Core i7-3960X Extreme Edition CPU Review

[nextpage title=”Introduction”]

Intel is releasing today its most high-end computing platform, the socket LGA2011 (LGA2011) platform for a new generation of Core i7 processors codenamed “Sandy Bridge-E,” which support the new quad-channel memory architecture. The top-of-the-line model is the Core i7-3960X, which comes with a base clock of 3.3 GHz, Turbo Boost technology pumping the CPU clock up to 3.9 GHz when performance is needed, six processing cores with Hyper-Threading technology, PCI Express 3.0 controller, and more. Let’s see how this beast performs.

This new CPU will be offered for around USD 1,000, which is similar to the previous Intel offerings for the ultra-high-end market, such as the Core i7-980X and the Core i7-990X. Intel is also releasing today the Core i7-3930K for an MSRP of USD 555 (USD 600 at Newegg.com), which has a base clock rate of 3.2 GHz and a Turbo Boost clock rate of 3.8 GHz, and a smaller L3 memory cache (12 MB instead of the 15 MB found on the Core i7-3960X). Intel will release a third LGA2011 model next year, the Core i7-3820, with a base clock rate of 3.6 GHz and a Turbo Boost clock rate of 3.9 GHz, four processing cores with Hyper-Threading technology, and 10 MB of L3 memory cache. Intel didn’t reveal the cost of this CPU.

In Figures 1 and 2, we have a comparison of the physical size of the new Core i7-3960X Extreme Edition to a socket LGA1366 and a socket LGA1155 Core i7 processors.

Figure 1: Core i7-3960X (LGA2011, left), Core i7-990X (LGA1366), and Core i7-2600K (LGA1155)

Figure 2: Core i7-3960X (LGA2011, left), Core i7-990X (LGA1366), and Core i7-2600K (LGA1155)

[nextpage title=”The LGA2011 Platform”]

The new LGA2011 platform features the new quad-channel memory architecture, and it is coming to succeed the previous top-of-the-line platform from Intel, the LGA1366, which features triple-channel memory architecture.

This new generation of socket LGA2011 Core i7 processors supports DDR3 memories up to 1,600 MHz officially, and 2,133 MHz unofficially. In comparison, the LGA1366 supports DDR3 memories up to 1,066 MHz using the triple-channel architecture. This way, we have an increase in memory bandwidth not only because of the increase from 192 (triple-channel) to 256 bits (quad-channel), but also because of the increase in the maximum clock rate supported.

So, while the LGA1366 platform has a maximum memory bandwidth of 25,584 MB/s (25.6 GB/s) when using 1,066 MHz memories, the new LGA2011 platform has a memory bandwidth of 51,200 MB/s (51.2 GB/s) when using 1,600 MHz memories or 68,256 MB/s (68.3 GB/s) when using 2,133 MHz memories, which was the case during our review.

Since each memory module is a 64-bit entity, four memory modules are required for the quad-channel architecture to be enabled. You can read more about this architecture by reading our “Everything You Need to Know about the Dual-, Triple-, and Quad-Channel Memory Architectures” tutorial.

If fewer than four memory modules are installed, the system will work under single-, dual-, or triple- channel, depending if you installed one, two, or three memory modules, respectively.

The chipset released for the new LGA2011 platform is the Intel X79.

The new LGA2011 CPUs have a total of 40 PCI Express 3.0 lanes for video cards. This allows the following video card configurations:

• Two video cards at x16 and one video card at x8
• One video card at x16 and three video cards at x8
• One video card at x16, two video cards at x8, and two video cards at x4

In Figures 3 and 4, you can see the new LGA2011 socket. Unlike previous sockets, it has two levers that must be moved to install and remove the CPU instead of only one.

Figure 3: LGA2011 socket

Figure 4: LGA2011 socket

[nextpage title=”Intel Liquid Cooling”]

Together with the LGA2011 platform, Intel is launching an optional liquid cooling solution, simply called “Intel Liquid Cooling.” This solution is actually manufactured by Asetek, will retail between USD 85 and USD 100, and is compatible with sockets 1155, 1156, and 1366 as well. In Figure 5, you can see this CPU cooler and, in Figure 6, its main specification. This was the cooling solution we used during our tests.

