| Intel Core i7-3960X Extreme Edition CPU |
| Reviews - Featured Reviews: Processors | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Written by David Ramsey | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Monday, 14 November 2011 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Intel Core i7-3960X Extreme Edition CPU Review
Manufacturer: Intel Corporation Full Disclosure: The product sample used in this article has been provided by Intel. Intel's "Extreme Edition" CPUs have always represented the company's top consumer offerings. Typically priced in the $1,000 range, they have unlocked multipliers, lots of cache, and lots of cores. But until now, Intel's top Extreme Edition offering, the Core i7-990X CPU, was based on the older Gulftown architecture, and the performance gap between this CPU and the newer Sandy Bridge architecture Core i7-2600K and 2700K is pretty damn narrow, especially considering that the latter costs less than a third the price of the former. But now Intel's made a Sandy Bridge Extreme Edition, with six physical cores and a staggering 15 megabytes of cache. Benchmark Reviews takes the new Intel Core i7-3960X Extreme Edition CPU around the benchmark course, testing it against the best CPUs Intel and AMD have to offer. Intel's Sandy Bridge processors, especially the Core i5-2500K and the Core i7-2600K, set very high performance bars when they were introduced in early 2011. The new architecture significantly increased the instructions per clock (i.e. they were faster at the same clock speed than older CPUs), and the 32nm fabrication process enabled amazing overclocks, with 4.6GHz and higher frequencies being easily obtainable with air cooling. A 2600K CPU will beat a 980X CPU in several benchmarks, and the older CPU only really dominates in heavily threaded applications that can make use of its extra cores.
So now we have a hex-core Sandy Bridge processor and new X79 Express chipset to support it. Intel Core i7-3960X FeaturesThe following information is courtesy of Intel
Intel (SKU) SpecificationsIt's getting confusing in Intel-land, with the "Core i7" brand now straddling no fewer than three platforms: LGA1155, LGA1366, and now LGA2011. The LGA2011 group initially comprises three processors, two of which are available now, and one scheduled for the first quarter of 2012. I've added the current top of the LGA1155 line, the Core i7-2700K, for comparison.
As you can see, the new LGA2011 CPUs are all quite similar, with differences in cache, cores, and "unlockability" being the main differences. All require new motherboards with the X79 chipset, and all of them boast Intel's new Turbo Boost 2.0 technology (introduced with the original Sandy Bridge processors). In the case of the Core i7-3960X, this means a boost of up to 300MHz when five or more cores are under load, and to 600MHz when fewer than five cores are under load. And just in case you're wondering: yes, X79 motherboards again change the spacing of the cooler mounting holes, so you'll need either a new cooler or an adapter for your existing cooler. This is understandable, though, since the new chip is absolutely gigantic, at 52.5 x 45mm. The picture below shows a Core i5-2500K on the left and a Core i7-3960X on the right. 
If someone had shown me this chip without telling me what it was, I'd have assumed it was an eight-core Xeon or something! Intel does not provide a standard cooler with any Sandy Bridge Extreme CPU, but will make air and water coolers available. For this review they provided their Asetek-sourced "Thermal Solution RTS2011LC", which they estimate will go for between $85 and $100 at the retail level. This cooler appears almost identical to the Antec KÜHLER H2O 620, but Intel says Asetek made changes specific to this model. 
Join me in the next section as I take a look at the Sandy Bridge Extreme architecure, and how it differs from the LGA1155 Sandy Bridge processors. Sandy Bridge Extreme ArchitectureArchitecturally, the Sandy Bridge Extreme CPUs are very similar to the existing Sandy Bridge lineup. Each core in the i7-3960X is pretty much the same as the core in an i7-2700K. The main differences between the Core i7-3960X and the Core i7-2700K are:
Most users will welcome the lack of the weak "iGPU", although its removal does mean that you can't use Intel Quick Sync for video transcoding. The most important new feature, in my opinion, is the 40 PCI-E lanes, which is a vast improvement over the stingy 16 lanes available on LGA1155 Sandy Bridge CPUs. Lane contention on an LGA1155 system with multiple video cards results in diminished performance or even non-functionality of some other interfaces that require PCI-E bandwidth, like USB 3.0 and SATA 6G ports. This won't be a problem with LGA2011. Technically, the new PCI-E lanes meet the PCI-E 3.0 specification, which means that at least in theory they should be able to deliver double the bandwidth of existing PCI-E 2.0 lanes. Some day, when PCI-E 3.0 peripherals actually exist, we might see some performance improvement...although even the beefiest current graphics cards aren't bandwidth-limited. 
