch wider overall, so be sure to plan for that if you have tighter case mechanicals to work with. The board is really sharp though with blacked-out accents trimmed in red. The X79 Dark also has a nice GUI UEFI BIOS. On that note, we'll dig into the BIOS a bit later but we went through a couple revisions in this review and EVGA is still getting things tacked down a bit it seems.
For our power requirements, EVGA also sent over their new SuperNOVA 1000 G2 power supply. The 1000 G2 is, you guessed it, rated for 1000 Watts continuous total power and is 80 PLUS Gold certified with up to 90% efficiency under load. It also comes with a 10 year warranty and has a fully modular design.
Built with a heavy gauge aluminum frame, the SuperNOVA 1000 G2 is based on a Super Flower Leadex design with an extremely quiet dual ball bearing fan. The modular cables are reasonably long as well, especially the PCIe power cables, though the ATX power cable could be longer maybe. The SN 1000 G2 is built like a tank and had zero issues maintaining solid voltage regulation and ripple control under load, when we were pushing envelope on our CPU. We expect to stress it a bit more in the months ahead with multi-GPU setups as well but don't expect it to have any issues there. In fact, the SuperNOVA 1000 G2 will likely serve as a test bench PSU for many months or years to come possibly here.
Next, we have a couple of offerings from Corsair to complete the build, with their Vengeance Pro Series 32GB quad-channel kit and H90 self-contained liquid cooler.
Corsair's 32GB version of this kit is currently hard to find in a 2666MHz rated flavor, though 2133 and 16GB kits are fairly plentiful. In fact, Corsair currently lists only up to 2400MHz (CAS 10) kits on their site at this time. That said, Corsair rates these bad boys for 2666MHz at 11,13,13,35 timings. We ran them at the 2400MHz CAS10 settings for purposes of our high speed overclocked testing. Regardless, the high speed bin here affords maximum flexibility when overclocking. We also found the modules played well with a couple of other test systems that are currently in house, so compatibility shouldn't be an issue, so long as you tweak timings accordingly, depending on clock speed.
And finally we have our all-important cooling solution, which came by way of Corsair's H90 liquid cooler. The H90 based on a larger fan and radiator design with a 140mm footprint to provide a bit more capacity and headroom under load, versus traditional single fan/radiator self-contained setups.
The H90 uses an Asetek pump and bracket and with motherboard control of its pump and fan speed via a 3-pin (CPU block) and 4-pin (fan) connectors. This setup worked well for us and in our case, we set the unit up for maximum power settings on both pin headers. Incidentally, Corsair recommends setting up the H90 with its fan as an intake on the back of your case, though our setup was on an open air bench, so it didn't matter for our purposes.
The nice thing about EVGA's X79 Dark motherboard is that there is a fair amount of open real estate around the CPU socket, which also allows memory heat spreaders and voltage regulator heatsinks to get plenty of air flow, in addition to easier navigation with the CPU water block and tubing. Another thing to note is the right angle ATX power connector orientation on the board. I personally really like it as I find it allows for passback of the cable to the back side of the motherboard tray, in cases that support that sort of setup. Some users might find this to be a major pain though, depending on your preference.
Though it was totally unplanned, as you can see, we have a decidedly well-coordinated black and red theme going on here with EVGA and Corsair components in our setup. As you might surmise, however, good looks only equate to beauty and not necessarily brawn. So let's fire it up.
|The Setup, Vital Signs and Tactics|
|EVGA's X79 Dark gives you pretty much any tool you'd want at your disposal for overclocking. The primary home screen greets you with a clean GUI that gives a graphical display of RAM installed, PCIe lanes used, CPU and Memory voltages, as well as VRM and CPU temps.
In general, Ivy Bridge-E overclocking is a fairly straight-forward endeavor. You can either increase the CPU's multiplier or the base clock or "BCLK" frequency. If you spend any time at all overclocking with Ivy Bridge-E chips, you quickly find out that, while adjusting the BCLK for a few MHz here and there is possible, it also contributes wildly to instability. So it's easiest to just stick with multiplier and voltage adjustments until you reach your peak, stable clock speed. EVGA does provide some quick settings for BCLK speeds of 125MHz, 133MHz and beyond, but for us, again, playing with these options ended up as an exercise in futility. We tried dozens of combinations but ultimately found the most head room at 1.48 to 1.5V on the VCore setting and then driving CPU multiplier settings up from there. This also limits you somewhat, however, in memory speed, since the memory interface clock is a derivative of the BLCK as well. Ultimately, we found our memory speed topped out at 2400MHz (driven from a 100MHz BCLK) with a 1.65 - 1.7V DIMM Voltage setting.
One thing we should note is that we performed our tests with EVGA's 2.04 revision BIOS for the X79 Dark. Though this BIOS version reportedly has issues with Intel Turbo Boost settings, at the time of testing, it was the most stable BIOS for overclocking. EVGA has since released version 2.07 which fixes some of these issues and we can confirm it contributes even more to stability while overclocking.