Figure 5: Intel Liquid Cooling

Figure 6: Intel Liquid Cooling specifications

[nextpage title=”The Reviewed CPUs”]

In the tables below, we compare the main features of the CPUs we included in this review. We are basically comparing the new Core i7-3960X to the previous fastest and most expensive CPU from Intel, the Core i7-990X. AMD doesn’t have any product that competes directly with these CPUs, as the most high-end AMD processor, the FX-8150, costs USD 270, less than one third of the price of the CPU we are reviewing. Nevertheless, we included this AMD CPU in our comparison, even though we know that it is unfair.

The LGA2011 and LGA1366 CPUs support Hyper-Threading technology, meaning that each processing core is recognized as two by the operating system. Therefore, the two Core i7 processors we included in our review are recognized as 12-core products.

CPU	Cores	HT	Internal Clock	Turbo Clock	Max Turbo	Core	Technology	TDP	Socket	Price
AMD FX-8150	8	No	3.6 GHz	3.9 GHz	4.2 GHz	Zambezi	32 nm	125 W	AM3+	USD 270
Core i7-990X	6	Yes	3.46 GHz	3.73 GHz	3.73 GHz	Gulftown	32 nm	130 W	1366	USD 1,000
Core i7-3960X	6	Yes	3.3 GHz	3.9 GHz	3.9 GHz	Sandy Bridge-E	32 nm	130 W	2011	USD 1,050

The pric
es were researched at Newegg.com on the day we published this review.

TDP stands for Thermal Design Power, which tells the maximum amount of heat the CPU can dissipate. The CPU cooler must be capable of dissipating at least this amount of heat.

Below we compare the memory cache configuration of the reviewed CPUs. The L2 memory cache of the AMD FX-8150 is shared by each pair of cores. Thus, it has four L2 caches instead of eight.

CPU	L1 Cache	L2 Cache	L3 Cache	Memory Support	Memory Channels
AMD FX-8150	64 KB + 64 KB per core	2 MB x 4	8 MB total	Up to DDR3-1866	Two
Core i7-990X	32 KB + 32 KB per core	256 KB per core	12 MB total	Up to DDR3-1066	Three
Core i7-3960X	32 KB + 32 KB per core	256 KB per core	15 MB total	Up to DDR3-2133	Four

[nextpage title=”How We Tested”]

During our benchmarking sessions, we used the configuration listed below. Between our benchmarking sessions the only variable device was the CPU being tested and the motherboard, which had to be replaced to match the different CPU sockets.

Hardware Configuration

• Motherboard (Socket AM3+): ASUS Crosshair V Formula (0813 BIOS)
• Motherboard (Socket LGA1366): MSI Big Bang XPower (1.6 BIOS)
• Motherboard (Socket LGA2011): Intel DX79SI (S1.0280B BIOS)
• CPU Cooler: Intel Liquid Cooling/AMD stock
• Memory: 16 GB DDR3-2133/PC3-17000, four G.Skill Ripjaws Z F3-17000CL9Q-16GBZH memory modules
• Hard Disk Drive: Western Digital Black Caviar 1 TB (WD1001FALS, SATA-300, 7,200 rpm, 32 MB buffer)
• Video Card: Radeon HD 6970
• Video Monitor: Samsung Syncmaster 305T Plus
• Power Supply: Antec TruePower New 750 W

Operating System Configuration

• Windows 7 Ultimate 64-bit
• NTFS
• Video resolution: 2560×1600 60 Hz

Driver Versions

• AMD video driver version: Catalyst 11.10
• AMD chipset driver version: 3.0.816.0
• Intel Inf chipset driver version (socket LGA2011): 9.2.3.1022
• Intel Inf chipset driver version (socket LGA1366): 9.1.2.1008

