Update (9/3/2013): The following story was originally published on July 16th, 2013. After receiving a second Core i7-4960X sample, we validated the correctness of our numbers and updated the piece with official information from Intel.
Recently, Gartner published numbers showing that shipments of PCs dropped a staggering 11 percent in the second quarter of this year, primarily attributed to tablets replacing entry-level machines. Wall Street, at least, is all doom and gloom about the PC’s future prospects.
But the boutique builders I talk to say that interest in super-fast gaming systems is at an all-time high thanks to the efficiency of certain processor and graphics architectures. So, while now might be a bad time to get stoked about mainstream hardware, performance-oriented power users have some pretty quick components to choose from.
None of this is news to enthusiasts. In fact, two and a half years ago, Intel’s Sandy Bridge architecture was serving up compelling performance under 100 W. Those were exciting times. The Ivy Bridge architecture that followed nudged our benchmark results forward a bit, but dropped power to less than 77 W. That was pretty cool, too. More recently, Haswell added another few percentage points to the performance picture, but bumped maximum consumption back up to 84 W.
Now, if you’re upgrading an old Core 2- or Phenom II-based machine with a $5000 boutique build, the latest parts are going to feel wicked-fast, no matter how incremental the previous two or three generations look on paper. The difference is simply less perceptible to those of us working with these components day in and out.
The point is that, for a do-it-yourselfer like me, Sandy Bridge was exciting, Ivy Bridge a little less so, and Haswell…well, I called that one The Core i7-4770K Review: Haswell Is Faster; Desktop Enthusiasts Yawn.
We all know where Intel’s collective mind is: the mobile space where those Gartner guys are telling us the low-end PCs continue getting slaughtered. In that context, spending $350 on a -4770K and another $250 on an LGA 1150-capable motherboard just to keep up with the Kardashians doesn’t sound so hot.
If, a year and a half ago, you snagged a Core i7-3930K (which won a very rare Best Of award from us in Intel Core i7-3930K And Core i7-3820: Sandy Bridge-E, Cheaper), you’d still be sitting pretty, potentially overclocked to 4.5 or 4.6 GHz, and outclassing the -4770K in a great many threaded applications. You’d also be using the same X79-based platform. And, with the revelation that Intel’s upcoming Ivy Bridge-E architecture will drop into an LGA 2011 interface, you’re also going to face your first opportunity in two years to buy something faster.
Meet Ivy Bridge-E, The Upgrade Path For X79 Express
From most angles, the Ivy Bridge-E-based parts look a lot like Sandy Bridge-E, except for the adoption of Intel’s Ivy Bridge architecture. That means a handful of IPC-oriented improvements in the core, cache, and memory controller, similar to what we described in Intel Core i7-3770K Review: A Small Step Up For Ivy Bridge. Of course, gone is the emphasis on graphics. That means Ivy Bridge-E is really about the updated core, a memory controller rated for 1866 MT/s (instead of 1600), official PCI Express 3.0 compliance (remember, Sandy Bridge-E only claimed 8 GT/s signaling support), and 22 nm manufacturing. Ivy Bridge-E-based CPUs are also unlocked up to 63x multipliers (versus SNB-E's 57x), you should be able to hit memory data rates beyond 2400 MT/s, Ivy Bridge-E supports XMP 1.3 (compared to SNB-E's XMP 1.2), and you'll have access to real-time ratio, voltage, and power limit settings.