Of course, with a new CPU and new socket comes a new supporting chipset, the X79 Express. 
Looking at the block diagrams of the Z68 and X79 chipsets side by side, the X79 seems functionally identical to the Z68, except that it lacks the digital display support and Intel Smart Response Technology. The lack of the latter is disappointing, since our tests with Intel Smart Response Technology showed that its use of an SSD as an intelligent cache to a hard drive could dramatically improve storage performance. As with the Z68 and earlier P67 chipsets, there are 14 USB 2.0 ports and 6 SATA ports, of which only two are SATA 6G. Notably missing is Intel's "Light Peak" (aka "Thunderbolt"), which has been used as Intel's excuse for not supporting SuperSpeed USB 3.0. And it's really odd that only two of the SATA ports are SATA 6G, since 6G devices are becoming more common, especially among SSDs. For a cutting-edge platform, this is impossible to justify. At least AMD gives you a full six SATA 6G ports. Disappointments with the chipset aside, let's get to testing this setup. Processor Testing MethodologyFor this review, I compared the Intel Core i7-3960X at stock and overclocked speeds on an ASUS Sabertooth X79 motherboard to an Intel Core i7-2600K on an ASUS P8P67 motherboard, an Intel Core i7-980X on an ASUS Rampage III Extreme motherboard, and an AMD FX-8150 on an ASUS Crosshair V Formula motherboard. The Intel platforms all used DDR3-1600 memory at 9-9-9-24 timings, whereas the AMD platform used DDR3-1866 memory at the same 9-9-9-24 timings. The primary drive for all platforms was a Western Digital VelociRaptor 300G, and the same Radeon HD6850 video card was use for all platforms as well. Intel X58 Express Test Platform
Intel P67 Express Test Platform
AMD 990FX Test Platform
Intel X79 Express Test Platform
Benchmark Applications
SiSoft Sandra Memory BandwidthSiSoftware Sandra (the System ANalyser, Diagnostic and Reporting Assistant) is an information & diagnostic utility. It should provide most of the information (including undocumented) you need to know about your hardware, software and other devices whether hardware or software. It's available in five versions, ranging from the free "Lite" version to the Sandra Enterprise version. Sandra fully supports and exploits multi-processor and multi-core systems, NUMA memory, Hyper-Threading, and MMX, SSE, SSE2, SSE3, SSSE3, SSE 4.1, SSE 4.2, AVX, and FMA instructions. It comprises a massive suite of test and reporting features...but the one I'm interested in is the memory bandwidth test: SiSoft's memory bandwidth test provides individual and aggregate scores across a variety of memory operations. For this test I'm reporting the aggregate score. Does the new quad-channel memory system really make that much of a difference? 
And the answer is yes...yes it does. The AMD FX-8150 comes in the loser here, even though it's running its dual-channel memory at a slightly higher speed than the Intel platforms. The 2600K's improved memory controller more than makes up for its one-channel deficit relative to the 980X's triple-channel setup, but just look at those 3960X scores: memory bandwidth (at least as measured by Sandra) virtually doubles the next best score. AIDA64 Extreme Edition TestsAIDA64 is a full 64-bit benchmark and test suite utilizing MMX, 3DNow! and SSE instruction set extensions, and will scale up to 32 processor cores. An enhanced 64-bit System Stability Test module is also available to stress the whole system to its limits. For legacy processors all benchmarks and the System Stability Test are available in 32-bit versions as well. Additionally, AIDA64 adds new hardware to its database, including 300 solid-state drives. On top of the usual ATA auto-detect information the new SSD database enables AIDA64 to display flash memory type, controller model, physical dimensions, and data transfer performance data. AIDA64 v1.00 also implements SSD-specific SMART disk health information for Indilinx, Intel, JMicron, Samsung, and SandForce controllers. All of the benchmarks used in this test— Queen, Photoworxx, ZLib, hash, and AES— rely on basic x86 instructions, and consume very little system memory while also being aware of Hyper-Threading, multi-processors, and multi-core processors. Of all the tests in this review, AIDA64 is the one that best isolates the processor's performance from the rest of the system. While this is useful in that it more directly compares processor performance, readers should remember that virtually no "real world" programs will mirror these results. 
