According to Mick Jagger, it’s lonely at the top. Intel might agree. After all, for the past five years, the company has put an increasingly large gap between its fastest desktop processors and AMD’s own best efforts. Enthusiasts tend to lament the fact that a lack of intense competition means they pay more for high-end hardware. But, if you’ve been around long enough, you know that Intel’s Extreme Edition CPUs were always thousand-dollar affairs and, once upon a time, AMD’s vaunted FX-series chips used to be worth their $700+ asking prices.
The fact that the $1000 price point persists today, eight years later, means Intel recognizes the extremely limited market for these flagship desktop processors and isn’t about to push one of its crown jewels even further out of reach.
It comes as little surprise, then, to see yet another Extreme Edition processor hovering around $1000. But this behemoth is very different than what came before.
The prior generation of flagship parts based on Gulftown came armed with six physical cores and up to 12 MB of shared L3 cache. They boasted LGA 1366 compatibility, extending the useful lives of pricey X58 Express motherboards, helping soften the blow of $500+ processor upgrades. No such luck this time; you’re facing a pricier investment.
Sandy Bridge-E, Gulftown’s successor, employs an LGA 2011 interface, requiring new motherboards based on Intel’s X79 Express Platform Controller Hub. It also comes armed with an integrated quad-channel memory controller, necessitating four-module memory kits. Oh, and then there’s the fact that Intel isn’t planning to bundle its new chips with coolers, requiring a separate purchase there, too.
Meet Sandy Bridge-E
Intel is announcing three Sandy Bridge-E-based models today, but only two will be available through the end of 2011: Core i7-3960X and Core i7-3930K. The third, Core i7-3820, is slated for a Q1 2012 introduction.
Gulftown is big; Sandy Bridge-E is much bigger
2011 pins require a lot of space
All three employ the same die, which is composed of 2.27 billion transistors and measures 434 square millimeters (making it a very big chip). In comparison, quad-core Sandy Bridge parts are made up of 995 million transistors and measure 216 square millimeters, while six-core Gulftown CPUs incorporate more than 1.1 billion transistors in a 248 square millimeter die.
Of course, Sandy Bridge-E was never intended to be a desktop processor exclusively. Rather, it’s going to emerge in the first part of next year as Xeon E5 for single- and dual-socket servers/workstations. In that context, the CPU’s size and complexity makes more sense. After all, Westmere-EX (at the heart of Intel’s more enterprise-oriented Xeon E7 family) is a 2.6 billion-transistor die occupying 513 square millimeters of space.
Core i7-3960X, with two cores and 5 MB L3 cache disabled
When Sandy Bridge-E surfaces as Xeon, it’ll offer up to eight processing cores and 20 MB of shared L3 cache. As a desktop CPU, however, it’s limited to as many as six cores and up to 15 MB of shared L3. Intel achieves this by disabling two cores and four of the die’s 16 slices of shared L3 cache.
Of course, that configuration only applies to Core i7-3960X. Core i7-3930K, which also features six cores, dips down to 12 MB of cache, revealing Intel’s ability to very granularly disable pieces of the shared L3 to suit its needs. The upcoming Core i7-3820 will employ four cores and 10 MB of shared L3 cache—essentially half of a Sandy Bridge-E die. Each core includes 32 KB of L1 instruction and L1 data cache, plus a dedicated 256 KB L2 cache.
| Sandy Bridge-E Family | |||||||
|---|---|---|---|---|---|---|---|
| Base Clock | Max. Turbo | Cores / Threads | L3 Cache | TDP | Memory | Price | |
| Core i7-3960X | 3.3 GHz | 3.9 GHz | 6 /12 | 15 MB | 130 W | 4-Channel DDR3-1600 | $990 |
| Core i7-3930K | 3.2 GHz | 3.8 GHz | 6 / 12 | 12 MB | 130 W | 4-Channel DDR3-1600 | $555 |
| Core i7-3820 | 3.6 GHz | 3.9 GHz | 4 / 8 | 10 MB | 130 W | 4-Channel DDR3-1600 | TBD |
The clocks on all three SKUs range up and down as well. The -3960X starts at 3.3 GHz and, through the same second-gen Turbo Boost technology introduced with Sandy Bridge, speeds up to 3.9 GHz. The -3930K starts at 3.2 GHz and hits a peak of 3.8 GHz in lightly-threaded workloads. Finally, the -3820 will start at 3.6 GHz and reach frequencies of up to 3.9 GHz with Turbo Boost.
Of course, the X- and K-series chips are also multiplier-unlocked, making those stock clocks pretty much meaningless for most enthusiasts planning to tweak their systems. Intel calls the -3820 “partially unlocked.” In all actuality, it gets six 100 MHz bins above its maximum Turbo Boost setting of 3.9 GHz, translating to a ceiling of 45x.


Intel is using the same cores found in its Sandy Bridge-based CPUs. Turning off Turbo Boost, setting similar base clocks, and running a couple of single-threaded apps demonstrates the efficient execution Sandy Bridge brings to the table compared to Thuban or Zambezi.