Intel's Ivy Bridge-E die; six CPU cores are clearly visible, along with shared L3 cache and the memory controller up top
You still get 40 lanes of PCI Express connectivity, divisible into as many ports as you need for four-way CrossFire and SLI. You’re still dealing with a quad-channel memory controller, though the higher data rate increases peak bandwidth to 59.7 GB/s from 51.2 GB/s. And you’re dropping Ivy Bridge-E into the aging X79 Express platform. The good news is that your old motherboard still works; you don’t have to buy a new one. Unfortunately, the chipset only offers two SATA 6Gb/s ports, it doesn’t feature native USB 3.0, and you don’t get to enjoy new capabilities like SATA Express, which is expected to surface alongside Haswell-based 9-series chipsets early in 2014.
| Core i7-4960X | Core i7-4930K | Core i7-4820K | Core i7-3970X | |
|---|---|---|---|---|
| Code Name | Ivy Bridge-E | Ivy Bridge-E | Ivy Bridge-E | Sandy Bridge-E |
| Base Clock Rate | 3.6 GHz | 3.4 GHz | 3.7 GHz | 3.5 GHz |
| Maximum Turbo Boost | 4 GHz | 3.9 GHz | 3.9 GHz | 4 GHz |
| PCI Express Link Speed | 8 GT/s | 8 GT/s | 8 GT/s | 8 GT/s |
| TDP | 130 W | 130 W | 130 W | 150 W |
| Processor Cores | 6 | 6 | 4 | 6 |
| Shared L3 Cache | 15 MB | 12 MB | 10 MB | 15 MB |
| Max. Memory Data Rate | DDR3-1866 | DDR3-1866 | DDR3-1866 | DDR3-1600 |
| Processor Interface | LGA 2011 | LGA 2011 | LGA 2011 | LGA 2011 |
| Price | $990 | $555 | $310 | $1020 (Street) |
Again, Core i7-4960X is a six-core part with 15 MB of shared L3 cache. No doubt that’ll disappoint the folks who were hoping a 22 nm process would make it easier for Intel to arm enthusiasts with eight or 12 cores (it actually does, as we saw in Intel's 12-Core Xeon With 30 MB Of L3: The New Mac Pro's CPU?). But, at the same ~$1000 price point, there’s really no reason to give you a more complex CPU when it’s already charging $1900 for an eight-core Xeon E5-2687W. And so, anyone considering a move from today’s Core i7-3970X can expect an additional 100 MHz base frequency, the same 4 GHz peak Turbo Boost clock rate, and the other incremental improvements.
The Core i7-4820K is a little more interesting. Realizing that there was almost no reason at all anyone would want a multiplier-locked, quad-core -3820, Intel gives its successor an unlocked ratio. It’s still a quad-core chip with 10 MB of shared L3 cache based on a previous-gen architecture in a previous-gen platform. But perhaps the additional PCI Express connectivity, memory bandwidth, and L3 cache, coupled with the ability to overclock, makes the -4820K competitive against Intel’s Haswell-based Core i7-4770K.
And then there’s the Core i7-4930K, which retains six cores, sheds a little of its shared L3 cache (dropping to 12 MB), and drops a little frequency (200 MHz base clock and 100 MHz peak Turbo Boost), but also costs close to half of what you’d pay for the flagship. That’s the model we were most excited about last generation. We reserve some of that excitement today, realizing that enthusiasts who bought a -3930K probably won’t step up to a -4930K for another $500+ dollars.
Curiously, all three Ivy Bridge-E-based parts are accompanied by 130 W thermal design power limits. Remember that the move from Sandy to Ivy Bridge yielded a more complex CPU with a significantly lower power ceiling, thanks in no small part to a shift from 32 to 22 nm manufacturing. Here, we have the same process transition. The die shrinks from Sandy Bridge-E's 434 square millimeters down to 257. Intel even cites a lower transistor count for Ivy Bridge-E (1.86 billion versus SNB-E's 2.27 billion). And yet, this new CPU shares the same 130 W rating as Core i7-3960X and -3930K. Keep an eye on this. Power may just become Ivy Bridge-E’s greatest strength.