The Queen and Photoworxx tests are synthetic benchmarks that iterate the function many times and over-exaggerate what the real-world performance would be like. The Queen benchmark focuses on the branch prediction capabilities and misprediction penalties of the CPU. It does this by finding possible solutions to the classic queen problem on a chessboard. At the same clock speed, theoretically the processor with the shorter pipeline and smaller misprediction penalties will attain higher benchmark scores. Like the Queen benchmark, the Photoworxx tests for penalties against pipeline architecture. The synthetic Photoworxx benchmark stresses the integer arithmetic and multiplication execution units of the CPU and also the memory subsystem. Due to the fact that this test performs high memory read/write traffic, it cannot effectively scale in situations where more than two processing threads are used, so quad-core processors with Hyper-Threading have no real advantage. The AIDIA64 Photoworxx benchmark performs the following tasks on a very large RGB image:
The 3960X produces the highest scores here, massively so in the case of PhotoWorxx. Note how the improved Sandy Bridge core architecture enables the much less expensive four-core I7-2600K to turn in performance similar to the very expensive six-core 980X. It was an incredible CPU in its day, but you don't have to spend that much money to get that kind of performance now. And if you do want to spend that much money, well, that's why the 3960X exists. 
AMD's "Bulldozer" FX-8150 turns in a pretty credible performance in these tests, even coming close to the 3960X's performance in the Hash test. But the new Intel CPU still wins. 
Intel's Clarksdale and subsequent CPUs have dominated the AES test due to their Advanced Encryption Standard New Instructions (AES-NI), which dramatically accelerate AES code. AMD's own implementation of AES-NI makes its first appearance in Bulldozer-based CPUs, and in the ASUS motherboard turns in performance competitive with the Intel CPUs except for the 3960X. The new CPU's performance is simply staggering, over 70% faster than the already fast 980X. Let's move on to the PCMark Vantage benchmark. PCMark Vantage TestsPCMark Vantage is an objective hardware performance benchmark tool for PCs running 32- and 64-bit versions of Microsoft Windows Vista or Windows 7. It's well suited for benchmarking any type of Microsoft Windows Vista/7 PC: from multimedia home entertainment systems and laptops, to dedicated workstations and high-end gaming rigs. Benchmark Reviews has decided to use a few select tests from the suite to simulate real-world processor usage in this article. Our tests were conducted on 64-bit Windows 7, with results displayed in the chart below. TV and Movies Suite
Gaming Suite*
Music Suite
* EDITOR'S NOTE: Hopefully our readers will carefully consider how relevant PCMark Vantage is as a "real-world" benchmark, since many of the tests rely on unrelated hardware components. For example, per the FutureMark PCMark Vantage White Paper document, Gaming test #2 weighs the storage device for 100% of the test score. In fact, according to PCMark Vantage the video card only impacts 23% of the total gaming score, but the CPU represents 37% of the final score. As our tests in this article (and many others) have already proven, gaming performance has a lot more to do with the GPU than the CPU, and especially more than the hard drive or SSD (which is worth 38% of the final gaming performance score). 
