Intel Ivy Bridge-E 4960X CPU Review (LN2 inside)

CPU by leeghoofd @ 2013-09-09

After upgrading the performance of the mainstream platform by the Haswell family, the Intel engineers also devoted time to perform an overhaul on their high end platform, LGA 2011. The Sandy Bridge-E  was a worthy successor of the old and aged LGA1366. Intel waited close to two years for manufacturing the new high end flagship CPU model available to the masses, this in contradiction with the launch schedule on some previously leaked roadmaps. As with the transition from Sandy to Ivy Bridge on the LGA1155 socket, we spot many similarities with the new Ivy Bridge-E. A die shrink to the 22nm process and alike architectural improvements should boost the performance and efficiency. Time to unravel the brand new Intel flagship LGA2011 CPU: the i7-4960X CPU.

Introduction

Ivy Bridge-E is the 22nm die shrink or a Tick as Intel refers too for their high end LGA2011 series. Thus not a complete architectural overhaul as the upcoming Tock version will be. Nevertheless, same as with the transition from Sandy Bridge to Ivy Bridge, there was more to the story then just a smaller manufacturing process.

 

 

The Intel i7-4960X on the left and our already heavily tortured i7-3960X on the right

 

 

The biggest improvement of the 22nm process, is that it allows for faster switching frequencies, yet at the same voltages, meaning that the processors in general become more efficient.  Also by using the 22nm process it becomes cheaper for Intel to produce, as they are able squeeze more transistors in the same die size.

Remember the Tri-gate Transistors from Ivy Bridge? These improve the current flow, as it can travel now in three dimensions. Beforehand it was only available to work in one plane. From left to right the conventional design, the 3D design and on the right the utilized Tri-gate system of the Ivy bridge and Ivy Bridge-E series.

 

 

The new 6-core Ivy Bridge-E desktop models are no longer 2 core disabled 8 core Xeon CPU versions. Intel went straight for a native 6 core design, instead of deriving their hexacore i7 CPUs from the high end server Xeons series. Again this allows for cheaper production for Intel and a lower power consumption for the end user. The ten or twelve core versions will only be available in the aforementioned Xeons server models (E5 2600 V2)

 


The PCI-Express 3.0 is now official, no more secrecy about it as with the Sandy Bridge-E lineup. The biggest advantage of LGA2011 CPUs over the more mainstream processors is the ability to divide 40 PCIe lanes over the different slots, thus allowing dual GPUs running at x16 speeds or even 4 cards at x8 speeds. This all without the need of an extra latency-creating PLX bridge chip.

This time no coinciding motherboard chipset launch, yet reutilizing the old and trustworthy X79 chipset. This means however: no native USB3.0, only support for 2 x SATA 600 and 4 x SATA 300 (unless extra controllers are installed) and no Thunderbolt on the exisiting X79 boards. Maybe a missed opportunity as the X79 chipset is getting or is for some already outdated. However on the other side, it allows exisiting X79 users the chance to upgrade to a brand new Ivy-E CPU by just a bios flash.

 

3 Different Models

Three models will be released for the LGA2011 Ivy Bridge-E series: the entry model being the quad-core i7-4820K, the mainstream hexacore i7-4930K and the Extreme model: the i7-4960X CPU. Biggest change versus Sandy Bridge-E is that the 4-core 4820K model is now a K skew model, allowind full multiplier selection. A great and applauded update Intel! Alike the Sandy-E, the Ivy Bridge-E models are sold without a cooler. Thus an aftermarket cooler is required and aim straight for a beefy air cooler or high end AIO or even better a good water cooling build.

We compare the old versus the new in the below chart:

 

 

As you see no drastic changes versus Sandy Bridge-E, besides slightly higher running clocks and therefore also higher Turbo speeds. The Cache Size is the same for the Hexacore versions, the quad core i7-4820K gets an extra 2MB. The TDP remains rated at 130W. The official memory support is now updated : 1866MHz versus 1600MHz before.