Switching Turbo Boost back on and running Core i7-3960X in parallelized and single-threaded titles gives us a better impression of what that technology does for performance.
In an application like iTunes, which is only able to utilize one core, Turbo Boost improves performance by 12.8%. In 7-Zip (well-optimized to use available cores), it increases performance by 10.8%. The second number is surprisingly high because Turbo pushes an additional three 100 MHz bins when five or six cores are active and none of the technology’s triggers are tripped. As a result, it's tackling our compression workload at 3.6 GHz instead of 3.3.
Quad-Channel Memory Control
All three models sport the same quad-channel memory controller officially rated for data rates as high as 1600 MT/s. That’s up to 51.2 GB/s of bandwidth, in theory.
Although we used Intel’s DX79SI motherboard for all of the performance benchmarking in this piece, the platform had a hard time scaling beyond DDR3-1600 using any of the quad-channel kits in our lab. So, I swapped into an Asus Rampage IV Extreme that arrived a little later, and used a set of G.Skill Ripjaws Z F3-17000CL9-4GBZH modules rated for DDR3-2133 to test scaling.



Jumping from DDR3-800 to -1066 to -1333 and then -1600 yields impressive throughput gains, reflected in some solid performance improvements (WinRAR) and other, less notable speed-ups (3ds Max).
For now, my recommendation is to grab the lowest-latency quad-channel DDR3-1600 kit for which you’re willing to pay. Thomas is already working on our first high-end memory round-up aimed specifically at X79, though, which promises to shed more light on the sensibility of the kits with gobs of headroom for overclocking.
PCI Express 3.0
Sandy Bridge-E’s memory capabilities are decidedly more relevant in the server space. However, the design’s PCI Express connectivity might appeal to power users with an affinity for multi-card graphics subsystems.
One of the most controversial points in my Sandy Bridge-E preview was the fact that PCI Express 3.0 was stated "not supported" by the manufacturer of my test platform. Multiple outside sources indicated that third-gen PCIe was planned, but wouldn’t be available, possibly until 2012.
It turns out that PCI Express 3.0 is, in fact, supported by Sandy Bridge-E (and the preview was updated to confirm 8 GT/s support the day after it went live). But because there weren’t (and still aren’t) any third-gen devices available yet, validating the feature was problematic. In fact, as you can see in the image below, Intel is still only officially guaranteeing that PCI Express 2.0 works, and probably will continue to do so until we see some hardware with a third-gen interface.
Nevertheless, Intel’s Core i7 datasheet confirms PCI Express 3.0 compliance, enabling up to 1 GB/s of bandwidth per lane, per direction.

Forty lanes of connectivity are built into all three Sandy Bridge-E-based processors, divisible into as many as 10 different ports. Most motherboard vendors will probably configure their platforms for three or four graphics cards, resulting in two x16 slots and one x8 slot, or one x16 link and a trio of x8 connectors.
Compared to Sandy Bridge and its 16 lanes of second-gen PCIe on each CPU, Sandy Bridge-E’s 40 lanes are downright decadent. But don’t expect that improvement alone to yield better gaming performance with multiple GPUs installed. After all, we’ve already proven that you can get exceptional performance from three graphics cards with a three-part series on the matter. At least until we’re able to test with PCI Express 3.0-capable devices, there’s no reason to demand the connectivity Sandy Bridge-E offers when an LGA 1155 platform equipped with Nvidia’s NF200 bridge isn’t even starved for throughput.
Where Sandy Bridge-E could positively impact gaming performance is with its enhanced processing power. Nvidia says the CPU’s speed-up is able to maximize scaling with three of its GeForce cards installed, and we’ll be testing that claim shortly. PCI Express isn’t credited with the improvement, though. The six Sandy Bridge cores and big L3 cache get that honor.
The Patsburg chipset (code name for the silicon on which X79 Express centers), as Intel originally planned it, was to have as many as 14 ports of storage connectivity. Six of them were SATA-based (2 x 6 Gb/s and 4 x 3 Gb/s), while as many as eight emanated from a separate, integrated storage controller. In its most decked-out form, that controller would have offered eight SAS 6 Gb/s ports. It also would have borrowed four of the processor’s 40 third-gen PCI Express lanes to create a x4 link dedicated to storage traffic.
Apparently, we’ll still see that souped-up rendition of Patsburg in 2012. But the company either couldn’t or didn’t enable it in X79 Express, leaving the chipset with the same two 6 Gb/s and four 3 Gb/s SATA ports we’ve seen for almost a year on the P67 Express chipset.
When you start looking around at the rest of X79’s features, you realize that, while the platform bears a new name, it’s pretty much P67 Express. You get the same 14 USB 2.0 ports, the same integrated gigabit Ethernet MAC, eight lanes of second-gen PCI Express, and HD Audio.
Now, Intel does include a driver called Rapid Storage Technology Enterprise 3.0, which is designed to facilitate the additional data protection servers and workstations will need once the more advanced versions of Patsburg emerge. For all of our testing, though, Microsoft’s native AHCI drivers are fine.
As a result, all of the changes inherent to the Sandy Bridge-E/X79 platform, at least on the desktop, are attributable to the processor. We’ll have to wait until next year for a more advanced platform—which motherboard vendors don’t seem to be sure what to do with yet, by the way.
I’m frankly not too concerned, though. Do I really need SAS support? No. Do I even need more than six SATA ports? Not really. More than anything, it’s a shame that Intel wasn’t able to incorporate native USB 3.0 support.
Intel isn’t bundling its Sandy Bridge-E-based Core i7s with a cooling solution of any sort. Perhaps the company surmised that few power users willing to drop 10 bills on a flagship processor stick with boxed heat sinks. Instead, it’s selling a closed-loop water cooler and cost-optimized heat sink/fan combination separately, allowing enthusiasts to pick between two options from Intel or a host of other third-party products.
Intel expects the liquid propylene glycol-based cooler, sourced from Asetek, to sell for somewhere between $85 and $100. That part is compatible with Sandy Bridge-, Ivy Bridge-, Sandy Bridge-E-, and future Ivy Bridge-E-based processors. Several other third-party liquid coolers are expected to accommodate motherboards with the LGA 2011 interface, too.
Intel’s heat sink and fan combo should go for less than $20, though that part is aimed at government and business customers who might look to Sandy Bridge-E for an entry-level workstation. That’s not to say there won’t be plenty of high-end air coolers for LGA 2011, though. Noctua sent us its NH-D14 SE2011 for our upcoming holiday gift guide, and we were able to get all six cores stable at 4.6 GHz running Prime95 for an hour with it on the Intel DX79SI board.
Now, I don’t put much stock in overclocking results from processors sent out to reviewers. But I spent some time polling system builders who were preparing to sell overclocked Sandy Bridge-E machines at launch, and it seems like 4.5 or 4.6 GHz is a realistic target for enthusiast-oriented PCs that (hopefully) won’t fizzle out and die within a month.
Overclocking Sandy Bridge-E
There are only three Sandy Bridge-E-based processors at launch, and only two of them are expected to be available. That pair of chips is multiplier-unlocked, making overclocking a fairly simple matter of setting a ratio and fine-tuning with BCLK settings.