The TV and Movies suite concentrates on video playback and transcoding, but only uses two threads at a maximum, so performance here is based more on clock speed and IPC (instructions per clock) than anything else. Here we see one of the closest performance clusters of this review, with the overclocked 3960X only 38% faster than the FX-8150 and 8% faster than the 2600K. The Gaming benchmark relies on the hard disk and video card for over 50% of its score (see the Editor's Note above), and we're using the same HDD and video card for all platforms, so the Intel decisive win in this test simply means that Vantage's gaming code is more optimized for Intel processors. Bear in mind, however, that most "real world" games will not show this difference; generally, in games, your video card matters most, followed by the clock speed (not number of cores) of your processor. The PCMark Vantage gaming test can use up to 16 threads, so it's very strange that the eight core FX-8150 turns in markedly lower scores than the four-core Intel processors, even considering Hyper-Threading. Unlike the Gaming test, the Music test results have more real-world relevance, since multi-threading is much more common in music transcoding applications than it is in games. Even so, the advanced Sandy Bridge cores mean that the four-core i7-2600K beats the six-core i7-980X. The 3960X wins again, albeit not by much. The wins are piling up in the Intel column. Let's move on to CINEBENCH. CINEBENCH R11.5 BenchmarksMaxon CINEBENCH is a real-world test suite that assesses the computer's performance capabilities. CINEBENCH is based on Maxon's award-winning animation software, Cinema 4D, which is used extensively by studios and production houses worldwide for 3D content creation. Maxon software has been used in blockbuster movies such as Spider-Man, Star Wars, The Chronicles of Narnia, and many more. CINEBENCH Release 11.5 includes the ability to more accurately test the industry's latest hardware, including systems with up to 64 processor threads, and the testing environment better reflects the expectations of today's production demands. A more streamlined interface makes testing systems and reading results incredibly straightforward. The CINEBENCH R11.5 test scenario comprises three tests: an OpenGL-based test that models a simple car chase, and single-core and multi-core versions of a CPU-bound computation using all of a system's processing power to render a photo-realistic 3D scene, "No Keyframes", the viral animation by AixSponza. This scene makes use of various algorithms to stress all available processor cores, and all rendering is performed by the CPU: the graphics card is not involved except as a display device. The multi-core version of the rendering benchmark uses as many cores as the processor has, including the "virtual cores" in processors that support Hyper-Threading. The resulting "CineMark" is a dimensionless number only useful for comparisons with results generated from the same version of CINEBENCH. What a nice, linear set of results! CINEBENCH really uses all the threads it can, and the more threads a CPU can dispatch, the better it's going to do on this test.
Although CINEBENCH is a synthetic benchmark, it does show one of the areas where all the cores and speed you can afford will be used well: professional-level rendering tasks. 
In single-core rendering, the 2600K and 3960X return virtually identical scores, both substantially ahead of the older 980X. This shows how much more efficient the Sandy Bridge core is compared to the older Gulftown core. CPU-Dependent 3D GamingBenchmark Reviews continually evaluates the various tests and benchmarks we use, and we have switched from Ubisoft's Far Cry 2 benchmark to CAPCOM's Street Fighter IV benchmark. Street Fighter IV uses a new, built-from-scratch graphics engine that enables CAPCOM to tune the visuals and performance to fit the needs of the game, as well as run well on lower-end hardware. Although the engine is based on DX9 capabilities, it does add soft shadows, High Dynamic Range lighting, depth of field effects, and motion blur to enhance the game experience. The game is multi-threaded, with rendering, audio, and file I/O all running in different threads. The development team has also worked to maintain a relatively constant CPU load in all parts of the game so that on-screen performance does not change dramatically in different game scenarios.
I ran the Street Fighter IV benchmark at its lowest resolution (640x480) with all graphical features turned down to the minimum possible settings. This makes the video card much less of a factor in the results, biasing towards processor performance. Again, we see efficient Sandy Bridge cores trouncing the older cores of the 980X, and the 3960X winning by a relatively small margin. PassMark PerformanceTest 7.0The PassMark PerformanceTest allows you to objectively benchmark a PC using a variety of different speed tests and compare the results to other computers. PassMark comprises a complete suite of tests for your computer, including CPU tests, 2D and 3D graphics tests, disk tests, memory tests, and even tests to determine the speed of your system's optical drive. PassMark tests support Hyper-Threading and systems with multiple CPUs, and allow you to save benchmark results to disk (or to export them to HTML, text, GIF, and BMP formats). Knowledgeable users can use the Advanced Testing section to alter the parameters for the disk, network, graphics, multitasking, and memory tests, and create individual, customized testing suites. But for this review I used only the built-in CPU tests, which aren't configurable. The CPU tests comprise a number of different metrics. The first three I'll look at are integer performance, floating point performance, and a benchmark that finds prime numbers.
Intel's strength has always been its integer units, and we see AMD turning in barely over half the performance of the i7-2600K. This is the single biggest factor in Intel's performance advantage over AMD. AMD's floating point performance, on the other hand, stomps all over Intel with the exception of the overclocked i7-3960X. Now if only AMD could find a benchmark that really used a lot of floating point math! AMD's FX-8150 also dominates in the Prime benchmark, again with the exception of the 3960X.
SSE stands for "Streaming SIMD Extensions", and are instructions that handle multiple chuncks of data per instruction (SIMD = Single Instruction Multiple Data). SSE instructions work on single-precision floating point data and are typically used in graphical computations. SSE was Intel's response to AMD's "3D Now", which itself was a response to Intel's MMX instructions. Don't you love competition? AMD's current implementation is actually quite good: notice how it beats the 980X? But the 3960X still easily trounces all comers in both benchmarks.