Software Used

• 3DMark 11 1.0.2
• Adobe After Effects CS4
• Adobe Photoshop CS4 Extended + GamingHeaven Photoshop Benchmark V3
• Cinebench 11.529
• Deus Ex: Human Revolution
• DiRT3
• Lost Planet 2
• iTunes 10.2
• Media Espresso 6.5
• PCMark 7 1.0.4
• Starcraft II: Wings of Liberty – Patch 1.3.4
• VirtualDub 1.9.5 + MPEG-2 Plugin 3.1 + DivX 6.9.2
• WinZip 15.5

Error Margin

We adopted a 3% error margin. Thus, differences below 3% cannot be considered relevant. In other words, products with a performance difference below 3% should be considered as having similar performance.

[nextpage title=”PCMark 7″]

The new PCMark 7 performs a series of tests and gives scores in the following categories: An overall score called PCMark; a Productivity score, which is the system performance when using applications such as web browsing and home office applications; a Creativity score, which is the system performance when viewing, editing, converting, and storing photos and videos; an Entertainment score, which is the system performance when recording, viewing, streaming, and converting TV shows and movies, importing, organizing, and browsing music, and gaming; and a Computation score, which indicates the processing performance of the system. Let’s analyze the results.

The overall PC Mark 7 score of the Core i7-3960X was 5.9% higher than Core i7-990X’s and 2.9% higher than AMD FX-8150’s.

 

The Core i7-3960X PC Mark 7 Productivity score was similar to the Core i7-990X’s and 7.5% higher than the AMD FX-8150’s.

The Core i7-3960X PC Mark 7 Creativity score was 6.5% higher than the Core i7-990X’s and 12.9% higher than the AMD FX-8150’s.

The Core i7-3960X PC Mark 7 Entertainment score was 9.2% higher than the Core i7-990X’s and 14.2% higher than the AMD FX-8150’s.

The Core i7-3960X PC Mark 7 Computation score was 23% higher than the Core i7-990X’s and 41.4% higher than the AMD FX-8150’s.

[nextpage title=”DivX Encoding”]

We converted a full-length DVD movie to DivX format using VirtualDub and saw how long it took to complete this conversion. DivX codec is capable of recognizing and using not only more than one CPU (i.e., more than one core) but also the SSE4 instruction set.

The movie we chose to convert was “Star Trek – The Motion Picture: Director’s Cut.” We copied the movie to our hard disk drive with no compression, so the final original file on our HDD was 6.79 GB. After compressing it with DivX, the final file was only 767.40 MB, which is quite remarkable.

The results below are given in seconds, so the lower the better.

On DivX encoding, the Core i7-3960X was 15.8% faster than the Core i7-990X and 45% faster than the AMD FX-8150.

[nextpage title=”Photoshop CS4″]

The best way to measure performance is by using real programs. The problem, though, is creating a methodology using real software that provides accurate results. For Photoshop CS4, there is a methodology created by the folks at GamingHeaven that is very accurate. Their script applies a series of 15 filters to a sample image. Then we wrote down the time taken for each filter to run. At the end, we had the results for each individual filter, which we added up to get the total time taken to run the 15 filters. The results below are given in seconds, so the lower the number the better.

On Photoshop CS4, the Core i7-3960X was 22.9% faster than the Core i7-990X and 36.2% faster than the AMD FX-8150.

[nextpage title=”After Effects CS4″]

After Effects is a very well-known program for video post-production, used to add animation and visual effects in videos. To evaluate the performance of each CPU running this program, we ran a workload consisting of a number of compositions that applied several filters and effects to a variety of input file types such as PSD (Photoshop), AI (Illustrator), EPS and TIF. After each filter was applied, the composition was rendered to an uncompressed AVI file with the same resolution as the input files. The results below are the time each CPU took to finish the whole batch, given in seconds, so the lower the number the better.

On After Effects CS4, the Core i7-3960X was 14.6% faster than the Core i7-990X and 107.1% faster than the AMD FX-8150.