The third model, Core i7-3820, isn’t unlocked, though. Instead, it’s “partially unlocked.” Intel lets you set a multiplier six bins higher than the highest Turbo Boost clock rate: 3.9 GHz. The result is a maximum of 4.5 GHz using ratios exclusively.
Unlike the Sandy Bridge/Cougar Point combination, however, the clock generator for this platform is external, which makes it easier for you to make BCLK modifications without running afoul of buses more sensitive to deviation. In order to facilitate this, you get a handful of strap ratios that increase the BCLK without pushing subsystems like PCI Express out of spec. In short, though Core i7-3820 overclocking is technically not unlocked, a combination of extra multiplier settings and PCIe/DMI ratios should make it possible to find the quad-core chip’s limit fairly freely, too.
| Test Hardware | |
|---|---|
| Processors | Intel Core i7-3960X (Sandy Bridge-E) 3.3 GHz (33 * 100 MHz), LGA 2011, 15 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled |
| Intel Core i7-990X (Gulftown) 3.43 GHz (26 * 133 MHz), LGA 1366, 12 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| AMD FX-8150 (Zambezi) 3.6 GHz (18 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled | |
| AMD Phenom II X4 980 BE (Deneb) 3.7 GHz (18.5 * 200 MHz), Socket AM3, 6 MB Shared L3, Power-savings enabled | |
| AMD Phenom II X6 1100T (Thuban) 3.3 GHz (16.5 * 200 MHz), Socket AM3, 6 MB Shared L3, Turbo Core enabled, Power-savings enabled | |
| Intel Core i7-2600K (Sandy Bridge) 3.4 GHz (34 * 100 MHz), LGA 1155, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i5-2500K (Sandy Bridge) 3.3 GHz (33 * 100 MHz), LGA 1155, 6 MB Shared L3, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i7-920 (Bloomfield) 2.66 GHz (20 * 133 MHz), LGA 1366, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Motherboard | Intel DX79SI (LGA 2011) Intel X79 Express Chipset, BIOS SI.0280B |
| Asus Rampage IV Extreme (LGA 2011) Intel X79 Express Chipset, BIOS 0067 | |
| Asus Crosshair V Formula (Socket AM3+) AMD 990FX/SB950 Chipset, BIOS 0813 | |
| Asus Rampage III Formula (LGA 1366) Intel X58 Express, BIOS 0505 | |
| Asus Maximus IV Extreme (LGA 1155) Intel P67 Express, BIOS 0901 | |
| Memory | Crucial 32 GB (4 x 8 GB) DDR3-1333, MT16JTF1G64AZ-1G4D1 @ DDR3-1600 at 1.65 V on Socket AM3+ and LGA 2011, DDR-1333 at 1.65 V on LGA 1155 |
| Crucial 24 GB (3 x 8 GB) DDR3-1333, MT16JTF1G64AZ-1G4D1 @ DDR3-1066 at 1.65 V on LGA 1366 | |
| Hard Drive | Intel SSD 510 250 GB, SATA 6 Gb/s |
| Graphics | Nvidia GeForce GTX 580 1.5 GB |
| Power Supply | Cooler Master UCP-1000 W |
| System Software And Drivers | |
| Operating System | Windows 7 Ultimate 64-bit |
| DirectX | DirectX 11 |
| Graphics Driver | Nvidia GeForce Release 280.26 Nvidia GeForce Release 285.62 for all SLI testing |
| 3D Game Benchmarks And Settings | |
|---|---|
| Benchmark | Details |
| Crysis 2 | Game Settings: Ultra Quality Settings, Anti-Aliasing: Disabled, V-sync: Disabled, High-Quality Textures: Enabled, DirectX 9 and DirectX 11, 1680x1050, 1920x1200, 2560x1600, Demo: Central Park |
| DiRT 3 | Game Settings: Ultra Quality Settings, Anti-Aliasing: Disabled and 8x AA, Anisotropic Filtering: Disabled, Sync Every Frame: No, 1680x1050, 1920x1080, 2560x1600, Demo: Built-in Game Demo |
| World of Warcraft: Cataclysm | Game Settings: Ultra Quality Settings, Anti-Aliasing: 1x AA and 8x AA, Anisotropic Filtering: 16x, Vertical Sync: Disabled, 1680x1050, 1920x1080, 2560x1600, Demo: Crushblow to The Krazzworks, DirectX 11 |
| Battlefield 3 | Game Settings: Ultra Quality Settings, Anti-Aliasing: 4x MSAA, Vertical Sync: Disabled, 1680x1050, 1920x1080, 2560x1600, Demo: Going Hunting, DirectX 11 |
| Audio Benchmarks and Settings | |
| Benchmark | Details |
| iTunes | Version: 10.4.10, 64-bit Audio CD ("Terminator II" SE), 53 min., Convert to AAC audio 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) |
| Video Benchmarks and Settings | |
| Benchmark | Details |
| HandBrake CLI | Version: 0.95 Video: Big Buck Bunny (720x480, 23.972 frames) 5 Minutes, Audio: Dolby Digital, 48 000 Hz, Six-Channel, English, to Video: AVC Audio: AC3 Audio2: AAC (High Profile) |
| MainConcept Reference v2.2 | Version: 2.2.0.5440 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 |
| x264 Software Library | AMD-Supplied AVX- and XOP-Optimized builds, TechARP's x264 HD Benchmark 4.0, Modified to accommodate new versions of x264 and CPU-Z 1.58 |
| Application Benchmarks and Settings | |
| Benchmark | Details |
| WinRAR | Version 4.01 RAR, Syntax "winrar a -r -m3", Benchmark: 2010-THG-Workload |
| WinZip 14 | Version 14.0 Pro (8652) WinZip Commandline Version 3, ZIPX, Syntax "-a -ez -p -r", Benchmark: 2010-THG-Workload |
| 7-Zip | Version 9.20 (x64) LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5", Benchmark: 2010-THG-Workload |
| Adobe Premiere Pro CS 5.5 | Paladin Sequence to H.264 Blu-ray Output 1920x1080, Maximum Quality, Mercury Playback Engine: Hardware Mode |
| Adobe After Effects CS 5.5 | Create Video which includes 3 Streams Frames: 210, Render Multiple Frames Simultaneously: on |
| Blender | Version: 2.59 Syntax blender -b thg.blend -f 1, Resolution: 1920x1080, Anti-Aliasing: 8x, Render: THG.blend frame 1 |
| Adobe Photoshop CS 5.1 (64-Bit) | Version: 11 Filtering a 16 MB TIF (15 000x7266), Filters:, Radial Blur (Amount: 10, Method: zoom, Quality: good) Shape Blur (Radius: 46 px; custom shape: Trademark sysmbol) Median (Radius: 1px) Polar Coordinates (Rectangular to Polar) |
| ABBYY FineReader | Version: 10 Professional Build (10.0.102.82) Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| 3ds Max 2012 | Render Space Flyby, 1440x1080, from Y: RAM Drive |
| Adobe Acrobat X Professional | PDF Document Creation (Print) from Microsoft PowerPoint 2010 |
| SolidWorks 2010 | PhotoView 360, 01-Lighter Explode.SLDASM Benchmark File, 1920x1080 Render, 1.44 Million Polygons, 256 AA Samples |
| Visual Studio 2010 | Miranda IM Compile, Scripted |
| Synthetic Benchmarks and Settings | |
| Benchmark | Details |
| PCMark 7 | Version: 1.0.4 |
| 3DMark 11 | Version 1.0.2 |
| SiSoftware Sandra 2011 | Version: 17.80 Processor Arithmetic, Multimedia, Cryptography, Memory Bandwidth, .NET Arithmetic, .NET Multimedia |