The Compress and String benchmarks are both integer-based, but the FX-8150 actually does pretty well against Intel except for the 3960X. There's no doubt: this thing's a game-changer. Now, let's move onto some more real-world applications. Handbrake Media EncodingIt's a truism that consumer-level computer performance reached the "fast enough" point years ago, where increases in system performance don't make things any faster for most people. Web browsing, e-mail, word processing, and even most games won't benefit dramatically from a super-fast CPU. There are some exceptions, though, and media encoding is one of them: transcoding video, especially high-definition video, can bring the strongest system to its knees. Fortunately, media transcoding is one of those things (depending on the design of the code, of course) that scales really well with both clock speed and the number of cores, so the more you have of both, the better your results will be. The free and open-source Handbrake 0.95 video transcoder is an example of a program that makes full use of the computational resources available. For this test I used Handbrake 0.95 to transcode a standard-definition episode of Family Guy to the "iPhone & iPod Touch" presets, and recorded the total time (in seconds) it took to transcode the video.
Here's a nice, linear progressive decrease in encoding times as we move "up the CPU scale". Comparing the encoding times of the four-core i7-2600K and the six-core Sandy Bridge Extreme processors, there's almost perfect scaling with the two additional cores of the latter, with the encoding time dropping only a little less than you'd expect. x264 HD Benchmark 3.19Tech ARP's x264 HD Benchmark comprises the Avisynth video scripting engine, an x264 encoder, a sample 720P video file, and a script file that actually runs the benchmark. The script invokes four two-pass encoding runs and reports the average frames per second encoded as a result. The script file is a simple batch file, so you could edit the encoding parameters if you were interested, although your results wouldn't then be comparable to others.
The AMD Bulldozer does well here, matching the performance of the i7-2600K. But cores win in this scenario, and the six-core Extreme processors simply walk away from the rest of the field. 
The 2600K takes a surprising dive here, but again the six core CPUs are in their element. SPECviewperf 11 testsThe Standard Performance Evaluation Corporation is "...a non-profit corporation formed to establish, maintain and endorse a standardized set of relevant benchmarks that can be applied to the newest generation of high-performance computers." Their free SPECviewperf benchmark incorporates code and tests contributed by several other companies and is designed to stress computers in a reproducible way. SPECviewperf 11 was released in June 2010 and incorporates an expanded range of capabilities and tests. Note that results from previous versions of SPECviewperf cannot be compared with results from the latest version, as even benchmarks with the same name have been updated with new code and models. SPECviewperf comprises test code from several vendors of professional graphics modeling, rendering, and visualization software. Most of the tests emphasize the CPU over the graphics card, and have between 5 and 13 sub-sections. For this review I ran the Lightwave, Maya, and Seimens Teamcenter Visualization tests. Results are reported as abstract scores, with higher being better. LightwaveThe lightwave-01 viewset was created from traces of the graphics workloads generated by the SPECapc for Lightwave 9.6 benchmark. The models for this viewset range in size from 2.5 to 6 million vertices, with heavy use of vertex buffer objects (VBOs) mixed with immediate mode. GLSL shaders are used throughout the tests. Applications represented by the viewset include 3D character animation, architectural review, and industrial design. MayaThe maya-03 viewset was created from traces of the graphics workload generated by the SPECapc for Maya 2009 benchmark. The models used in the tests range in size from 6 to 66 million vertices, and are tested with and without vertex and fragment shaders. State changes such as those executed by the application- including matrix, material, light and line-stipple changes- are included throughout the rendering of the models. All state changes are derived from a trace of the running application. Siemens Teamcenter Visualization MockupThe tcvis-02 viewset is based on traces of the Siemens Teamcenter Visualization Mockup application (also known as VisMockup) used for visual simulation. Models range from 10 to 22 million vertices and incorporate vertex arrays and fixed-function lighting. State changes such as those executed by the application- including matrix, material, light and line-stipple changes- are included throughout the rendering of the model. All state changes are derived from a trace of the running application.