 

 

Since we talked about the K skew models, there's still a bit more extra under the hood for the extreme enthusiasts. The CPU ratio has been lifted to 63X versus 57X before. The three gear ratios are still present: 1.00x, 1.25x and 1.66 allowing for finer coarse increments. Nevertheless the DRAM dividers are still just in 266MHz increments, which is a pity. On the Rampage IV Extreme board max workable selectable divider is 2400MHz, so to achieve higher DRAM speeds we need to select a higher CPU ratio: 1.25x or 1.66x

 

 

With the release of the Ivy Bridge-E the Xtreme Tuning Utility, XTU was also updated to support the new processors. At HWBot.org the community has already thoroughly tested with this new overclocking tool and it has been approved by the HWBot staff as an official benchmark. Cool features are that you can share profiles, compare settings and scores. Plus the benchmark doesn't take too long to complete. Check out the scores at the XTU zone.

 

 

No changes in XMP profile either, still Xtreme Memory Profile 1.3 is valid, ofcourse in a quad channel configuration to maximize the performance of this platform.

 

Test Setup and Methodology

Normally with each press kit, Intel includes one of it's own high end boards for testing purposes. Due to the fact there is no successor yet for the X79 chipset, we opted to utilise the ASUS ROG Rampage IV Extreme motherboard and flashed with the 4301 Ivy Bridge-E compatible bios. This board would also serve as the perfect testbed for our extreme cooling tests.


 

TEST SETUP :

  • Intel Core i7 3960X/4960X
  • Corsair H80i cooler
  • ASUS Rampage Extreme IV (4301 bios) (4401 bios for LN2)
  • 8GB Corsair Dominator PC12800C8 (4 x 2GB) Platinum 2800C11 (4 x 4GB)
  • AMD 7970HD Catalyst 13.4 driver
  • Western Digital 1TB Green Caviar
  • Corsair HX 1050 PSU
  • Corsair Vengeance case
  • Windows 7 Professional 64Bit SP1 fully patched.

Note: All the Intel CPUs were run with the enhanced Turbo, meaning all cores run at the same speed !!

All tests are conducted three times and the average run is then calculated. During the first pages, the out of the box performance is displayed. Then we look a bit closer clock per clock versus it's main rivals. Memory divider tests are up next and finally the Overclocking part is split up in first, the daily possible overclocks and then to conclude the testing we go subzero to push the envelope.

Stock 2D Results Part I

Let's start off the stock tests. Keep in mind that the CPUs are all running at max possible Turbo speeds.  First up the quick single threaded 1M SuperPi test. The outcome depending on pure MHz and efficiency of the instructions per Clock, all the rest  (DRAM timings,...) hardly matters. No wonder the new i7-4960X is the top contender here with the cores running at 4000MHz. The Haswell generation is right on its heels, but the extra clocks make the difference here. In the Wprime32 multithreaded test the i7-4960X is again untouchable. Let's see what happens if we launch the longer test versions of these applications.

 

 

SuperPi 32M gets a huge boost versus the Sandy-Bridge E versions. Take note that these are not the sole outcome of the 100MHz higher core clock (4GHz versus 3.9GHz for the 3960X). In fact we spot a similar pattern, as when the transfer was done from Sandy to Ivy Bridge (LGA1155). A far higher efficiency was achieved clock per clock by the new design, depending on the benchmark up to 15% improvements were noticeable. In SuperPi 32M the Haswell generation is superior thanks to superfast Cache and Ultra low latencies.

 

 

The Quad Channel DRAM configuration delivers insane numbers in the AIDA 64 Bandwidth test. However as mentioned before it's not all about bandwidth. The Quad channel DRAM setup on the X79 motherboards have got higher latencies (60 versus 50-45) versus the Dual channel configurations of the Ivy and Haswell LGA1155/50 motherboards. Spot the nice Copy boost in the AIDA Bandwidth test versus the Sandy Bridge-E processors.

 

 

 

In the X264HD encoding test we spot that the new improvements in design and instruction set again pay off. An 8-9 FPS performance increase versus the predecessors in the Pass1 test and even a 3-4 FPS increase in Pass2 is impressive. Again this outcome is alike with the Sandy versus Ivy Bridge comparison. The 6 cores and 12 Hyperthreaded ones make these LGA 2011 Ivy Bridge-E CPUs insane crunching or rendering monsters. In Cinebench Release 10 we note that there's a solid performance increase in both the single and multi-threaded test versus the slightly lower clocked predecessor.

