Lack of competition hurts any market. For example, with the departure of NVIDIA from Intel chipset maker list (solved by licensing NVIDIA IP for $1.5 billion), the semiconductor giant from Santa Clara slowed down the high-end roadmap. For example, the last performance chipset from Intel was launched in November 2008, i.e. exactly three years ago, while the mainstream chipset part went through a billion-dollar recall due to botched SATA controller.
Every wait comes to an end, though. At IDF, Intel showed us what X79 and Sandy Bridge-E would be packing, and was met with quite a few mixed reactions as it simply didn't appear to be a big improvement over Sandy Bridge's P67/Z68 chipset. Today, we will be taking a look at both the X79 chipset and the processor when we review the DX79SI and Core i7 3960X processor. What question is, can Intel surprise?
The Intel Core i7-3960X is the flagship processor of the Sandy Bridge-E release of Intel's Tock cycle of processor development. For those that are unaware, Intel follows a Tick-Tock architecture design philosophy which involves a new architecture followed by an improvement upon that same architecture as a result of a better manufacturing process or some other performance improvements.
However, with Sandy Bridge-E, Intel claims that they are a Tock to the Sandy Bridge Tick. We're not sure how that exactly works, since SNB-E is the original SNB minus the integrated graphics, plus four CPU cores and a doubled memory controller.
The Core i7-3960X will be followed by a Tick next year by the name of Ivy Bridge which should be manufactured on Intel's new 22nm manufacturing process relative to Sandy Bridge's 32nm process.
Getting back to the processor itself, the Core i7-3960X is the fastest and most expensive processor that Intel currently offers to consumers. The processor itself is a 6-core (sexacore/hexacore) design which is built off of the previous quad-core design of the first Sandy Bridge processors. If you look at the diagram, you'll notice that there are 2 unused cores, these two cores are for the server version that Sandy Bridge-E is based off of, Sandy Bridge-EP. In total, Sandy Bridge-E amounts to about 2.27 billion transistors, which is the highest transistor count Intel has ever seen on a desktop processor. Because of this, Intel has decided to put a heavier focus on the L3 Cache (which Intel calls 'smart cache').
The physical Core i7-3960X die has 20MB cache, but only 15MB are enabled for all 6 cores to share. This is in contrast to the original Sandy Bridge processors like the 2600K which only came with 8MB, meaning that Intel has nearly doubled the amount of shared L3 cache.
Since Intel is still on the 32nm manufacturing process for this processor, Intel is forced to make the die bigger in size to accommodate the extra cores and cache and a huge 415mm2 die is the end result. The Sandy Bridge-E processors also all come with support for Quad-Channel memory which means that you have doubled the memory bandwidth from the original Sandy Bridge. Theoretically, with DDR3-1600 you should reach 50GB/s. You can run memory in XMP mode at DDR3-1866 and DDR3-2133, reaching 58.3 and 66.6GB/s respectively.
The added memory bandwidth should serve the added cache and two more cores well and improve performance in applications that thrive from memory and cache bandwidth. These are all in addition to having hyper-threading and Intel's Turbo Boost 2.0 which enables added performance.
There are three members of the Sandy Bridge-E family and they are detailed in the graph below compared and contrasted against the existing Sandy Bridge processors.
As you can see from the pricing, those prices are per 1,000 tray processors, so you can likely expect an extra 10-20% to be tacked on once retail hits. Newegg is already selling them for $1,049. We're already starting to see some bare/non-retail Core i7 3960X's as well.
In the grand scheme of things Intel has brought back their extreme processors with the same pricing structure as they've had in the past with their Core i7 990X and 980X and older extreme processors all priced over $1k.
Architecturally, the Core i7-3960K really isn't much of a departure from the Sandy Bridge that was released earlier this year with the exception of the additional cores, cache and added memory channels. The Core i7-3960X also clocks in lower than most of its predecessors in both the Nehalem, Gulftown, and Sandy Bridge generations. While we're not quite sure why Intel necessarily did this, we assume it mostly has to do with the 130W processor TDP which they probably wanted to remain within. There will be 150W workstation processors (branded Xeon E-Series), and hopefully it gives us more overclocking headroom.
The Motherboard - DX79DSI
The motherboard that Intel supplied us with was the Intel DX79SI - Siler which is based upon the Intel X79 chipset. This chipset was looked at to be as the 'future' one for Intel but it really disappointed a lot of people once they found out that yet another 'new' Intel chipset wouldn't natively support USB 3.0 which is extremely disappointing to a lot of motherboard manufacturers and enthusiasts that just want native USB 3.0.
Bear in mind that Intel had a lot of issues in development of almost every Sandy Bridge chipset, and X79 is no exception. USB 3.0 and Serial Attached SCSI (SAS) are the two features which the company culled from the final release as the whole project ran late. Even so, the chipset supports up to 14 USB 2.0 ports with 6 coming on the rear I/O and 8 coming via USB 2.0 header.
The X79 chipset also supports 6 native SATA ports with only 2 of them coming in the SATA 3.0/6G type, which is not an improvement over the previous Sandy Bridge motherboard chipsets of P67/Z68. This is because Intel has decided that the desktop variant of Patsburg (X79) would not be getting the added bandwidth to the SATA ports that was initially rumored and as a result of that there would be no improvements to storage performance.
If you look at our DX79SI you can clearly see there is a spot on the PCB for 4 more SATA ports, but Intel clearly made the PCBs before they decided to gimp the Patsburg chipset in order to avoid delay into 2012 (which is what happened with Romley, the SNB-E workstation/server platform - which requires SAS and will only debut at the end of first quarter).
The motherboard itself comes with 8 DIMM slots supporting up to 64GB of RAM with a maximum supported OC frequency of 2300MHz. Intel still only officially supports 1600MHz as a native memory clock speed, a missed opportunity as JEDEC made DDR3-1866 and DDR3-2133 official standard as well. The board also comes with a humongous LGA 2011 socket to fit the Core i7 Sandy Bridge-E processors in. It also comes with two USB 3.0 ports in the back and two more USB 3.0 ports via header (3rd party controller). This board is supposed to also support PCIe 3.0 as most Sandy Bridge processors when paired with properly designed motherboards (right switches and resistors) are capable of the 8GT/s that is part of the PCIe 3.0 spec.
Officially, though, this board isn't technically qualified for PCIe 3.0 quite yet so it's sort of a quasi-PCIe 3.0 classification until the classifications are completely verified. The Intel X79 Siler board also comes with 10-channel audio and dual Gigabit Intel LAN which are completely unrelated to the chipset itself but are rather design decisions instead.
The Intel Core i7-3960X processor is quite literally the fastest desktop processor that Intel offers right now and as such we will be primarily focusing on this processor in our review, not forgetting that there is a more affordable non-extreme i7-3930K sexa-core variant and another i7-3820 quad-core variant. Our system setup for this review will be mostly consisted of Intel components, but we will be reviewing other X79 boards as well.
Our system that we did this review with was based on the following parts:
The DX79SI is a reference board which is designed to still support those who want to overclock. The board itself has 8 DIMM slots, multiple USB ports, audio ports, and PCIe slots. In general, this board is designed like most motherboards out there with a few caveats that we believe aren't necessarily good design decisions.
Here is the CPU with the heatsinks and 8 DIMM slots, with the blue slots being the primary coming in from right to left (from this perspective).
Our first complaint about this board was that Intel opted for only two USB 3.0 ports on the back of the board when there is clearly room for at least 2 more. Engineers also went for a very poor/odd placement of the CPU power and CPU fan headers. The CPU power connector is located in the almost dead center of the board which makes it unlike most other boards out there and also goes against the way that many cases' cable management is designed. Because the CPU power is located there, in many cases people may have a very hard time connecting the CPU power connector after they've installed their cooling solution inside of a case.
If the cooling solution is water, it will be somewhat easier. Thankfully, our Dimastech bench allows us to basically run our cables from the left or right side of the board which enables easy installation of the CPU power connector regardless of its location.
The two USB 3.0 ports (Blue) and the fan header (red) as well as the dual gigabit LAN, firewire, and audio.
This brings us to the CPU fan header and its poor location. Normally, the CPU fan header is located next to the CPU socket where Intel decided to place the CPU power connector. This switch presents a problem for multiple reasons. First of all, the fan header is located to the left of the first 4 RAM slots which means that if any RAM is installed(which there will be), the cable will have to go around the slots and snake its way all the way around the socket to the actual fan. The problem for us was that the fan cable that comes off of the Intel provided liquid cooling solution is just barely long enough to connect to this header without any RAM installed and with RAM installed it is physically impossible. As such, we were forced to use a different CPU fan header and switch the settings in the BIOS.
You can clearly see the unpopulated silkscreen part of the board where SATA (or SAS) headers clearly belong.
Since Intel decided that their Patsburg (X79) chipset wouldn't have more than two SATA 3.0 ports and four SATA 2.0 ports natively, they effectively made their reference board one that is extremely limited from a storage standpoint as you can only essentially get 1 CD/DVD drive, 2 SSDs in RAID 0, and 3 more hard drives. Anything beyond that will essentially amount to needing to purchase a 3rd party SATA controller which seems a bit ridiculous to us considering the price of this board and the processor in it.
We also didn't like the PCIe slot layout of the board in the fact that the board placed the PCIe X16 slots (when in SLI/CrossFire mode) too closely together. It has been the standard design practice of most motherboard manufacturers to leave quite a bit of room between the X16 enabled slots in order to promote better cooling. The problem for us, though, is that this board doesn't accommodate our water-cooling setup of our EVGA GTX 480's which simply cannot be placed so closely together as the Intel board requires in order to have dual X16. If we wanted to place our cards on this board in their proper configuration, they would have to be in the X16 and X8 slots which would slow down our cards' performance on a platform that is supposed to make them faster.
The one thing we are glad to see in this board's design is a very good heatsink design that isn't obnoxious, but at the same time still gets cooling of critical board components done. Some X79 boards out there are either hideous in their aesthetics or they have active cooling with an added fan on one or more places on the chipset which seems unnecessary when you consider that Intel's own reference design doesn't have a fan.
Intel Goes Liquid Cooling
When it comes to liquid cooling, there's no doubt that Intel made the right, at and easy decision of going with an Asetek LCLC-HP (High Performance Low Cost Liquid Cooling) solution. Intel's branded solution is named the RTS2011LC for LGA 2011 liquid cooling. This cooler will also fit on previous generations of Intel's processors such a 1366, 1155 and 1156. Unfortunately for the legacy 775 users, you're going to get left out.
This cooling solution itself isn't anything special or anything to write home about other than the fact that it has a unique blue fan and the Intel logo in blue on the combination block/pump assembly. From our experience in the past, these LCLC solutions have performed pretty well and general perform nearly identical to each other in terms of thermal performance. As such, we expect to be able to get some decent mild overclocks with this cooler but nothing extreme. We will detail our results in our temperature and power section towards the end of the review.
For benchmarks we will be taking a look at a broad array of benchmarks running on this system at stock clocks as well as various overclocked settings to show you guys what kind of performance you can expect out of SNB-E (Sandy Bridge-E) and X79. We will also do our best to compare against the previous generation of Sandy Bridge and AMD's latest and greatest offering (FX-8150) when possible.
SiSoft Sandra 2012
We ran a battery of tests in SiSoft's brand new 2012 Sandra benchmark so that you guys could get an idea of what kind of performance the Core i7 3960X is capable of and how it stacks up against the older Intel and AMD processors out there. Because of the relative newness of Sandra 2012 and the time constraints that we had, we didn't get a chance to benchmark it against AMD's Bulldozer or the older Intel processors, but here are some screen shots of our results compared against what SiSoft benched on Sandra 2012.
Sandra 2012 - Processor Arithmetic
In this test we saw that the Core i7 3960X scored pretty well and scored just as well as or better than the 2600K. In the Dhrystone test, the Core i7 3960X scored 2.21GIPs (152.49-150.28) - lower than the 2600K. Even though this is a sexa-core versus a quad-core, bear in mind that Core i7-3960X is clocked 100MHz lower and also has a much lower Turbo speed. In the Whetstone test, though, the Core i7-3960X definitely took the cake by scoring 119.45 GFLOPS over the Core i7 2600K's 96.52GFLOPs signifying an improvement of approximately 20%.
Sandra 2012 - Multimedia
In the Multimedia test, Sandra 2012 tests the Multi-Media Float in (MPix/s) and Multi-Media Integer in (MPix/s). In this test, the Core i7-3960X obliterates everything in sight by quite a large margin. In the Integer test the Core i7-3960X scores 296.52 MPix/s while the 2600K only scores 239.63, a difference of 20% (once again). In the Float test we saw the margin widen between the i7-2600K and Core i7-3960X even more with the Core i7-3960X scoring 405.79 MPix/s and the Core i7 2600K scoring 306.68 MPix/s. This is a huge improvement of about 33%.
Sandra 2012 - Multi-Core Efficiency
The Multi-Core Efficiency test once again combines two tests, but in this case it combines Inter-Core Bandwidth with Inter-Core Latency. Both seem to be pretty good measurements of what could be considered to be a proper definition of multi-core efficiency. In the Inter-Core Bandwidth test we see the Core i7-3960X blast past all the other processors including the Core i7-2600K. The Core i7-3960X shows 35.64GB/s of inter-core bandwith over the Core i7-2600K's 24.87 which represents nearly a 50% increase in terms of inter-core bandwidth. This can be directly attributed to the increase in the L3 shared cache to 15MB.
In the Inter-Core latency test, the i7-3960X scores worse than the i7-2600K, expectedly so considering that there are 6 cores to deal with rather than 4 in the i7-2600K which should expectedly increase the latency.
Sandra 2012 - Memory Bandwidth
In the memory bandwidth test, we really don't expect anything other than a complete blow out from the Core i7-3960X simply because of the quad-channel capability of the X79/SNB-E platform and the clock speeds that it is capable of. In the integer memory bandwidth test the X79/SNB-E destroys the X58/990X platform in quite a big way. The X58 chipset is triple channel which is already better than the original Sandy Bridge's dual channel.
As you can see in the graph above, the Core i7-3960X combined with the X79 score 37.48 GB/s in memory bandwidth which is nearly a 50% improvement upon a similarly clocked (RAM) X58 setup. If you look at the Sandy Bridge (green) graph, you will notice that it only scores 23.93GB/s and that is while the RAM is clocked at 2133MHz unlike the X79 and X58 which are both at 1600MHz. We will detail the 32GB performance of X79 in another later review as we didn't have enough time to properly explore the benefits of going from single quad-channel to dual quad-channel setups.
We initially ran an earlier version of AIDA64 which gave us some interesting benchmarks to talk about showing that the Core i7-3960X didn't quite perform up to par, that was until an update rolled out and we were notified that there was an AVX optimized version that would improve performance. Upon re-testing everything we saw improvements between 10 and 20% which we would say are pretty significant and as a result we scrapped our benchmarks and reran them with the ones you see below.
AIDA64 CPU Tests
The Core i7-3960X barely edges out the previous-gen i7-990X
Here the Core i7-3960X destroys the competition by a wide margin.
Once again, the Core i7-3960X wins by a broad margin, almost double the i7-990X
In these four CPU tests that we ran in AIDA we really didn't see anything surprising other than in the Hashing test where the Core i7-3960X didn't actually win by any margin at all, but rather lost to the 12-core Opteron 2431. While the i7-3960X does have 12 threads thanks to its hyper-threading, it appears that the sheer quantity of cores in the Opteron gave it an edge in terms of raw hashing power. However, the workstation version will come with 8 cores enabled and 16 threads, so we'll see how SNB-E will fare against the old Opteron. The new 16-core Opteron debuted today... so AMD still has a chance to remain the winners in this benchmark.
As one can see, in tests like the AES test Intel has made some significant improvements to AES performance in the Core i7-3960X as it doubles the i7-2600K's score and nearly doubles the Core i7 990X's as well. Overall, quite an impressive performance.
AIDA64 FPU Tests
Here we see the 3960X win by about 20% over the 990X
In this test the 3960X beats the 2600K by a whopping 50%
The results continue to be astonishing once again with the i7 3960X beating the i7 2600 by more than 50%
In the four FPU tests, the 3960X once again scores first in almost all of tests by quite a large margin except for one. In the FPU SinJulia test the 3960X actually loses to the older 990X, this could be due in part to the fact that SinJulia is based upon the floating point X87 instruction set rather than the x64 instruction set, so the Core i7-3960X doesn't really improve X87 operation performance. It does, though score about 50% better than the 2600K which could be attributable to the added cores and memory bandwidth.
AIDA64 Memory Tests
Here we can see the $999 3960X just barely beat out the $315 2600K
Interestingly enough, in this test the i7 2600 with only dual-channel 1333 beats out the i7-3690X with quad-channel 1333.