The SPECviewperf suite is a good example of a real-world test of applications that would normally be the province of a high-end workstation: the individual tests comprise code and models from real applications, running scripts that do real work. Here we see that four Sandy Bridge cores again trounce six Gulftown cores, but here's a surprise: they also (at stock clock speeds) beat six Sandy Bridge Extreme cores, if only slightly. SPECapc LightwaveSPECapc (Application Performance Characterization) tests are fundamentally different from the SPECviewperf tests. While SPECviewperf tests incorporate code from the various test programs directly into the benchmark, the SPECapc tests are separate scripts and datasets that are run against a stand-alone installation of the program being benchmarked. SPECapc group members sponsor applications and work with end-users, user groups, publications and ISVs to select and refine workloads, which consist of data sets and benchmark script files. Workloads are determined by end-users and ISVs, not SPECapc group members. These workloads will evolve over time in conjunction with end-users' needs and the increasing functionality of PCs and workstations. For this test, I ran the SPECapc "Lightwave" benchmark against a trial installation of Newtek's Lightwave 3D product. The benchmark, developed in cooperation with NewTek, provides realistic workloads that simulate a typical LightWave 3D workflow. It contains 11 datasets ranging from 64,000 to 1.75 million polygons and representing such applications as 3D character animation, architectural review, and industrial design. Scores for individual workloads are composited under three categories: interactive, render and multitask. The benchmark puts special emphasis on processes that benefit from multi-threaded computing, such as animation, OpenGL playback, deformations, and high-end rendering that includes ray tracing, radiosity, complex textures and volumetric lighting. The test reports three scores: Animation (multitasking), Animation (interactive), and Rendering. The numeric scores represent the time it took to complete each section of the benchmark, in seconds, so lower scores are better. I've found the SPECapc Lightwave 3D test to be an excellent indicator of overclock stability. In many cases, overclocked systems that will make it through every other benchmark here will crash in this test. 
Bear in mind that what this benchmark does is use scripts to control a stand-alone instance of Lightwave, so in that sense it's more indicative of real-world performance than the embedded Lightwave code in SPECviewperf. We again see very similar performance from the i7-2600K and the i7-980X, but the 3960X still wins decisively in all three tests. BlenderBlender is an open-source, free content creation suite of 3D modeling, rendering, and animation capabilities. Originally released in 2002, it's available in versions for Mac OS X, Windows, Linux, and several Unix distributions. It supports rigid and soft-body objects and can handle the draping and animation of cloth, as well as the rendering and animation of smoke, water, and general particle handling. Our Blender test renders multiple frames of an animation of a rotating chunk of ice, with translucency and reflections. Rendering of this model uses ray-tracing algorithms and the program reports the rendering time for each of the animation's 25 frames. The results are a summation of the rendering times for all frames and the lower the score, the better. 
Blender is limited to a maximum of 8 threads, so it's interesting to see that the Sandy Bridge Extreme CPU still performs better than the standard Sandy Bridge. Perhaps the extra cache helps. POV-RayThe Persistence of Vision ray tracer is a free, open source 3D modeling program that uses ray-tracing algorithms to generate realistic three dimensional images. Ray tracing is very computationally intensive, and the POV-Ray program has a handy built-in benchmark to let you check the performance of your system. AMD wins this round, posting stock-clocked results that are 26% better than Intel, although the FX-8150's lead narrows to a mere percentage point when both processors are overclocked.
Another i7-3960X win wraps up the performance testing portion of this review. Core i7-3960X OverclockingBenchmark Reviews has always been a fan of overclocking, with its historical roots of getting better performance from budget components. It might seem silly (and, frankly, kinda is) to overclock Intel's $990 Core i7-3960X CPU...but more performance is always better, right? The non-Extreme Sandy Bridge CPUs limit your overclocking in a number of ways: first, since the base clock is used to derive most of the other clocks in the system, to overclock by increasing the base clock frequency is virtually impossible, since increases of more than a few MHz will render the entire computer unstable. This leaves turbo ratio overclocking as the only real mechanism, and while all Sandy Bridge processors can increase their stock turbo multipliers slightly, only the unlocked "K-series" CPUs allow enough headroom for significant performance increases. And you must ensure that the chipset on the motherboard you're running allows this: putting the latest Core i7-2700K in an H67 motherboard will result in a functional computer, but one without any overclocking ability, which is limited to the P67 and Z68 chipsets. The X79 chipset brings a separate base clock back into the overclocking world, so this time-honored mechanism is again available. The CPU itself adds a new twist: in addition to increasing the voltage to the chip to support higher speeds, you can now increase the sustained maximum and burst maximum current draw as well. You should consider these overclocking results as preliminary: time pressure and the fact that the only cooler available was Intel's Asetek-sourced liquid cooler (which, while certainly competent, isn't a super-high-end cooler) means that we'll probably see better results in the future. Still, I achieved a solid 4.8GHz clock, sustainable on all cores under load, simply by increasing the turbo multiplier and tweaking ASUS' advanced digital power system on their new TUF Sabertooth X79 motherboard. This motherboard's active VRM cooling helped, too. The chart below shows how overclocking affected the 3960X's performance on each benchmark, with the stock clocked benchmark score normalized to 1.0 and the overclocked benchmark score represented as how much faster it was than the base clocked benchmark.