 

Stock 2D Results Part II

Let's continue the stock testing with another one of Maxon's benches, this time Cinebench R11.5 64Bit. Again the Ivy Bridge-E doesn't fail to impress from the moment all cores get into motion. Once we get our hands on the more humble 4930K and 4820K CPUs, things will get more interesting budget wise.

 

 

 

Both The Fritz chess benchmark and the rendering program POVRAY v3.62 (single threaded) nicely show the step forward Intel took by moving their LGA2011 CPUs to the 22nm Ivy Bridge-E architecture. The beta Winrar test remains plus minus on par with the performance of the  I7-3960X.

 

Stock 3D Results

Time to put the Radeon HD 7970 card to the test. In Futuremark's 3DMark11 we observe a healthy boost in the PhysX score. For the total outcome of the benchmark, the new CPU architecture is again that tiny bit faster. However this is purely synthetic and of course closely monitored by benchers. Gamers will hardly feel the differences in-game between a quad core 2600K or this hexacore power monster. The graphics card is mostly the limiting factor between smooth gameplay or jerkovision in a modern PC configuration. Not much to mention on this page, either the single GPU is completely limited to put out more FPS or the game relies a bit more on extra crunching power to display a minimalistic higher FPS. For most gaming environments the mainstream Haswell or even the older Intel models are more then sufficient.

 

 

 

 

Quick Clock per Clock Comparison

Of course all these enhanced Turbo clocks are a nice feature, however to really compare different CPU architecture's performance, one must compare clock per clock. Thus we opted to retest both hexacore CPUs with all cores running at 4000MHz and disabled the Intel speedstep technology.

For reference we added the Haswell i7-4770K also running at 4GHz clocks.

 

 

In superPI the new improvements to the design show what the Ivy Bridge-E generation is capable off. The Haswell CPUs with their superfast Level 1 cache are unbeatable clock per clock. In Wprime, which is a core crusher, the new Ivy Bridge-E remains ahead of the similar clocked Sandy Bridge-E CPU. The 4-core Haswell 4770K CPU has got no chance against the hexacore processors.

 

 

 

And the story continues, the new Ivy Bridge-E is, depending on the benchmark tested, in general 5-10% faster then its predecessor, no matter being it Cinebench or the X264HD encoding test. Once all the cores get hammered, the i7-4960X shows its colours.

 

 

 

 

RAM Dividers Tested

Is high end DRAM worth the cash yes or no? That question is always debatable. It all will depend on the usage and the price you want to pay for your PC. Some will do it just for the bragging rights, others will remain sceptic and stick with slow 1333MHz DDR3 DRAM.

Memory manufacturers always advertise that more is better. Sales and PR marketing at their best of course. We tested the following speeds and timings to unravel (again) some of the mysteries. From stock 1600MHz to a whopping 2800MHz DRAM speed.

We took the following speeds as they seem to tbe he most comon sold at the [M]Fav shop. Take note that DRAMs at similar speeds , but with tighter timings might jield better performance then displayed here.

  • 1600MHz at C8-8-8-24 Command rate 1T
  • 2133MHz at C9-11-10-27 Command rate 1T
  • 2400MHz at C11-13-13-31 Command rate 2T (1.25x ratio)
  • 2800MHz at C11-14-14-35 Command rate 2T (1.25x ratio)

Of course there's more going on then just upping clocks and adjusting the primary timings. The Bios will adjust also secondary and other timings to maintain stability. Sometimes at the cost of performance.

SuperPi 32M displays a 9 second difference between 1600MHz and 2133MHz, which is nice. From 2133MHz on, the performance increase is far less steep. For benchers and tweakers, the higher and most efficient they can dial in their setup, the happier they get :P. For daily users 1866-2133MHz is plenty !

 

 

 

It will all depend a lot on the programs which the user will run to see if them high costly DRAMs are really worth every penny. The AIDA Bandwidth test outcome might encourage you to go bonkers. However, sometimes the subtimings and or latencies get penalized, making it perform less impressive in real world applications.

The X264HD v4.0 test shows again scaling with extra DRAM speed in the Pass 1 test. The Cinebench R11.5 64Bit CPU test and VGA test show very mild scaling, nothing earthshattering. In the 3DMARK11 test, only the Physx test gets a solid boost, for the rest not much to tell...