| Test Hardware | |
|---|---|
| Processors | Intel Core i7-4960X (Ivy Bridge-E) 3.6 GHz (36 * 100 MHz), LGA 2011, 15 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled |
| Intel Core i7-4770K (Haswell) 3.5 GHz (35 * 100 MHz), LGA 1150, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i7-3770K (Ivy Bridge) 3.5 GHz (35 * 100 MHz), LGA 1155, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i7-2700K (Sandy Bridge) 3.5 GHz (35 * 100 MHz), LGA 1155, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i7-3970X (Sandy Bridge-E) 3.5 GHz (35 * 100 MHz), LGA 2011, 15 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i7-3930K (Sandy Bridge-E) 3.2 GHz (32 * 100 MHz), LGA 2011, 12 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| AMD FX-8350 (Vishera) 4.0 GHz (20 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled | |
| AMD A10-5800K (Trinity) 3.8 GHz (19 * 200 MHz), Socket FM2, 4 MB Total L2 Cache, Turbo Core enabled, Power-savings enabled | |
| Motherboard | MSI Z87 Mpower Max (LGA 1150) Intel Z87 Express, BIOS 1.2B1 |
| MSI Z77 Mpower (LGA 1155) Intel Z77 Express, BIOS 17.8 | |
| MSI X79A-GD45 Plus (LGA 2011) Intel X79 Express, BIOS 17.2 | |
| MSI 990FXA-GD80 (Socket AM3+) AMD 990FX/SB950, BIOS 13.2 | |
| MSI FM2-A85XA-G65 (Socket FM2) AMD A85X, BIOS 2.0 | |
| Memory | G.Skill 16 GB (4 x 4 GB) DDR3-1600, F3-12800CL9Q2-32GBZL @ DDR3-1600 at 1.5 V |
| Hard Drive | Samsung 840 Pro 256 GB, SATA 6 Gb/s |
| Graphics | Nvidia GeForce GTX Titan 6 GB |
| Power Supply | Corsair AX860i, 80 PLUS Platinum, 860 W |
| System Software And Drivers | |
| Operating System | Windows 8 Professional x64 |
| DirectX | DirectX 11 |
| Graphics Driver | Nvidia GeForce Release 320.18 |
Existing X79-based motherboards require a firmware update to support the processor. We had to seek this out specifically, but cannot say which platform was used for benchmarking the Ivy Bridge-E-based CPU.
| Benchmark Configuration | |
|---|---|
| Adobe Creative Suite | |
| Adobe After Effects CS6 | Version 11.0.0.378 x64: Create Video which includes three Streams, 210 Frames, Render Multiple Frames Simultaneously |
| Adobe Photoshop CS6 | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Adobe Premeire Pro CS6 | Version 6.0.0.0, 6.61 GB MXF Project to H.264 to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| Audio/Video Encoding | |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| Lame MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| HandBrake CLI | Version: 0.98: Video from Canon Eos 7D (1920x1080, 25 FPS) 1 Minutes 22 Seconds Audio: PCM-S16, 48,000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677: MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, 2 Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Adobe Acrobat X | Version 10.0.0.396: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encryption |
| Autodesk 3ds Max 2012 and 2013 | Version 14.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080 |
| Blender | Version: 2.64a, Cycles Engine, Syntax blender -b thg.blend -f 1, 1920x1080, 8x Anti-Aliasing, Render THG.blend frame 1 |
| Visual Studio 2010 | Version 10.0, Compile Google Chrome, Scripted |
| File Compression | |
| WinZip | Version 17.0 Pro: THG-Workload (1.3 GB) to ZIP, command line switches "-a -ez -p -r" |
| WinRAR | Version 4.2: THG-Workload (1.3 GB) to RAR, command line switches "winrar a -r -m3" |
| 7-Zip | Version 9.28: THG-Workload (1.3 GB) to .7z, command line switches "a -t7z -r -m0=LZMA2 -mx=5" |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.1.0, Benchmark Only |
| SiSoftware Sandra 2013 | Version 2013.01.19.11, CPU Test = CPU Arithmetic / Multimedia / Cryptography / Memory Bandwidth / Cache Bandwidth |