Although PCMark 7 is able to utilize at least 16 threads (confirmed by Futuremark), it gains far less from parallelism than it does from high clocks and an architecture able to maximize IPC. As a result, you see the thousand-dollar Core i7-990X fall behind Intel’s much cheaper Core i7-2600K.




Because Core i7-3960X incorporates the best of Sandy Bridge in a six-core configuration, it manages to eke out a first-place overall finish.
The rest of the tests also go well for Sandy Bridge-E—particularly the computation metric. Only once is Core i7-2600K able to sneak past the flagship part.
Do these results warrant spending $1000 on a new processor? Decidedly not. However, PCMark 7 centers on applications included with Windows 7, most of which aren’t as demanding as the real-world workloads that characterize the rest of our benchmark suite. So, let’s keep moving…

3DMark’s Performance preset is dialed in to tax graphics processing above all else. But a secondary emphasis on processor-based physics yields an advantage to Core i7-3960X in the overall suite score.

The broken-out graphics subtest demonstrates just how close (most) of these CPUs are to each other when the bottleneck is shifted to the GPU.


However, looking at physics performance on its own reiterates Sandy Bridge-E’s dominance.


Sandy Bridge-E has little trouble jumping to the top of the Arithmetic test, ahead of Intel’s outgoing -990X. Given Sandra 2011’s synthetic nature, it’s no surprise to see it exploiting all aspects of these eight processors.


Using SSE 4.1 (integer) and 2 (floating-point), Core i7-3960X slides right past Core i7-990X for the number one spot. Those figures improve dramatically with the implementation of AVX, though.

One of the things I noticed in Intel Core i7-3960X (Sandy Bridge-E) And X79 Platform Preview was that Sandy Bridge-E enabled significantly better AES256 bandwidth than Gulftown or Sandy Bridge. That advantage persists in the C1 stepping, nearly doubling Core i7-2600K’s result in the Cryptography benchmark. Intel confirms that it made changes to enhance AES throughput, but doesn’t expound on what it did.
Not satisfied, I did a little digging and started pulling memory modules. With three channels of memory, Sandy Bridge-E achieves 8 GB/s AES256 bandwidth. Two channels facilitate 5.43 GB/s. And a single channel of memory installed yields 2.72 GB/s. It seems that AES-NI is very much constrained by throughput (given that it's accelerated in hardware, and consequently very easy to execute), so it looks like the changes Intel suggested are tied to its memory controller, rather than its AES-NI implementation.
Hoping for some correlation to real-world performance (and a reason to get more excited about four 64-bit channels on the desktop), I ran a few tests in the latest stable build of TrueCrypt using the built-in benchmark and a 1 GB buffer. Despite a mean result of 3.8 GB/s in single-channel mode and 5.2 GB/s using two channels, performance fails to scale beyond that, indicating a bottleneck other than the speed at which the processor can encrypt and decrypt data.