Our goal here was to show what a typical end user can expect to achieve overclocking Ivy Bridge-E with off-the-shelf components and less than extreme cooling. There's nothing exotic about our setup for a reason. With a little finesse and a bit of patience, our results are very achievable within say 100MHz either way.
Core i7-4960X Stock With 1866MHz DDR3 Memory
Core i7-4960X Overclocked To 4.6GHz With 2400MHz DDR3 Memory
As it turned out, our maximum, relatively stable clock speed ended up being 4.6GHz for the CPU with 2400MHz on the memory. To achieve this we had to take core voltage on the CPU up to 1.47V minimum though you might need a little more, depending on your cooling solution and particular CPU. Corsair's Vengeance memory was stable at default voltage for 2400MHz CAS 10 but we did dial it up slightly to 1.7V just for good measure.
Our first test was to power through a run of Cinebench R11.5. This proved to be no issue for the chip and in fact all of our benchmark runs were perfectly stable at this clock speed and voltage. However, further stress testing with Prime95 yielded more interesting and different results.
With Prime95 we had to bump core voltage up to 1.485V and in this shot as you can see it's sagging just a bit. With EVGA's 2.07 BIOS the motherboard and CPU combination didn't blue screen, but though the CPU was only hitting 72ºC or so, we did see what looked like throttling or stalls (utilization dropping to zero and worker thread stalls) on an individual core level. That said, Intel specifications are that the Ivy Bridge-E will throttle at 90ºC, so we weren't seeing throttling in the traditional sense. This was after fairly heavy stress testing and build-up but it did indicate that we were scaling beyond the thermal capacity of our setup. As it turned out, it wasn't Corsair's H90 cooler that needed help, but rather the motherboard power FET array. Interestingly, the motherboard VREG monitor wasn't showing crazy high temps either but once we strapped a small fan directly over the array, the CPU no longer showed these random core stalls and we were able to run the Prime95 torture test with max power draw and temperature settings for extended periods. We would speculate that power delivery was being affected somehow as temperatures scaled.
As it turns out, keeping the circuit area around the CPU cool can be just as important with self-contained water coolers. You don't get the additional air flow around the socket area that you typically do with standard fansinks; at least we didn't with out open air setup. In a well-ventilated case with intake and exhaust fans, this may be less of an issue.
One other small side oddity of EVGA's BIOS with the X79 Dark is that the PCI Express lanes are setup default to PCIe Gen 2.0 specifications. As we found out, overclocking seemed to be hampered quite a bit when plugging in a Gen 3 graphics card like our GeForce GTX 770, if this setting was left at Gen 2. Once we toggled the slot settings to Gen 3, all was right in the world again.
Regardless, this is about what we considered as the upper limit of our Ivy Bridge-E setup with mainstream, off the shelf cooling. Let's take a look at the numbers and see what upside performance can be had at various settings as well. We're done with the pain. Let's look at the gain.
|Overclocked Ivy Bridge-E Benchmarks: SANDRA, SunSpider|
SANDRA TestingAs a quick sanity check we took at some SANDRA numbers first. SANDRA stands for "System ANalyzer, Diagnostic and Reporting Assistant." We ran three of the built-in subsystem tests that partially comprise the SANDRA 2013 suite with Intel's new Core i7-4960X processor (CPU Arithmetic, Multimedia, and Memory Bandwidth). The scores reported below were taken with the processor running at its default clock speed of 3.6GHz (4GHz Turbo) with 32GB of DDR3-1866 RAM running in quad-channel mode on the EVGA X79 Dark motherboard and then again overclocked at 4.6GHz with DDR3-2400 settings on our Corsair Vengeance Pro series memory.
Ivy Bridge-E Stock (left) and Overclocked (right) - Multimedia Test
Memory Bandwidth Stock (left) and 2400 MHz Overclock (right)
In SANDRA's Arithmetic tests we see a gain of about 27 - 30 percent and again in Floating Point/AVX performance we see about the same. In terms of Memory Bandwidth, Corsair's 32GB kit scales up another 26% in memory bandwidth from 41.6GB/sec to 52.5GB/sec at 1866MHz CAS10 settings to 2400MHz CAS10.
SunSpider shows a massive gain for the overclocked Ivy Bridge setup, posting a time to complete in almost half the time of our stock 3.6GHz CPU / 1866MHz Memory system.
|PCMark 7, Cinebench and Lame MT|
Futuremark PCMark 7 TestingFuturemark's PCMark 7 is the latest version of the PCMark whole-system benchmarking suite. It has updated application performance measurements targeted for a Windows 7 environment and uses newer application metrics to gauge relative performance. Below we're listing the total PCMark score for each test system.
With PCMark 7 we see a much smaller gain achieved by the overclock Core i7-4960X test bed; in fact it only posted a modest 3 - 4 percent gain (3.7 to be exact). In this test specifically, disk IO affords you larger gains versus CPU and Memory throughput, which frankly is the way it works with most real-world light duty creativity and productivity apps.
Cinebench, on the other hand is a totally different animal...
Cinebench R11.5 Testing
Cinebench R11.5 is a 3D rendering performance test based on Cinema 4D from Maxon. Cinema 4D is a 3D rendering and animation suite used by animation houses and producers like Sony Animation and many others. It's very demanding of processor resources and is an excellent gauge of pure computational throughput.
With Cinebench it's all about CPU and Memory throughput for crunching its rendering workload. Here we see a similar 27 percent gain, like we did in our SANDRA testing, with respect to multi-threaded performance. Single-threaded gains were more modest, on the order of about 17 percent.
LAME MT Testing
In our custom LAME MT MP3 encoding test, we convert a large WAV file to the MP3 format, which is a popular scenario that many end users work with on a day-to-day basis to provide portability and storage of their digital audio content. LAME is an open-source MP3 audio encoder that is used widely in a multitude of third party applications.
Performance was more tightly group for our lightly-threaded LAME testing, though there was a solid gain in production run time of 4 seconds for multi-threaded processing. Generally speaking we're looking at a 15 - 20 percent advantage for the overclocked test bed.
|Bioshock Infinite and Total War: Shogun 2|
|In our gaming tests, we've run two sets of numbers here to illustrate a point. Specifically, if you're "GPU-bound" in any game engine or title, cranking up CPU or Memory speeds isn't going to buy you much. We will, however, toss in the one caveat that if you have a multi-GPU setup, small overhead from inter-GPU communication can be alleviated with stronger CPU and Memory subsystem performance.
Regardless, we ran both Shogun Total War 2 and Bioshock Infinite benchmarks, at high resolution / high image quality settings as well as medium resolution / medium image quality settings to show you the total picture.
Total War: Shogun 2 and Bioshock Infinite Game TestsTotal War: Shogun 2 is a turn-based strategy game that loads up lots of tons of battle units on the screen at once, along with DX11 terrain tessellation and ambient occlusion. It can be a pretty sizable overall stress test for midrange setups.
And thankfully our setup is anything but "midrange." Here, the overclocked Core i7-4960X Ivy Bridge and 2400MHz memory setup posts almost negligible gains at high res but a modest 12 percent kicker at 720p with medium settings.
Bioshock Infinite is based on the Unreal Engine 3 but with added effects that do place a bit more stress on the graphics subsystem. Overall it's less workload than Total War: Shogun 2, so we turned up the resolution a bit to 2560X1600 on the high end tests.
More of the same here. At high resolution, we're almost completely GPU-bound but at 720p with medium IQ settings we see about a 10% gain in frame rate.
|Wrap It Up, I'll Take It|
Performance Summary and AnalysisOverclocking Intel's latest $1K flagship CPU allowed us to wring a full 1GHz more clockspeed out of the chip from its base frequency of 3.6GHz, but only 600MHz more over Intel's built-in 4GHz Turbo speed boost. In reality, for most workloads this equates to a modest 15 percent clock speed kicker that remains constant across all six cores of the CPU. However, with the help of some faster memory speeds as well, we were able to realize gains on the order of 20 - 25 percent, for standard compute workloads. Gaming at high resolutions, on the other hand, is not an area where CPU throughput is going to offer much upside, but you knew that. That's what your graphics processor (GPU) is for.
With a six-core chip at 4.6GHz and 32GB of DDR3-2666 memory, folks might call you "baller."
What was more interesting to us were the subtle nuances that we had to accommodate to reach a stable, sustainable overclock. We'll underscore the fact that all of our testing was done on an open air setup, on a test bench. We had to resort to direct airflow over the FET power array cooler on the X79 Dark motherboard as well, to achieve good stability at 4.6GHz. As a safe bet fallback, we'd suggest 4.5GHz is even more attainable for setups with less than optimal air flow, though 4.6GHz can be had without much sweat. However, again, if you're using a self-contained water cooler, you're going to want some extra airflow around the socket area and FET array to keep temperatures nominal and power delivery clean; that is if your case of choice doesn't already accommodate for this.
Also, though you can play with BLCK timings to your heart's content, we were hard pressed to realize a higher clock speed by tweaking timings for this setting. Our best results were realized at a flat 100MHz BLCK.
At the end of the day, it was an interesting and fun exercise taking Ivy Bridge-E to its limits. There's definitely performance headroom to spare in within Intel's six-core beast, and though the better deal is likely overclocking a $594 Core i7-4930K, if you've shelled out the pesos for Intel's top dog, you're probably going to want maximize ROI. And that's what overclocking is all about. Ultimately, we picked up an additional 600MHz of top-end clock speed with commonplace cooling methods. With a little more elbow grease, if you're the type that likes to push the envelope, you might well be able to realize more.