Using the same GeForce GTX Titan as our Haswell launch coverage, we see that Ivy Bridge-E doesn’t do anything for single-card graphics performance in 3DMark 11 (which is what we’d expect, given that both platforms yield a full 16 lanes at 8 GT/s).
In contrast, the processor-bound Physics module demonstrates a small bump in favor of the Core i7-4960X over -3970X. More pronounced is the -4960X’s 30%+ improvement over Core i7-4770K.

There’s very little gain over the Sandy Bridge-E flagship in SiSoftware’s Sandra Arithmetic sub-test.

The same goes for the Multimedia benchmark. In fact, Core i7-4770K yields better numbers in the integer component thanks to its AVX 2 support.

It’s possible that we could get more memory bandwidth from Core i7-4960X using a quad-channel DDR3-1866 memory kit. However, we only had access to 1600 MT/s for this story, so we used the same G.Skill kit from our Core i7-4770K launch piece. We already know this platform isn’t particularly bandwidth-constrained on the desktop, though, so we don’t expect any real-world benefit beyond this 41 GB/s mark.

When we sort by L1 cache throughput, the Haswell architecture’s doubled theoretical max yields almost 1 TB/s, while Ivy Bridge-E ducks in under 800 GB/s. On paper, Haswell should also push twice as much L2 bandwidth as well. We haven’t observed this yet, though. In contrast, Core i7-4960X, sporting six cores with 256 KB of L2 each, pushes more aggregate bandwidth, nearly hitting 500 GB/s. The extra cores also help with shared L3 bandwidth, given more stops along the ring bus.

We use two distinct Photoshop benchmarks, one of which fully taxes each processor’s x86 cores using well-threaded filters, and another that is OpenCL-optimized to leverage graphics resources. Don’t compare the black and red bars above—they’re only together to save space (and your scrolling finger).
Core i7-4960X pretty much ties the Sandy Bridge-E-based -3970X in our CPU-based metric. You'd assume that Ivy Bridge-E would have an advantage, right? It turns out that we also discovered that Core i7-3770K wasn't any faster than -2700K in this test more than a year ago. On the other hand, all of Intel’s six-core CPUs outmaneuver the company’s quad-core offerings.
The OpenCL-accelerated metric is less consistent, seeming to favor architecture over core count or clock rate. For instance, the Haswell-based Core i7-4770K places first, followed by the Ivy Bridge-based Core i7-3770K and -4960X. Four- and six-core Sandy Bridge-based processors group up in a third clump, while AMD’s CPUs fall pretty far back.

The architectural benefit of adopting Ivy Bridge helps Intel’s Core i7-4960X claim a first-place finish, though nobody’s going to upgrade from Sandy Bridge-E for a 5% speed-up. A more demanding render project might create a larger delta between CPUs, and we’re working on sourcing a real-world project to test this theory.

After Effects is a special case where adding cores doesn’t always help performance as the available memory per core shrinks. That’s why you see the Core i7-3970X and -3930K in the middle of the chart. Between its 16 GB of DDR3-1600 and the Ivy Bridge architecture, though, Intel’s Core i7-4960X manages to tie the first-place -4770K.

Both of our 3ds Max benchmarks favor Intel’s hexa-core processors, and the Ivy Bridge architecture is enough to earn Core i7-4960X the top spot yet again. Core i7-3970X isn’t far behind though, and it’s not hard to imagine an overclocked Core i7-3930K topping both flagship chips.

There’s a fairly consistent pattern in play. Threaded workloads that would have favored Sandy Bridge-E are just a few percent faster on Ivy Bridge-E. In this case, we’re looking at about a 5% speed-up. All three LGA 2011-based CPUs are quite a bit quicker than the LGA 1150/1155-based models.