The memory bandwidth advantage of a quad-channel DDR3-1600 bus is incredibly evident in Sandra 2011, which manages to realize around 37 GB/s from a potential 51.2 GB/s theoretical maximum.
Impressive though that number is, keep it in context. Sandy Bridge, with its dual-channel DDR3 memory controller, already showed that it wasn’t particularly starved for memory bandwidth in most desktop software. Practically, there won’t be many apps able to exploit those big throughput numbers. Perhaps that’ll change in the first quarter of next year when Sandy Bridge-E turns into Xeon E5 for dual-socket servers.



This page represented one of the best showings for AMD’s FX processor in AMD Bulldozer Review: FX-8150 Gets Tested. Adding the Core i7-990X and Core i7-3960X, however, pushes AMD's flagship down the stack.
3ds Max easily leverages the 12 logical processors presented by both six-core CPUs, additionally putting Sandy Bridge’s architectural benefits to good use as the -3960X ducks in under two minutes.
Really, this is a good example of how heavier-duty apps stand to benefit from more workstation-oriented hardware. Neither the Core i7-920 or Phenom II X4 980 processors are very old. However, Intel’s new flagship finishes this workload in almost half the time of those two quad-core models.

Photoshop CS 5.1 was another application that let AMD’s FX shine last month (it still does well, with a third-place finish). However, Intel’s previous and current flagships displace it.
Parallelism takes precedence over architecture, as Sandy Bridge-E and Gulftown perform pretty similarly (though Phenom II X6 isn’t able to get through the workload as effortlessly as the two Intel chips and AMD’s own FX).

Premiere Pro is an interesting test, particularly because it leverages our GeForce GTX 580 to turn what used to be an almost hour-long workload into a sub-one-minute walk in the park using Intel’s Core i7-3960X.
The Sandy Bridge architecture is partially responsible for this, evidenced by a comparison to the Core i7-990X. But so are extra cores, demonstrated by a side-by-side with the Core i7-2600K.
The FX-8150 doesn’t do too badly here, given AMD’s $249 MSRP. It’s unfortunate that the chip is still selling for closer to $280 online more than a month after its launch.

Sandy Bridge-E, Gulftown, and Sandy Bridge (-2600K) all fall within five seconds of each other in our After Effects render job. That’s hardly a compelling reason to spend $1000 on an upgrade. However, if you’re coming from something older than a Core i7-920 or Phenom II X4 980, the speed-up is more palpable.

Another first-place finish for the Core i7-3960X in Blender represents a five-second victory over Intel’s Core i7-990X and an eight-second win over the Core i7-2600K.
Those narrow advantages are far less impressive than the near halving of the Phenom II X6 1100T’s showing.

The six-core Intel processors score a big win in SolidWorks, though the Core i7-3960X’s design allows it to outpace the Core i7-990X easily. Again, if you’re a workstation user, the gains attributable to Sandy Bridge-E compared to an older Core i7 or Phenom II X4—both of which we still consider very capable CPUs—are sizeable.

OCR isn’t a workload for which we’d normally tap a thousand-dollar chip. However, ABBYY’s FineReader 10 does scale based on available core count, granting the Core i7-3960X a first-place finish. One of our most commonly-recommended CPUs, Intel’s Core i5-2500K takes exactly twice as long to finish this benchmark. How’s that for perspective?

All of the tests up until now have painted Core i7-3960X in a pretty positive light by virtue of optimizations for threading, which keep all of Sandy Bridge-E’s cores busy. But Lame is single-threaded, so the only advantages this new chip has are its clock rate, IPC, and whatever gains Intel can enable with Turbo Boost.
Not surprisingly, then, the Core i7-3960X comes in right around the Core i7-2600K—a CPU about one-third of its cost. The rest of the field follows behind. Without question, this, like the FX last month, is a processor primarily intended to tackle threaded workloads. The big difference is that it also presents solid single-core performance too, rather than sliding backward, which is what we see FX-8150 do.

The same story presents itself in WinZip, roughly. The -3960X and -2600K swap places, yet remain practically tied.

WinRAR is a completely different animal. Not only does it exploit all six of the Core i7-3960X’s cores, but it also demonstrates an affinity for higher-performance memory. Add to that the clear benefits attributable to Intel’s Sandy Bridge architecture and it’s no wonder the incoming flagship does so well, notably outpacing the outgoing top-end model.

Also able to utilize all of a six-core/12-thread processor’s resources, 7-Zip favors the Core i7-3960X just like WinRAR did.
Because this test is well-threaded, FX-8150 delivers a nice gain over the Phenom II X6 1100T, falling just one second short of Intel’s Core i7-2600K.

The creation of a PDF document from a PowerPoint 2010 presentation runs fastest on the Core i7-3960X, but only by a second. The fact that all of the Sandy Bridge-based chips finish within four seconds of each other suggests that the workload only taxes one thread, and favors Intel’s most current architecture over the Nehalem design that came before.
All of AMD’s chips bring up the rear. And because the Phenom II X4 offers better IPC than FX-8150, it’s able to outperform the most recent release. Unfortunately, Phenom II X6 1100T gives up too much clock rate to keep up, despite the fact that its Turbo Core technology dithers at up to 3.7 GHz.