Here we see a veritable tie between the i7-2600K and i7-3960X
In the memory tests, AIDA64 tests for copy, read, write and latency. Surprisingly enough, The Core i7-3960X does not score 1st in the majority of the tests, but rather 2nd in most of them. In the case of the memory copy and memory write tests, it loses to the Core i7-2600K and in the memory latency test it loses to AMD processors.
The reigning champion is the AMD FX-8150 which comes in at a snappy 45.8ns beating the Athlon64 X2's 47.5ns which also happens to beat the Core i7-3960X. Hopefully Intel can work more on their memory latency with Ivy Bridge as the AMD FX-8150 is currently the memory clock speed champion due to having such a good memory controller (even though it is only dual-channel).
AIDA Cache and Memory Benchmark
Here we take a look at the performance of the Core i7-3960K with its different levels of cache and its memory performance.
You can also see the improvements that we achieved on the memory and cache performance once we overclocked the processor to 4.86GHz. You can notice here that the L1 cache latency actually decreased over Nehalem, i.e. the original Core i7 processor.
As you can see from our overclocks, the Core i7-3960X when overclocked sees huge improvements to L3 cache bandwidth as well as latency in memory and cache. We managed to get sub 1ns latency in the L1 cache with over 151GB/s of bandwidth. This is the first time we're seeing sub 1ns latency after the early days of Pentium 4.
In Cinebench R11.5 we weren't quite sure what to expect, but once we got our set up running correctly we were amazed at the numbers we were seeing.
With the Turbo-enabled Core i7-3960X, we got a score of 10.36 - nearly double that of the Nehalem generation and nearly double that of the currently the fastest AMD processor FX-8150, which scored only 5.86.
If you look at the benchmark screen shot itself you can see other scores we posted with the Core i7-3960X and the 9.68 score represents the processor without Turbo Boost enabled and the 9.53 score represents the Core i7-3960X running with only 4GB of RAM at 1866MHz in dual channel mode. This was done to show the differences in performance between different memory and processor configurations.
We also overclocked the Core i7-3960X to 4.86GHz and managed to get an amazing score of 14.06 in Cinebench R11.5. What is amazing is the fact that a six-core, twelve-thread processor operating at 4.86GHz beat eight-core sixteen-thread Xeon system running at 3.33GHz. A single piece of silicon blasted two pieces of silicon, quad-channel memory trumped six-channel memory. In the end, architecture is still king.
Futuremark 3DMark 11
Since we're mostly interested in CPU performance we will focus more on the Entry Level (E) lower resolution score of 3DMark 11 and check on the CPU (physics) score as well. In 3DMark 11 we saw the Core i7-3960X score an E9005 with a Physics score of 11096.
This is in contrast to our recently benchmarked FX-8150 which scored a combined score of E7896 and a physics score of 6481, barely 60%. This doesn't bode well for the FX-8150 even though it does cost just a quarter the price. However, a Core i7-3930K can be had for a little under twice the price of the FX-8150. Note that both of these scores were taken with the exact same graphics card and the driver version.
In PCMark 7 we ran the Intel Core i7 3960X and DX79SI through a full battery of tests and got a PCMark score of 4415. This compares against the recent test we did of the FX-8150 which scored a 4206 yielding an improved score of about 200+ points purely on switching out the motherboard, CPU and RAM which admittedly are all pretty hefty upgrades in terms of price. So, in that sense PCMark 7 doesn't really show that the extra money spent on a Core i7 3960X really yields that much of a difference in PCMark 7.
LinX Linpack Benchmark
In LinX we ran the processor in both hyperthreaded and non-hyperthreaded modes in order to show the Linpack performance difference with the Core i7-3960X when hyperthreading is enabled or disabled and what some may be able to expect from this processor in the server world.
With the Core i7-3960X having Hyper-Threading on it scored a peak of 64.9275 GFLOPS. When running with Hyper-Threading off, there is a small gain to 69.1810 GFLOPS. This is still significantly better than what we tested with our last CPU review in the FX-8150 which only scored 30GFLOPS meaning that the Core i7-3960X is more than twice as poweful as the FX-8150 in terms of Linpack performance.
We ran two test suites in Passmark, the CPU and Memory marks. In the CPU Mark, we end up re-hashing a lot of the same benchmarks that we ran in the past.
In the CPU Mark, you can see that overall, the 3960X effectively is about 20% faster than the 2600K and 30% faster than the 990X. In the prim numbers test we can see that the Core i7 3960X struggles a bit compared to the other Intel processors and especially against the FX-8150 where it takes the cake. The same story can be said for floating point math where the FX-8150 is 26%. Otherwise, in almost every single test the Core i7 3960X pretty much wins by quite a bit.
In the Passmark Memory Mark test we were interested to see that the Core i7-3960X scored far ahead of all the other processors including the 990FX which it beats by about 20%. Interestingly enough, though, the i7-3960X doesn't bode as well against the i7-990X and i7-2600K in the rest of the tests, similar to what we saw in AIDA64, so we can be sure that our results both tests are not erroneous.
For video encoding, we ran two different tests. One being a benchmark and the other being a real-world test of handbrake.
In the x264 HD benchmark, we saw that the Core i7 3960X performs quite well over the 2600K in the second pass where it gets an almost 20% improvement. In the first pass, though it does only 4% better than the 2600K. meanwhile, in handbrake we managed to take a 1080P file and convert it into an iPhone resolution in 20 minutes, this was achieved at an average frame rate of 97.843 with zero errors and zero drops. This is in contrast to the FX-8150 we just tested which got an average frame rate of 95.5 and took a little over 21 minutes to accomplish.
Colin McRae's DiRT 3
In DiRT 3, we decided that we would just run the game at full resolution and simply compare that against what we got with our FX-8150 which was supposed to improve game performance because of its core optimizations. In DiRT 3 at full settings at 1920x1080 we got an average of 66 FPS with a minimum of 62 and a max of 73. This was in contrast to our results with the FX-8150 which scored an average of 66FPS with a minimum of 51 and a maximum of 75. In this game, it appears as though there is really no benefit to the Core i7-3960X and that the FX-8150 actually has a higher maximum FPS than the much more expensive Intel processor. Here, AMD's Game program shone under the limelight and produced clear results.
In Metro 2033 we ran the game at an ultra-low resolution (800x600) in DX9 which yielded an average FPS of 153.5 with a minimum FPS of 76 and a maximum of 201FPS. With the CPU overclocked to 4.8GHz, it managed to crank up to 171.06 average, a minimum of 118 and a max of 208. With everything set to ultra and the resolution at 1920x1080, we managed to get an average frame rate of 59.4 with a minimum of 24 and a max of 109 which is still playable with only one graphics card. With the FX-8150, we managed an average FPS of 50 and a minimum FPS of 23 with a max of 92. So, it appears in Metro 2033 there is actually quite a difference in terms of performance depending on the CPU as the average frame rate at max settings went up 20% from 50FPS to almost 60FPS.
Since Battlefield 3 is such a new title, we haven't really had a chance to compare this processor against anything else that we've tested but we fully plan on having Battlefield 3 as part of our test suite here at BSN* when it comes to testing game performance.
When we set BF3 to Ultra pre-set settings and ran the game at 1920x1080 resolution we managed an average FPS of 78.048 and a minimum of 64 FPS with a maximum of 109 FPS. In reality, we'd really like to be able to test the multiplayer rather than the single player, but the multiplayer varies far too much from game to game in order for us to reliably compare our results from card to card and cpu to cpu. Although, single player performance should be a relatively good measurement of what the engine's performance will be on your system and shouldn't deviate too far from single player performance.
When it came to overclocking, the Core i7-3960X and DX79SI were kind of a mixed bag. We saw quite a bit of performance improvements when overclocking this processor, but we felt that we had to pump way too much voltage into the processor to get it stable at 4.86GHz on water. Our biggest obstacle, we felt was the board itself and not the processor. We also noticed that in our stress test at 4.86GHz that the vCore was fluctuating quite a bit unlike it had been at stock which to us was an indication of not good enough power delivery to the CPU at higher vCores.
Intel's overclocking assistant built into the BIOS with the 1.25x gear ratio really made achieving 4.63GHz extremely easy and is a very nice assist for anyone looking for a good base to build off of when overclocking their processors. We're certainly interested to see this develop in the future with more overclocking assisted profiles and tools.
If you look at our overclocked Cinebench R11.5 score, you can see that once the Core i7-3960X is at 4.8GHz it is unbeatable by a mile. We also have a comparison of our AIDA64 scores at 4.86GHz as well just to show how big of an improvement a 47% OC can provide. We went from 9.68 to 14.06 with the 47% OC.
The Intel Extreme Tuning Utility for the DX79SI is definitely a welcome addition to Intel's boards and could make overclocking a much easier process for those that don't know where to look in the BIOS, but the truth is that most of the changes made in the ETU end up requiring a restart anyways and it really seems to lose a lot of its value if you have to restart anyways. Having something like EVGA's Eleet may be more beneficial to Intel if they can find a way to do clock speed bumps without actually requiring a restart. But then again, doing it in the BIOS after a restart is always our preferred method of overclocking and we don't really recommend doing it any other way.
Our biggest complaint, though was that because we had to feed the chip so much voltage in order to keep it stable the processor ended up hitting 91C under stress tests at 4.8GHz. We are quite sure that with a better VRM and overall better board design, we could probably hit 5.2 to 5.3 or even higher, especially if we go for a better cooling solution.
Power and Heat
When it came to power, as a whole the Core i7-3960X consumed around 235W under full load and at idle it sipped 112W. The processor itself, though, at idle consumed 15.7 Watts with only 3 Watts actually coming from the cores and 12.7 Watts coming from the uncore. At load, though, the CPU managed to consume 103W with 86W coming from the CPU cores alone.
You can also compare these stock power consumption numbers against our overclocked ones below.
If you're having a hard time figure out those 4.86GHz power consumption numbers, the CPU package as a whole is consuming 168.76W and the cores alone are consuming 143.49W and you can easily see the idle numbers illustrated in the graph above.
When it came to temperatures, the Intel provided liquid cooling solution did a pretty good job at keeping the CPU cores under load in the mid to high 50's with the cores hitting the high 80's low 90's at 4.86GHz. We were happy to see that the Intel liquid cooling was able to keep up with the processor at 4.86GHz even when it was nearing the tmax of the processor.
If you plan on overclocking this processor on this board as highly as we did, though, we would probably recommend looking at something like an H100 or better as that will run much cooler and give you more overclocking headroom if you want to have it. The Intel liquid cooling thermal solution isn't bad at all and definitely serves well as a mild overclock cooler, but anything above 1.46V is asking for trouble as that will likely push the CPU over its thermal threshold without proper cooling.
Stock CPU at 100% load
4.86GHz temperatures where we hit 91C on one of the cores, explaining why the motherboard needs to have better voltage control.
When it comes to the Core i7 3960X and the DX79SI there is no doubt that each of these is not cheap. The DX79SI will retail somewhere around $300 and the Core i7-3960X is already retailing between $1,049.99 and $1,099.99 depending on where you look. When you take that, and the fact that the CPU does not come with any thermal cooling solution, you begin to question where the value really is.
The board itself, really doesn't lend itself to the ~$300 price and the processor is expectedly expensive, but comes without any cooling. Yes, it will still come with a coveted 3-year warranty, but without a cooling solution you are either forced to spend ~$100 and buy the Intel solution or go elsewhere and buy your own, and frankly from our own experience we'd recommend you take that $100 and spend it on an H100 because it'll give you the best overclocking for your money, especially on this processor.
The Core i7-3960X and the DX79SI are most certainly a deadly combination for any kind of task that you may throw them at. With 16GB of fast memory and a good GPU, these parts can easily be combined to make the ultimate anything rig. The truth is that the Core i7-3960X is such a capable processor that it is really a shame to see that Intel spend more time and effort on the chipset side of things and make the platform as a whole shine. We really wish that X79 wasn't going to live up to the bad rumors, but it did indeed and as such, it feels like a glorified P67 with more RAM and PCIe lanes (both of which are really CPU dependant now, not chipset). But either way, if you want the bar-none fastest machine today, you have to get it with a Core i7-3960X and an X79 chipset. It is not perfect for reasons above, but it is as close as it can get.
As such, we must award the Core i7 3960X with our Editor's Choice Must have purely based upon its sheer performance and dominance of almost everything we could muster to throw at it.
We would also like to thank Intel, Kingston and Corsair for providing the components that we needed for this review and you can expect a review of the Kingston and Corsair RAM kits soon.