Based on Maxon’s Cinema 4D software, our scripted Cinebench test measure single- and multi-core processor performance.
Clearly, the threaded component of this benchmark favors Ivy Bridge-E, as Intel’s Core i7-4960X turns in a score 4% higher than the -3970X. And although Ivy Bridge-E also demonstrates an advantage over Sandy Bridge-E in the single-core metric, Haswell demonstrates the top result when we isolate one thread.

Optimizations for threading in ABBYY’s optical character recognition application set all three hexa-core processors apart from the rest of the field. Within that elite group, the Core i7-4960X edges out its predecessor by just two seconds. We’re talking about low single-digit percent differences.

As you might expect, once you subject Ivy Bridge-E to a single-threaded workload, it falls behind Haswell, which boasts greater IPC throughput. Instead, Core i7-4960X is on-par with Core i7-3770K—both based on the Ivy Bridge architecture.

Our Google Chrome compile workload, on the other hand, does exploit whatever compute resources it can, and so the Core i7-4960X edges out its predecessor by a hair. The quad-core -4770K finishes several minutes after IVB-E.

Fritz isn’t really a productivity app (unless you consider playing chess productive), but we’re putting it here anyway. The results from each processor are reflected in kilonodes per second. A node is a position on the chessboard. So, in the case of Core i7-4960X, Fritz evaluates nearly 20,000 thousand nodes per second, or 20 million. If you give the engine enough time to “think”, you’re going to get a pretty competitive computer opponent. Hope you brought your A-game.

Poor scaling in WinRAR yields unimpressive results. Most of the Intel processors clump up, clearly not affected by core count, clock rate, or architecture.

Better-optimized for multi-core processors, the three six-core CPUs stand out in our 7-Zip benchmark, the -4960X barely leading the pack.

Our WinZip chart includes several results, since we first test using the CPU cores, and then follow that up by enabling OpenCL acceleration to offload some of the work. Of course, we know from talks with Corel that the GPU only kicks in on files larger than 8 MB. Because our 1.3 GB archive is a mix of different sizes and types, only some of this benchmark is aided by turning on OpenCL.
The longest bar, in black, represents maximum compression, also performed on the CPU. That’s the one we’re sorting by, and the Core i7-4960X takes a first-place finish. The less-taxing processor-based test, in red, is won by Intel’s Haswell-based Core i7-4770K and followed by Core i7-4960X. OpenCL acceleration throws the numbers off slightly, favoring Haswell first, Ivy Bridge second, and Ivy Bridge-E third, though Intel’s processors all fall within a fairly tight margin.

Most of the video transcoding apps we use to test with are well-threaded, and TotalCode Studio (formerly MainConcept) is no exception. The Core i7-4960X takes the lead, followed by both Sandy Bridge-E-based chips. The quad-core competition trails behind as Core i7-4770K occupies fourth place.

The same applies to HandBrake, though Haswell isn’t quite as far behind Core i7-3930K in this benchmark.

iTunes (above) and LAME (below) are different in that they’re both single-threaded, which is why Haswell takes first place. Core i7-3970X and -4960X trade blows. We’d actually expect Ivy Bridge-E’s IPC advantage to carry it ahead more definitively, but that just doesn’t happen.
Regardless, these last two benchmarks gently make the point that the upcoming Ivy Bridge-E-based processors excel in well-threaded workloads and trail slightly to Haswell in lighter tasks.

Alright, so, overall I have to admit that those benchmarks were pretty boring. I sort of thought that might turn out to be the case when I heard that Ivy Bridge-E would be roughly the same six-core configuration, updated with a lightly-tuned architecture.
But, like all of the other processor reviews I work on, I made it a point to log power consumption as our scripted suite ran. I didn’t expect the results to be particularly noteworthy—after all, Intel is saying that Core i7-4960X has the same 130 W TDP as the Core i7-3930K.