We know from watching Windows’ Task Manager that our Miranda IM client compile test taxes multiple cores, which is why the Core i7-990X manages to slide into second place behind the Core i7-3960X. Intel’s other Sandy Bridge-based chips come in third and fourth, followed by three AMD CPUs. Intel’s Core i7-920 seems to lack the clock rate and architectural advantages of Sandy Bridge to compete in this discipline.

Our last known single-threaded test, iTunes, demonstrates the 100 MHz advantage that Core i7-3960X boasts over Core i7-2600K with only one thread active under Turbo Boost. The Core i5-2500K follows right behind, trailed by Core i7-990X.
Based on the age of the other CPUs in this test, the only real loser is FX-8150, which just came out, but cannot outpace the three-year-old Core i7-920.

FX-8150 performs more admirably in our MainConcept transcode job, where it not only bests the i7-920, but also the other two AMD processors tested and Intel’s Core i5-2500K.
It cannot overtake the -2600K, -990X, or -3960X, though. In fact, thorough optimizations for threading allow Sandy Bridge-E to take first place from the Gulftown-based chip.

Similarly, HandBrake readily takes advantage of available cores, so again Core i7-3960X snags first place. Everything else falls in behind, generally in the order we’ve come to expect from very well-threaded applications.
As with our FX-8150 review, I’m testing three games today. The first, Crysis 2, remains unchanged, as does the third, World of Warcraft. In between, though, we have DiRT 3 (rather than F1 2011). As a last-minute addition, a bit of Battlefield 3 scaling analysis makes it in at the end. For the other titles, two- and three-way SLI testing rounds out the game benchmarks.



There remains very little reason to favor one processor over the others in a graphically-demanding title like Crysis 2. Of course, our GeForce GTX 580 is the bottleneck, and from 1680x1050 to 2560x1600, performance remains fairly consistent.



Our F1 2011 charts really didn’t do AMD any favors. So, I switched to DiRT 3 and retested almost all of the platforms from the FX-8150 review, aside from the Core i5-2500K and Core i7-920.
The results are much more favorable to AMD, though the three benchmarked FX and Phenom II platforms still finish behind all of Intel’s. Interestingly, and perhaps expectedly, there’s really no benefit tied to the Core i7-3960X, which achieves similar performance as the Core i7-990X and Core i7-2600K.



Ironically, the most mainstream game in this comparison is the one best able to take advantage of Intel’s new $1000 processor. World of Warcraft doesn’t fully tax our GeForce GTX 580 graphics card, so swapping CPUs in and out does impact performance quantifiably.
Core i7-3960X consistently enables the best frame rates, followed by both Sandy Bridge-based chips. Gulftown and Bloomfield are next. And all three of AMD’s tested processors serve up disappointing numbers, comparatively.
Finally, though, when we hit 2560x1600 with 8xAA turned on, the graphics load is high enough to equalize all eight platforms.
Intel’s three most modern enthusiast-oriented platforms include X58, Z68, and now X79.
The first proffers 36 PCIe 2.0-compatible lanes, enabled through the X58 Express I/O Hub itself. The second facilitates 16 lanes of second-gen PCI Express on the processor and eight more lanes on the Z68 Express Platform Controller Hub, though you’ll usually only see the CPU’s PCIe used to drive one or two graphics cards. With X79, the 40 lanes of third-gen PCI Express are enabled by the processor, potentially facilitating a massive throughput advantage over either older chipset.
But because there aren’t any PCIe 3.0-capable devices available yet, that feature goes unutilized for now. Still, can Core i7-3960X show us a benefit to using it over prior enthusiast platforms?



In DirectX 9 mode, where the graphics demand is the lowest, Intel’s Core i7-3960X turns in the highest average frame rate. Utilizing DirectX 11, the outcome is too close to call.
There’s nothing to indicate that two PCI Express x16 slots are of any benefit to our GeForce GTX 580 cards in SLI compared to Z68’s two PCI Express x8 slots, though. And any advantage Core i7-3960X holds over Core i7-990X is more likely to be attributable to the processor’s performance itself.



Core i7-3960X fails to put significant distance on our other two three-way-capable platforms in three different resolutions. Sandy Bridge-E does hold a measurable advantage, it’s just very small.



Again, small differences in performance between Intel’s six-core offerings and its quad-core Sandy Bridge-based chips are just as likely to be tied to threading optimizations as PCI Express bandwidth. In fact, because the gaps get smaller as resolution increases, differences in processing performance again make the most sense.



DiRT 3 takes advantage of up to eight threads, though it’s not clear if it prioritizes physical cores over logical ones.
Core i7-3960X manages to score a more sizeable gain in three-way SLI at 1680x1050 and 1920x1080. The performance of all three platforms narrows at 2560x1600, though.



Core i7-3960X’s advantage was already apparent with a single card in World of Warcraft.
Poor scaling at 1680x1050 using SLI indicates that we’re processor-bound, and that this game could really use a faster CPU above all else (indeed, overclocked to 4.6 GHz, the -3960X yields nearly 150 FPS).
Because there’s little difference between no AA and 8x AA at 1920x1080, we can also be fairly confident in a processor bottleneck at that resolution, too. The Z68 and X58 platforms fall in behind, inhibited by slower CPUs.
Finally, at 2560x1600, we see the -3960X’s 8x AA score drop a bit relative to its performance without the feature. But look at everything else—from 1680x1050 to 2560x1600, the frame rates are pretty consistently even.