 

 

 

 

We also quickly tested a brand new 32GB high end kit from Corsair: the Vengeance Pro 32GB 2666C11 kit and with a little added VCSSA voltage to 1.35V, we got the XMP profile (11-13-13-35 2T) and even tighter timings (10-13-12-35-2T) perfectly stable for the [M] benchmark suite. This kind of quantity and speed was impossible to achieve on SandyBridge-E, the quality of the Integrated Memory Controller (IMC) took great steps forward.

 

  

 

The 16GB Corsair Platinum Hynix CFR kit did a SuperPi 32M run at 2933C11 at 1.725 Vdimm. For stability we had to up the VCSA voltage to 1.365V.

 

Overclocking

Time for the fun part and start to find the limits of this new processor architecture for daily usage or abusage :).

We worked our way slowly up and started to increase the multiplier and CPU Vcore voltage. Stability testing was done with a small half-hour prime95 Custom 12-12K test. Then the CPU had to remain stable during a SuperPi 32M, Cinebench R11.5, Fritz Chess test and finally complete an encoding X264HD test.

While we got 4725MHz rock stable at 1.43Vcore with the RAM running happily at 2800MHz, with ambient temps of around 20°C, the core temperatures during the prime test get according Realtemp in the 80°C range. Much will depend on your own Processor, cooling method, ambient temps and most important your own personal green or safe operating zone. My setups are on daily 24/7 and I prefer to find the sweetspot for temperatures and voltages. during During the testing we discovered once cranking the DRAM divider, this CPU achitectures prefers to be on the 1.25 strap.(board limitation)

 

 

Thus we settled for 100MHz less at 4600MHz at 1.36 Vcore, DRAM clocks were lowered at  2400MHz, to put less strain on the IMC. All this resulting in a small performance hit, however the temps were now hovering around the 70°C mark. Our particular ES sample clocked thus 100MHz better then the previous generation Sandy Bridge-E CPU. This of course is hit and miss as you might get an even beter or in worst case scenario a far worse clocking sample. One always should closely monitor voltages and temperatures of your own setup while stability testing!

 

 

Without any surprise, of course depending on the cores loaded, this i7-4960X CPU is a real powerhouse! Overclocking mileage can vary, most of the ES ( Early Samples) got easily up to 4600-4700MHz on air, retails in general are better. But again it's all about luck to find a good clocking, low voltage and low temperature operating processor. For the extreme overclockers we add to that list: scaling with cold and voltage, a good IMC under cold, no Cold bug or Cold boot Bug plz :)

 

 

 

 

Extreme Overclocking

Sandy Bridge and Sandy Bridge-E must have been the most boring processors ever developed for the extreme cooling guys. The architecture scaled hardly with added cold or voltage. They were great news for the watercooling guys as some could run their cherry CPU at higher clocks on a high end waterloop,  as some could do at -50°C. With Ivy Bridge Intel gave the overclocking community the means back to go flat out and really cool these processors to explore the maximum frequencies.

We tried the I7-4960X ES on the ASUS Rampage IV Extreme motherboard and got pretty decent results. As we heard some unconfirmed news of degredation issues at 1.8Vcore or higher, we opted to keep the Vcore at 1.75Vcore max.

 

 

  

 

We tried to run 3DMARK Vantage as this benchmark really scales with added cores and CPU speed and got almost 61K in the CPU test. The 4960X was running at  5700MHz at 1.74Vcore and ran without any hickup through the entire benchmark. We kept the CPU pot full of LN2 as it just evaporised... However we ran into some DRAM instability, sometimes posting in a Triple or Quadruple DRAM configuration. Changing voltages, adjusting timings, dropping DRAM multipliers didn't help 100% to solve the problem. We are suspecting the IMC coldbugged. Further testing might reveal more info on the matter.

 

 

 

 

One of HWBot's CPU test is WPrime, the 32 version was over at around the 3 second mark with the 12 cores in action, the 1024 test took around 86 seconds to complete. Take note that these benches were ran on the review Operating System and thus are not tuned for maximum performance :)

 

 

The Intel XTU test  was a whopping 1790 score .  The last test we did with this CPU was the UCBench test and got a new team and World record set at a mere 5737MHz. The CPU for sure has got a few more MHz left, but we try to spank it some more with some multi GPU configurations in the upcoming weeks.