In the chart above, data points are recorded every two seconds, and the end of the run is truncated to fit as much information as possible in the space available. Regardless of where each CPU seems to finish the complete suite, 30 minutes of idle time are tacked onto the end before our script automatically shuts the systems down. As a result, the average and total efficiency measurements include a long period of time where absolutely nothing is happening.
Without question, Core i7-4960X is more power-friendly than the 150 W Core i7-3970X, seen in green. But even the Core i7-3930K (in yellow) appears to register higher energy use during our suite.
To get a better idea of what the line graph really means, we average each processor’s results from the time we start the test until our log shows zero power use.

The system averages for each setup fall almost exactly where they should. The 77 and 84 W Ivy Bridge and Haswell CPUs drive the machines averaging the least power consumption. The 95 W Core i7-2700K and 100 W A10-5800K place third and fourth. Intel’s 130 W Core i7-3930K and -4960X take the next two spots, followed by AMD’s 125 W FX-8350 (which should probably be in front of the LGA 2011-based parts). Finally, Core i7-3970X lands in the back, swinging its sweaty 150 W TDP around.
Of course, the -3970X uses that big power budget to get things done faster. Let’s multiply the time it takes to cruise through our suite by average power use to give us efficiency in Wh.

Crazy, right? We already know that the Core i7-3970X is fast. But it needs so much power to get there that the Sandy Bridge-E processor isn’t very efficient in the process. Only AMD’s FX-8350 and A10-5800K use more energy getting through our benchmark suite.
The average consumption numbers showed us that Intel’s Core i7-3930K uses a lot less power than the flagship model, but is so much slower as a result of its cut-back shared L3 cache and lower frequency that it, too, ends up less efficient than Ivy Bridge-E. Even the Core i7-2700K shows up behind the Core i7-4960X.
It’d be almost impossible for the six-core -4960X to outperform Intel’s latest quad-core parts decisively enough to beat them in an efficiency race. But despite the single-threaded apps and half-hour idle period added to our log, Ivy Bridge-E does remarkably well.
Does Intel’s Core i7-4960X, specifically, get me all revved up about upgrading? Well, no. Not really. But then again, those thousand-dollar CPUs rarely do. What about the Core i7-4930K replacing Intel’s -3930K for $550? That’d be a tough sell for all of the same reasons. Mainly, it doesn’t push performance high enough to warrant a big price tag. Any interest in a Core i7-4820K? I’d be more inclined to bet on a -4770K/Z87 platform, if only for the newer chipset’s extra functionality.
As far back as April of last year we knew that Core i7-3770K was somewhere between zero and seven percent faster than Core i7-2700K, depending on the workload. Is it really a surprise that Ivy Bridge-E would only be a few percent faster than Sandy Bridge-E? At least on the performance end, Core i7-4960X is close to what we might have expected.
Still, it would be cool to see Intel configure Core i7s with 10 or 12 cores, like some of the planned Ivy Bridge-EP models. Instead, it looks like we’ll be waiting for Haswell-E to see the first eight-core enthusiast-oriented processors. As a result, Intel is really limiting the appeal of Ivy Bridge-E to power users building or buying brand new PCs. The bummer there is the two-year-old platform with two SATA 6Gb/s ports and no native USB 3.0. That’s hardly going to get an enthusiast worked up when Z87 is so much more fully featured. If you already own a Core i7-3960X or -3930K, you’re simply not going to sink another big chunk into single-digit percent gains.

Should Ivy Bridge-E fail to encourage upgrades or new system builds, I know who’s going to absolutely love this new architecture: the server and workstation segments. For what little gets added to performance, Ivy Bridge-E does some crazy-awesome things to power and efficiency. When you multiply out the gains across a rack, you’re looking at a lot less power, a lot less heat, and a lot less cooling.
Consider this a parting shot: Core i7-4960X is faster than Core i7-3970X and simultaneously about 30% more efficient. In the world of Xeon E5-2x00 v2 processors, that’s going to be killer. Want some proof? Go check out Intel's 12-Core Xeon With 30 MB Of L3: The New Mac Pro's CPU?