The same platform-oriented bottlenecks that kept GeForce GTX 580s from stretching any higher in two-way SLI keep a trio of cards from scaling, well, at all, really. WoW just isn’t one of those games that demands more than one really powerful GPU.
It does, however, put the extra performance enabled by Core i7-3960X to good use, as the game shows off a 20 FPS advantage over the next-closest finisher (Core i7-2600K) at all three resolutions.
Although this is a processor review and we’re most interested in how Sandy Bridge-E performs, it’s also important to take a closer look at gaming performance in more GPU-constrained titles to gauge whether the more potent CPU helps alleviate any overhead. With that in mind…
Last week, Nvidia called to let us know that Sandy Bridge-E really allowed three-way SLI to shine in games like Battlefield 3. It showed performance results up to 20% higher than Intel’s prior-generation platform, but it didn’t say whether it used the campaign or a multi-player map for testing. I really like the idea of benchmarking a 64-player rush match, yeah. But I just can’t accept that the results are reliable. I even reached out to Johan Andersson at DICE for guidance on testing, and he admitted there aren’t any good deterministic sequences to profile.
So, I fell back to the same campaign sequence used in Battlefield 3 Performance: 30+ Graphics Cards, Benchmarked, hoping that it was at least graphics-bound enough at Ultra settings to show off what three-way SLI can do.



It turns out that this sequence does demonstrate scaling at all three resolutions. A trio of GeForce GTX 580s yields great performance from 1680x1050 to 2560x1600. It just doesn’t shine significantly brighter on Core i7-3960X.
So, here’s my interpretation of Nvidia’s findings. It’s not that Core i7-3960X allows three-way SLI to stretch its legs in any particularly unique way. In a purely graphics-bound scenario, it scales almost as well on a $300ish Core i7-2600K or a $1000 Core i7-3960X. However, I suspect Nvidia did its benchmarking in a multi-player map, where processor performance is more influential. Less-powerful CPUs become bottlenecks with so much graphics muscle behind them, inhibiting scaling.
If anything, this serves as a reminder why gamers shouldn’t skimp on a processor and load up on GPUs. In a title like Battlefield 3, there are environments that tax graphics (the campaign) and others that exact a more demanding load on the CPU (multi-player). Balancing the two is critical. So, if you’re willing to splurge on three-way SLI, be prepared to also spend generously on a complementary platform. Today, Sandy Bridge-E, by virtue of its per-clock performance and six-core configuration, is unquestionably the best you can present to a trio of potent GTX 580s.
Sandy Bridge-E isn’t on the same tick-tock cadence as Intel’s desktop processors. We’re a few months away from the next “tick,” in the form of Ivy Bridge, and here we are evaluating the performance of a “tock”-derivative.
That’s not bad news by any stretch of the imagination. Intel nailed its 32 nm lithography node, giving the company a mature process on which to build this 2.27 billion-transistor, 434 square-millimeter processor. Gulftown, in comparison, is comprised of 1.17 billion transistors in a 248 mm2 die. Despite that disparity in size and composition, Intel manages to work Sandy Bridge-E into a 130 W TDP, just like the Core i7-990X.
| Processor | Idle System Power Consumption |
|---|---|
| Intel Core i7-3960X (Sandy Bridge-E) | 87 W |
| Intel Core i5-2500K (Sandy Bridge) | 90 W |
| Intel Core i7-2600K (Sandy Bridge) | 90 W |
| AMD Phenom II X4 980 (Deneb) | 100 W |
| AMD FX-8150 (Zambezi) | 111 W |
| AMD Phenom II X6 1100T (Thuban) | 114 W |
| Intel Core i7-990X (Gulftown) | 127 W |
| Intel Core i7-920 (Bloomfield) | 130 W |
And yet, at idle, Core i7-3960X behaves a lot like a 95 W Sandy Bridge desktop processor, dropping down to less than 90 W of system power use measured after 10 minutes of sitting on the Windows desktop. In comparison, Bloomfield (Core i7-920) and Gulftown (Core i7-990X) are much more power-hungry at idle.

After measuring idle system power use, I ran and logged complete runs of PCMark 7 to track average power use in a more graphically-representative way. In order to keep the chart from getting too muddled, I only left Core i7-3960X, Core i7-990X, and FX-8150. You can see, though, in looking at the peaks and dips, that Sandy Bridge-E is using less power than either of its competitors.
Averaging system power use gives us this list:
| PCMark 7 System Power Consumption, Logged In Two-Second Intervals | |
|---|---|
| Processor | Average Power Across One Run |
| Intel Core i5-2500K (Sandy Bridge) | 153 W |
| Intel Core i7-2600K (Sandy Bridge) | 155 W |
| Intel Core i7-3960X (Sandy Bridge-E) | 172 W |
| AMD Phenom II X4 980 (Deneb) | 184 W |
| Intel Core i7-990X (Gulftown) | 189 W |
| AMD Phenom II X6 1100T (Thuban) | 191 W |
| AMD FX-8150 (Zambezi) | 191 W |
| Intel Core i7-920 (Bloomfield) | 193 W |
Core i7-3960X indeed uses less power, on average, than the 130 W Gulftown- and Bloomfield-based parts. It’s bested only by Intel’s 95 W Sandy Bridge processors. Even AMD’s older Phenom II X4 and X6 chips are more power-hungry (as is FX-8150).
| Prime95 x64 Small FFTs System Power Consumption | |
|---|---|
| Processor | Power Consumption After 5 Minutes |
| Intel Core i7-2600K (Sandy Bridge) | 175 W |
| AMD Phenom II X4 980 (Deneb) | 221 W |
| AMD Phenom II X6 1100T (Thuban) | 230 W |
| Intel Core i7-3960X (Sandy Bridge-E) | 253 W |
| Intel Core i7-990X (Gulftown) | 263 W |
| AMD FX-8150 (Zambezi) | 264 W |
Of course, PCMark 7 doesn’t push processors to their limit. For that, we turn to Prime95, which I didn’t run in our FX-8150 review, but include here with most of the platforms in that story re-tested.
Core i7-3960X on Intel’s DX79SI board cuts peak power consumption by 10 W compared to Core i7-990X on Asus’ Rampage III Formula, which itself uses 1 W less than AMD’s FX-8150 on Asus’ Crosshair V Formula.
Core i7-2600K is the only CPU able to facilitate a less-than-200 W system power number, dipping in at 46 W less than the second-place finisher, AMD’s Phenom II X4 980 Black Edition.
Surprised at the savings, especially compared to the smaller and less complex Gulftown design, I asked Intel to help explain how Sandy Bridge-E could possibly use less power. The response was that two cores in the 2.27-billion transistor die are completely fused off, and that a number of other power/speed path improvements were made to help cut the CPU’s draw.

Across 16 different threaded and single-threaded applications, not counting games, Core i7-3960X is about 12% faster than Core i7-990X.
Price-wise, the processors are pretty comparable, though remember that Intel isn’t bundling a cooler with its Sandy Bridge-E-based CPUs. Also, you’ll need an X79-based motherboard and a quad-channel memory kit.
That’s an expensive list of upgrades for a 12% boost if you already own Core i7-990X. But if you have money burning a hole in your pocket, this flagship most certainly outperforms the old one.

When I first reviewed Intel’s Core i7-2600K and Core i5-2500K processors, I didn’t really consider the higher-end chip a solid value. After all, it offered a slightly higher clock, slightly more shared L3 cache, and Hyper-Threading.
However, compared to the Core i7-3960X, which is 212% more expensive, and only able to offer an average 17% performance improvement, the -2600K looks like a real winner. Don’t forget that you need a more expensive motherboard, more expensive memory kit, and an air or water cooler on top of the -3960X’s cost, as well.

The Core i5-2500K is the chip I’ve been recommending for almost a year now to anyone focused on getting the most performance per dollar with plenty of headroom to overclock.
Impressively, Core i7-3960X delivers a 28% average performance improvement over the i5-2500K—but at a 358% higher price (not including the pricier platform to go with it).

It’s a foregone conclusion that Core i7-3960X is faster than AMD’s FX-8150, but the comparison is an interesting one nevertheless.
With all benchmarks weighted evenly, we end up seeing Sandy Bridge-E best Zambezi by more than it beat the Sandy Bridge-based Core i5-2500K. This is a result of close finishes in the threaded apps, and bigger blow-outs in titles like iTunes, WinZip, and Lame.
Unfortunately, pricing that remains way above AMD’s initial estimation means you pay 253% more for the Core i7-3960X, which averages 32% better results across our benchmark suite. Compare that to Core i5-2500K, a processor that fares better in both performance and pricing metrics.
When I first learned about the three processors Intel planned to launch, I petitioned the company at its own Developer Forum in San Francisco to sample the Core i7-3930K either in addition to or instead of the Core i7-3960X. “There is no way,” I argued, “that the thousand-dollar model can come close to the value of a $550 SKU during a serious buying decision.”
That didn’t happen, though, and as such, I can’t see any reason to recommend paying $990, plus the price of a cooling solution, plus a new motherboard, plus a quad-channel memory kit for Intel’s Core i7-3960X. Stepping down to the $555 Core i7-3930K really only costs you 3 MB of shared L3 cache—at a savings that’d cover a really nice motherboard and water cooler. That’s the processor enthusiasts with money (and an SSD full of threaded workloads) should be lusting over.
What about the Core i7-3820? I wouldn’t hold my breath. Expected in the first quarter of next year, Intel won’t yet comment on its price, though we assume it’ll be high $200s/low $300s. The savings is nice, but it involves losing 2 MB of shared L3 cache, two cores, and the unlocked multiplier. Although six additional 100 MHz bins, plus some BCLK flexibility, will be available, you also have to remember that Ivy Bridge won’t be far off at that point.
Also slated to sell in quad-core configurations, Ivy Bridge-based CPUs will work in existing LGA 1155-equipped motherboards, they won’t necessitate new memory kits or coolers, and, depending on the IPC improvements Intel’s engineers extract, could wind up being faster than Sandy Bridge-E in software limited to eight threads or less.
If it’s value you seek, Sandy Bridge-E isn’t the platform for you. Ivy Bridge stands a much better chance of satisfying that niche. However, for the folks who bought into Bloomfield/Gulftown and skipped Sandy Bridge altogether in anticipation of today’s introduction, a marriage of the Sandy Bridge design, more cores, more cache, and more bandwidth yields impressive double-digit performance gains, on average.
Should we see PCI Express 3.0-capable hardware in the next couple of months, Sandy Bridge-E will have yet another opportunity to set itself apart. No other chipset includes this feature, and we expect graphics cards and RAID controllers to exploit it within the first half of 2012.
Just don’t feel compelled to splurge on the $1000 Core i7-3960X. We’re trying to get our hands on a Core i7-3930K—there’s a good chance that’s the Sandy Bridge-E-based chip for savvy enthusiasts looking to overclock.