[nextpage title=”Media Espresso 6.5″]

Media Espresso is a video conversion program that uses the graphics processing unit of the video card to speed up the conversion process. It is also capable of using Intel’s QuickSync technology available in the CPUs from this company. We converted a 449 MB, 1920x1080i, 18,884 kbps, MPG2 video file to a smaller 640×360, H.264, .MP4 file for viewing on a portable device such as an iPhone or iPod Touch.

Here we forced Media Espresso to use the CPU for the conversion process, not the GPU. The Core i7-3960X was 30.4% faster than the Core i7-990X and 54.3% faster than the AMD FX-8150.

[nextpage title=”WinZip”]

We used WinZip not only to measure compression time, but also decryption time. We measured the time each CPU took to decompress and decrypt 200 JPEG images, 125 of them at 10 megapixels and 75 of them at six megapixels. The total size of all the photos was around 830 MB. The results below are given in seconds, so the lower the number the better.

Decompressing and decrypting files, the Core i7-3960X was 10% faster than the Core i7-990X and 35% faster than the AMD FX-8150.

[nextpage title=”iTunes”]

We used iTunes to convert an uncompressed .wav file into a high-quality (160 Kbps) MP3 file, and checked how many seconds each CPU took to perform this operation. Therefore, the results below are given in seconds, so the lower the number the better.

On MP3 conversion, the Core i7-3960X was 9.4% faster than the Core i7-990X and 31.3% faster than the AMD FX-8150.

[nextpage title=”Cinebench 11.5″]

Cinebench 11.5 is based on the 3D software Cinema 4D. It is very useful to measure the performance gain given by having more than one CPU installed on the system when rendering heavy 3D images. Rendering is one area where having more than one CPU helps greatly, because usually, rendering software recognizes several CPUs. (Cinebench, for instance, can use up to 16 CPUs.)

Since we were interested in measuring the rendering performance, we ran the test called “Rendering x CPUs,” which renders a “heavy” sample image using all available CPU cores – either real or virtual, as on CPUs with Hyper-Threading technology, each CPU core is recognized as two cores by the operating system.

On Cinebench, the Core i7-3960X achieved a score 16.7% higher than Core i7-990X’s and 72.4% higher than AMD FX-8150’s.

 

[nextpage title=”StarCraft II: Wings of Liberty”]

StarCraft II: Wings of Liberty is a very popular DirectX 9 game that was released in 2010. Though this game uses an old version of DirectX, the number of textures that can be represented on one screen can push most of the top-end graphics cards to their limits (especially when the graphics settings are set at “Ultra”). StarCraft II: Wings of Liberty uses its own physics engine that is bound to the CPU and thus does not benefit from PhysX.

We tested this game at 1920×1200. The quality of the game was set to the “low” preset, disabling both anti-aliasing and anisotropic filtering. We then used FRAPS to collect the frame rate of a replay on the “Unit Testing” custom map. We used a battle between very large armies to stress the video cards.

On StarCraft II, the Core i7-3960X and Core i7-990X achieved the same performance level. The reviewed CPU was only 3.6% faster than the AMD FX-8150 on this game.

[nextpage title=”Lost Planet 2″]

Lost Planet 2 is a game that uses many DirectX 11 features, like tessellation (to round out the edges of polygonal models), displacement maps (added to the tessellated mesh to add fine grain details), DirectCompute soft body simulation (to introduce more realism in the “boss” monsters), and DirectCompute wave simulation (to introduce more realism in the physics calculations in water surfaces; when you move or when gunshots and explosions hit the water, it moves accordingly). We reviewed the video cards using Lost Planet 2 internal benchmarking features, choosing the “Benchmark A.” (We know that “Benchmark B” is the one recommended for reviewing video cards, however, at least with us, results were inconsistent.) We ran this game at 1920×1200 with graphics set at “low,” with no anti-aliasing and no anisotropic filtering. The results below are the number of frames per second generated by each system.

On Lost Planet, the three CPUs included in our comparison achieved the same performance level.

 

[nextpage title=”Deus Ex: Human Revolution”]

Deus Ex: Human Revolution is a DirectX 11 game, and we used in-game introduction to measure the number of frames per second using FRAPS. We configured the video resolution at 1920×1200 with low settings and anti-aliasing disabled.

On Deus Ex, the three CPUs included in our comparison achieved the same performance level.

[nextpage title=”DiRT3″]

DiRT3 is another DirectX 11 game. We measured performance using this game by running a race and then playing it back using FRAPS. We configured the video resolution at 1920×1200 with low settings and both anti-aliasing and anisotropic filtering disabled. The results below are in frames per second.

On DiRT3, the Core i7-3960X and Core i7-990X achieved the same performance level. The reviewed CPU was only 5.7% faster than the AMD FX-8150 on this game.

[nextpage title=”3DMark 11 Professional”]

The 3DMark 11 measures Shader 5.0 (i.e., DirectX 11) performance. We ran this program at 1920×1200 using the “Performance” profile.

This program provides three different scores: graphics, physics and combined.

The graphics score achieved by the three CPUs was the same. This was expected, since this score measures the performance of the video card, not the processor.

The physics score measures exclusively the physics performance of the system, a process that is typically done on the CPU. Here the Core i7-3960X achieved a score 13.6% higher than the Core i7-990X’s and 83.2% higher than the AMD FX-8150’s.

The combined score shows a balance between the graphics and the physics performance achieved by each system being tested. Here the Core i7-3960X achieved a score 4.6% higher than the Core i7-990X’s and 4.3% higher than the AMD FX-8150’s.[nextpage title=”Overclocking”]

Since the Core i7-3960X is an unlocked CPU, we have several adjustments available, including the adjustment of the clock multiplier for each core and the TDP and current limits of the CPU.

The “correct” way of overclocking Intel CPUs with Turbo Boost technology is to change the clock multiplier that each core will use when the Turbo Boost technology kicks in. This can be done on the motherboard setup or using the Intel Extreme Tuning Utility, which has a handy stress test to see if the system is stable.

We could put the Core i7-3960X to work with its six cores at 4.3 GHz at the same time by increasing the clock multiplier of all six cores from 33x to 43x, after increasing the CPU voltages (see Figure 7) and the current limit from 135 A to 165 A. We were afraid of increasing the current limit even further and burning our CPU, so if you have nerves of steel you may achieve an even better result.

Figure 7: Six cores working at 4.3 GHz

Then, we decided to reduce the number of active cores. This way, the total thermal dissipation is decreased and, because of that, we were able to increase the clock rate of the active cores. We disabled two cores for a total of four active cores and were able to increase the clock multiplier for these cores to x45, making them work at 4.5 GHz. We used the same voltages and current limit as before.

Figure 8: Four cores working at 4.5 GHz

Even on locked CPUs, you can increase the CPU base clock; however, second-generation Core i processors have very tight overclocking limits for this clock. The system crashed whenever we tried to increase the CPU base clock during the tests described above.

Once again, we were afraid of burning the CPU and probably didn’t reach all the overclocking potential of this CPU.

[nextpage title=”Conclusions”]

The new Core i7-3960X does live up to its expectation of being the fastest desktop processor released to date. Costing the same as the previous top-of-the-line desktop CPU, the Core i7-990X, it provides a performance improvement of up to 30%, but you should get a typical 10-15% performance increase on most programs.

Like its predecessors, the Core i7-3960X is not targeted to the average user, since it is too expensive for the performance gain you will get. It is also not worthwhile for gamers, since gaming performance is limited by the video card, not the CPU, if you already have a high-end CPU. For the average user looking for a fast CPU, our recommendation is the Core i5-2500 and for the high-end user, the Core i7-2600K.

However, for the professional user who works with video edi
ting and other CPU-intensive applications, this new CPU is an excellent option, as it will have your work done faster. The fastest you can finish jobs, the more jobs you can take, meaning more money that you can make.