The 13% overall performance improvement is less than I'd hoped for, for a 900MHz overclock, but bear in mind that not all benchmarks used all cores, and indeed didn't necessarily load the cores enough to elicit the maximum turbo ratio. In any case, the 3960X is a performance beast even at its stock settings. Sandy Bridge Extreme Final ThoughtsI think this is the first time I've tested a new processor that so decisively dominated the benchmarks. There's no doubt about it: the Core i7-3960X Extreme processor is a benchmark-smashing beast of a CPU, leveraging Intel's brilliant Sandy Bridge core architecture and sweetened with more cache and even more cores. Intel's "Extreme" series CPUs have always represented the best the company can produce, and it's instructive to look at the stock-clocked performance of this new CPU as compared to its Extreme series predecessor, the Core i7-980X. As with the overclocking chart in the previous section, the scores below show the base score normalized to 1.0, with the comparison score expressed in terms of this.
With a solid 29% average score improvement over the 980X on this mix of benchmarks, the 3960X represents one of, if not the most, significant performance leaps over a previous-generation CPU I've ever seen. There's just so much performance available that even the most avid enthusiast will have to ask themselves if they have any realistic use for it. Intel is, wisely, aiming at multimedia and content creators as the principal audience for this chip, since these are the classes of applications that can best use all the threads this CPU can handle. If you work with applications that can keep the 3960X's capacious maw full of data, you'll see amazing performance that will easily justify the cost of a new system in increased productivity.
There are other advantages inherent in the platform: many X79 motherboards have 8 DIMM sockets, which means that you can install up to 64 GB of RAM in your desktop system. For some applications, this will be a tremendous advantage, as will the 48 PCI-E lanes. For most of us, though, the Core i7-3960X makes as much sense as a 1,000 horsepower street car. It's fun at the dragstrip (running benchmarks), but a Core i7-2700K on a nice Z68 motherboard will be a fraction of the price and just as fast in "real world" terms. Intel Core i7-3960X ConclusionBenchmark tests should always be taken with a grain of salt. It's difficult to try and isolate the performance difference a single component in a computer system makes, especially when it's necessary to compare across different manufacturers and platforms. Complicating the matter is the fact that benchmarks change, a manufacturer may change the technical details of a product, and the retail price may change as well. So please use this review as just one piece of information, and do your research before making a buying decision.
We've known Sandy Bridge Extreme has been coming for some months now, and we even knew pretty much what it would be: a six-core CPU based on the Sandy Bridge core architecture. Things like the gigantic 15M cache and expanded overclocking options are nice surprises, and at this point we can only wonder what Intel has in store for the forthcoming 22nm "Ivy Bridge" CPUs. Industry buzz has it that Ivy Bridge CPUs will drop right into existing Socket 1155 motherboards with only a BIOS upgrade needed, and the 22nm fabrication process may unlock even more performance. Costing $1049.99 at Newegg, the top-end Sandy Bridge Extreme CPU is the most expensive consumer CPU (high end Xeon server processors can cost thousands of dollars each). Add the $300+ cost of an X79 motherboard to put it in, as well as a new cooler (remember that Intel doesn't package a cooler with this CPU) and probably some more memory, and the cost of upgrading your existing X58 system could easily run to $1500 or more. That's a lot of money, especially in this economy, but as I mentioned before, for some segments of the market, it's a no-brainer. Assume that 29% average performance increase over a 980X holds true in a production environment: suddenly, rendering work that took, say, four hours now takes less than three hours. That's a huge win. This makes the value proposition hard to assess: even the most avid gamer's super rig won't generate noticeably more frames per second than it would with a Core i7-2600K, but for a professional graphics designer who spends hours rendering, it's cheap at twice the price. But at the end of the day, regardless of how much sense this CPU might or might not make in your application, the Core i7-3960X firmly establishes Intel's dominance in performance computing at the consumer level. It's simply an amazing product. Pros:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Comments
"For most of us, though, the Core i7-3960X makes as much sense as a 1,000 horsepower street car. It's fun at the dragstrip (running benchmarks), but a Core i7-2700K on a nice Z68 motherboard will be a fraction of the price and just as fast in "real world" terms."
This is a new era in power that will remain almost completely unused and unnecessary for most people, like you say, even us enthusiasts will need to think twice, except those who have money to needlessly burn.
You would think in these economic times, R&D of high dollar items like this would slow down. Interesting. At the same time, my i7 920 cost 1000.00 back in 2008, so the price to power ratio has dramatically fallen too.
But rich fanbois have big egos to satisfy so there will always be market for for stuff priced way out of line with what it is really worth. Useless - if I'm paying $1000 for a CPU it better be a multi CPU server chip in a render farm - something where you can get a return on your money at least. just my 2 cents for an alternate perspective.
I think petty envy speaks always more to one's ego state. Anyone's needs that are "really satisfied" by todays computing power will probably have a "cute device" for their "cute needs" soon enuff. and good riddance.
My needs are far from even close to being satisfied... though cinebench over 12 is at least progressing ( not enough knowing the Ivy Extreme will be released in 2012. And hopefully that benchmark gives more distance )
Catch 22 is really depressing! There is an army of artists whose talent could bring yer boss machine to it's knees given the chance. Instead they compromise to console greed economics. And instead of buying hardware that can push the boundries a market without that entertainment actually believes that we are at "good enuff".
Thanks for the consolitic ball and chain!
Buy it and they will make it/ Make it and they will buy it.
What do "most" users do with their computers that does not desperately need more computing power? Don't they have an app for that yet?
Also, if by FPS you're referring to frames per second, I suggest you do a bit MORE reading here at Benchmark Reviews... so you'll learn that the CPU has very little to do with frame rates.
Look on the bright side: now that Intel's Core i7-3960X has significantly raised expectations, you might soon find AMD processors selling in the blue-light section over at K-Mart. I understand they fired their entire marketing staff of 1400 workers after Bulldozer launched, so the only people saying anything good about them these days are the fan boys, and even they're unhappy.
You're right about frame rates with some games, and I over-simplified my statement. Battlefield 3 is one such game, and DirectX-11 effects drive all processor cores to nearly 50% usage even when played with a GTX 590. Obviously, there are times when the CPU will make a difference in game performance, but that requires advanced technology at a time when most developers are keeping to console (a la Modern Warefare 3).
Amatures ... This David R guy sounds like he has it together though ...
This was a very comprehensive review. It left me with no questions about the comparison of power.
FSX: enough said.
For example, my 2600K at 4.9Ghz + water can run the SPECapc rendering suite in 443 seconds. From your chart in the review it shows an overclocked 3930X does it in 408 seconds.
Simply put, that would be a 10% improvement from what I have now for a 300% price tag. :-)
I just wanted to say the fact you include this test in your reviews is very helpful to me, so thanks again. :-)
So yes, you pay 300% more, but that is a one time cost compared to getting that 10% every single time you run a model.
I believe the 3d ivybridge will be a better bargain than these.
Idont know if ivybridge extreme will be compatible with x79 mobo but P67 mobos will all for ivybridge upgrades making the 2600k chips a better deal from the getgo.
If you're asking about six core non-extreme Sandy Bridges, well, there aren't going to be any.
The NVIDIA Maximus announced release sounds pretty sexy ( quadro and tesla sandwich please!). Kind of makes me wonder if the mercury engine might deserve a second look.
Would appreciate a review if could score a copy.
geforce.com/Active/en_US/shared/images/articles/x79/3way.jpg
Still will probably wait for ivy bridge - E as the cinebench numbers do not justify dropping $1500 for x79 porn. But if the 3 way scaling is even close to the truth then I am very happy that the IVY Bridge solution will most likely do as much.
Makes waiting a few more seasons for my new dragster pretty hard though.
I must say I am very satisfied!