 

Conclusion

The logic step to move on to the 22nm process, even for the high end Intel CPU gamma, was just a matter of time. Due to the small observed performance increase, we would label this new 22nm generation as a logic evolution. The most shocking performance step I witnessed was when Intel launched their so "called mainstream platform" with the Sandy Bridge processors. In many applications outperforming the high end LGA 1366 processors, without breaking a sweat. For daily users these were a pure treat: the K skew models could be overclocked to 5Ghz on air and even beyond on watercooling. The latter speeds however just for benching purposes. The hexacore versions, aka Sandy Bridge-E, were a solid replacement for the LGA1366 Gulftown processors. Now with Ivy Bridge-E, Intel continues to improve the performance per clock, allowing again better performance then before. More important is that it all can be achieved at a lower power consumption.

While the lack of native USB 3.0 support and especially Thunderbolt might be a setback for some end users. For once it's nice to get a new CPU generation which does not require a motherboard/chipset upgrade to demonstrate it's performance. And yes the new Ivy-E does perform. Sometimes achieving a 5-10% performance improvement over the predecessor, out of the box. Keep in mind that the i7-4960X runs at  slightly higher clocks then the i7-3960X. As mentioned before, due to the 22nm manufacturing process, the power consumption is reduced with about 20-30W under load.

 

 

One of the main concerns we had with the Sandy Bridge-E generation was that the DRAM speeds were usually limited to 2400MHz or even lower for daily stability when running high quantity kits. The 32GB 2666C11 CORSAIR Vengeance Pro kit was a test for the Integrated Memory Controller of the i7-4960X ES. Though even with tighter timings, the CPU remained stable during our test suite. With lower quantity aka 4GB Dimms 2800MHz or even higher is perfect possible. Ofcourse it's not all about speed and still a balance has to be found between MHz and DRAM timings to optimize the efficiency of the platform.

Intel refers to this CPU as the ultimate Gaming CPU. Indeed this is correct if one wants to build a multiple GPU and unlimited budget PC. However most game PCs we spot at LAN parties, mostly have one, max two graphics cards. For Gamers it's better to step down to the more mainstream Haswell or even Ivy Bridge CPUs, as mostly games do not even utilize more than 4 threads.  However the  quad core i7-4820K CPU might become a very interesting Processor. It bridges higher affordability with the multi GPU ability of the LGA2011 platform. The hexacore Ivy-E versions are more targeted at those that need all the crunching power they can get in a desktop PC. People that do a lot of video rendering, photo editing stuff, infact all programs that can properly address the available cores.

For those that don't care to cash out and want the fastest hexacore crunching processor out of the box, this i7-4960X perfectly fits that bill. For most, the i7-4930K will be more than fast enough or as fast, when slightly overclocked, this at a far better price/performance ratio. Once we get the 4820K and 4930K samples in the [M]Lab we will continue to test these new Ivy-E CPUs and might bring a more focused article versus the other more common CPUs on today's market.

Die hard Benchers are obliged to invest in one of the two six core versions to max out the scores in some 3D tests. If all CPUs can do at least 5600MHz under cold, it will reflect the performance of a 200MHz higher running Sandy Bridge-E CPU. A must buy then, let the CPU binning game start (again) to find them +6GHz gems.

Intel keeps on sucessfully improving their CPU lineup, no focus alike the competition on raw MHz. In 2013 it all evolves around efficiency tuning and reducing the power consumption. The price tag of the Extreme edition CPUs is always around the same 1000 dollar mark. Most will skip this CPU for the far more affordable 4930K edition as the overclock potential will be as good and in some cases even better. Bang for the buck the 4930K is unbeatable for those that really require heavy duty six core action.

To sum it up, with Ivy Bridge-E, the performance and power consumption improvement is present, nothing earthshattering, nevertheless again a solid and nice evolution for the Intel lineup.

Let's await the Tock revolution next year...

 

leeghoofd logging off...

 

Thanks to Jan, Kristof and Kim from Intel for providing us with the i7-4960X ES sample

 

 

To Sven from ASUS for the Rampage IV Extreme motherboard

 

Harry, Rick and Jake from Corsair for the Dominator Platinum and Vengeance Pro kits

 

 

And the crew from Tones.be for the years of support.

 

 

  翻译: