Haswell-E – Intel Core i7 5960X Review
The core factor
Well, dear boys and girls, it is our special pleasure to share a historical moment with you. Of course, some of you probably got bored with the MHZ and core count rush, but the launch of the first 8 core desktop CPU is nevertheless a very important moment in the history of computing. And not because it has 8 cores, but because one year ago the mobile market took lead for the first time when it comes to core count, launching the first 8 core chipsets.
It may not seem like a big deal but until now processing power was always a thing reserved for desktop computers, transistor count wise, MHz wise and core count wise. Well, the mobile chipsets cannot come close to desktop CPU’s when it comes to transistor count and frequency, but when it comes to core count there was a period of time where mobile chipsets had more cores then the high-end desktop CPU’s.
When you think about it, for more then 20 years core count was not an issue. Of course, the transistor count grew constantly according to Moore’s Law, we saw the progress from 4 bit CPU’s to 8, 16, 32 and finally 64 bit CPU’s, the power consumption also grew and of course the performance grew at an exponential rate. However, until 2005, when AMD launched the first dual core desktop CPU and Intel fought back with their Pentium D series, the desktop CPU’s had only one physical core.
Starting with 2005, multi core computing became a hot subject for the desktop market so the two CPU giants started the core count race. 2006 brought the first Intel native dual-cores (two cores on the same die – Conroe) and also the first quad-core (Kentsfield) which were made of two Conroe dual-cores on the same PCB.
Yorkfield integrated two Wolfdale dual-cores on the same die and Nehalem brought back the concept of hyper-threading. Gulftown was the first 6 core / 12 thread desktop CPU from Intel and starting with that moment (2010) we had to wait for more then 4 years for a higher core count desktop CPU. Both Sandy Bridge E and Ivy Bridge E had 6 cores and 12 threads and Intel didn’t consider we need more on a desktop… until now.
Well, the beast Intel Core i7 5960X puts an end to our wait integrating 8 cores and 16 threads in the same die. But the good news don’t stop here – unlike SB-E and IB-E, Haswell-E does not integrated more cores from the last generation, and more than that, with the launch of Core i7 5960X and X99 we also get a new type of RAM, DDR4. Therefore, it is pretty obvious that we have a lot to talk about today, so let’s get to work!!!
Haswell-E is a big step forward for the desktop market, bringing 8 cores along with the support for the new DDR4 RAM. However, the architecture is the same we already know very well from the mainstream Haswell, so there is no major news regarding architecture.
Even though we have heard rumors regarding a native 12 core design, the desktop Haswell-E has a native 8 core / 16 thread design, with 20MB L3 cache shared by all cores. The manufacturing process is the same 22nm with tri-gate transistors, and the 2.6 billion transistors fit comfortably in 356 mm2. IB-E had 1.86 billion transistors and a 257 mm2 die, while SB-E had 2.27 transistors and a big 434 mm2 die size.
OF course, when you squeeze 2.6 billion transistors and 8 cores into a small silicone die the clocks cannot go as high as the mainstream quad-cores, so Intel Core i7 5960X has a base clock of 3GHz, going up to 3.3GHz on 8 cores and 3.5 GHz when it only uses 1 or 2 cores. Compared to Core i7 4960X, which has a base clock of 3.6GHz for 6 cores and Core i7 4790K which has a base clock for 4GHz, the stock clocks we see on 5960X are really not that impressive. However, clock speed is not everything… at least not always.
The PCI-Express controller is similar with what we could see on Ivy Bridge-E, namely 40 PCI-E 3.0 lanes, more then enough for multi-card configurations (16x/16x, 16x/16x/8x or 16x/8x/8x/8x. Also, Haswell-E is the first desktop CPU that supports DDR4 in quad-channel configuration.
However, like any new standard, the DDR4 RAM will not impress us at first and their true potential will become obvious once the technology reaches maturity. The DDR3 technology has reached complete maturity but we should not forget that in the beginning DDR3 came in 1066MHz flavors while DDR4 is starting from 2133 MHz. The DDR3 RAM reached 3100MHz on some kits, while DDR4 can only reach 3000MHz at the moment, with huge timings.
However, this will change with time and we will see DDR4 kits that come from the factory with 4000MHz ratings and much more affordable prices compared with what we can see now. What we have to remember is that at the moment we cannot compare DDR4 performance with DDR3 performance because in the beginning they will most likely loose to the 7 year old mature technology. What we can see for sure is that DDR4 needs less power compared to DDR3 (1.2v).
However, the news don’t stop here. With Haswell-E launch we also get a new chipset, which offers 14 USb ports (6 x USB 3.0 and 8 x USB 2.0), 8 PCI-Express 2.0 lanes and 10 S-ATA 3 ports. And even though the socket still has 2011 pins, it is called LGA 2011 – v3 and it is not compatible with older LGA 2011 CPU’s. And one some extreme overclocking motherboards we will also see a similar socket compatible with Haswell-E which has 2084 pins…
Models and pricing
In the beginning we will have 3 new Haswell-E CPU’s: Core i7 5960X, Core i7 5930K and Core i7 5820K, which have the same price scheme with the CPU’s they are replacing. However, the only the Extreme version (5960X) has 8 cores – the K models only come with 6 cores and 12 threads.
Also, beside the clock difference, Core i7 5820K only has 28 PCI-Express 3.0 lanes, which means you will not be able to use it for 4-way multi-card configurations. It will however be able to support 2-way (16x / 8x) and 3-way (8x / 8x / 8x) configurations. Another interesting thing with Intel X99 is that the M2 port is connected to a PCI-Express 3.0 lane straight from the CPU which means we will see better performance numbers for such devices on X99.
The CPU’s launched today use the same Intel Turbo Boost 2.0 technology we saw at work in the last couple of years. That means that even if the base clock is really not that impressive, the CPU’s can work at a much higher clock, depending on the load. For instance, i7 5960X can go as high as 3.5 GHz on one core while i7 5930K can go up to 3.7 Ghz on one core. Maybe the most interesting version is the Core i7 5820K version, which costs 50 USD more than 4790K but it comes with 6 cores / 12 threads, a base clock of 3.3 GHz (3.6 GHz Turbo), 15MB cache L3 and DDR4-2133 quad-channel memory.
As usual, today we will test the very best of the best, the flagship model Intel Core i7 5960X, which you can admire in the pictures bellow…
ASUS Rampage V Extreme
Evidently, a new Intel architecture involves a new chipset (X99), a new socket (LGA 2011-v3) and obviously a new ASUS motherboard from the Rampage Extreme series. In this case we are talking about Rampage V Extreme, a motherboard dedicated to enthusiasts but mainly to extreme overclockers. The motherboard is shipped in a suggestive package, decorated with the specific Republic of Gamers logo, and it comes bundled with everything we would ever wish for in order to build a high-end rig.
Among the accessories we can also find the inevitable overclocking gadget called OC Panel, the 3T3R antenna for the onboard Wi-Fi 802.11ac card, and also an extra backplate (X-Socket II) that allows mounting of LGA1366 coolers or LN2 / DICE pots.
The motherboard itself retains the design used for the Z97 ROG series, having in return the specific connectivity options of the X99 PCH. Thus we have at our disposal a total of 12 SATA 3 ports (10 native, and 2 thanks to an additional ASMedia controller), out of which 4 ports can be converted when required into 2 SATA Express ports. We also have an M.2 port that can use 4 PCI-Express 3.0 lanes linked directly to the processor, a feature made possible on most Intel X99 motherboards.
Thanks to the 40 PCI-Express 3.0 lanes found in the 5960X and 5930K processors we could easily build configurations up to 4-way SLI / CFX, Rampage V being fit with 3 correctly spaced slots for this purpose. The 8 DDR4 memory slots can host up to 64GB running at frequencies of up to 3300MHz. The sound board is the same SupremeFX 2014 that we mentioned in other articles, so a Realtek ALC1150 with some touches.
The ‘crazies’ corner from the right side of the motherboard includes the Power and Reset buttons, the LN2 mode jumper, the Slow-Mode switch, and also the measuring points found directly underneath the 24-pin ATX connector. There we can also find a DIP switch which can be used to turn off any of the PCI-Express slots, and also the usual Debug LED. On the Rampage V Extreme we can find two new buttons that can help us reach the highest overclock: Safe Boot (if the settings were too aggressive, by pressing this button we can enter the BIOS using default settings but without losing what we had previously configured), and ReTry blabla.
We left for last the most important new feature brought by ASUS, namely OC Socket. By taking a look at the pictures down below we can easily notice that Haswell-E (left) has more pins than Ivy Bridge-E (right). This pins are not documented by Intel, therefore Shamino – the man behind the latest ASUS ROG motherboards – ordered a special socket from Foxconn that contains the missing pins from the original LGA 2011-v3. This new socket is internally named LGA 2084 because it has 73 more pins, and it will initially be available only on the ASUS X99 motherboards.
After 4-5 months of work in which the purpose of the missing pins was researched, the ASUS motherboards can boast with the following achievements brought exclusively by the use of this socket: more pins for powering the processor, higher memory and uncore frequencies attainable but also extra monitoring of the voltages generated by FIVR.
On the backpanel we can find 12 USB ports (10 x USB 3.0 and 2 x USB 2.0), a PS/2 port for keyboard / mouse, one RJ45 port (Intel i217-V), the connectors for the onboard WiFi 802.11ac and also for the SupremeFX sound board. The IO panel is completed by the CMOS Reset button and also by ROG Connect, both of which are illuminated.
Processor’s VRM is easy and solid, built with 8 phases controlled by a possibly IR3580 renamed controller. With Rampage V Extreme, ASUS completes the transition from NexFET to the more compact (with drivers included) PowIRstage built by International Rectifier (IR3555M) with a maximum amperage of 60A. The cooling system for the PowIRstage ICs is based on the DirectCU technology, which in my opinion is the only good application of this technology since we’ve already noticed it doesn’t work that great for video cards 🙂 The Microfine Allow Chokes coils are also something new for Rampage V Extreme, and they too are cooled.
ASRock X99 Extreme4
Further on we change our attention towards a motherboard aimed more at the mainstream market, namely ASRock X99 Extreme4. The motherboard was bulk, therefore we are going to skip the packaging and contents section since we didn’t had the pleasure of receiving them.
On the X99 Extreme4 we can mount up to 3 video cards in multi-card setup while the SATA connectivity is limited to the 10 SATA 3 ports provided by the Intel X99 PCH (more than enough in my opinion). ASRock did not fit this board with any SATA Express ports, but instead there is an M.2 port connected directly to 4 PCI-Express 3.0 lanes provided by the processor (Ultra M.2). The onboard sound card is Purity Sound 2 that we mentioned in previous articles, so a Realtek ALC1150 with some touches.
We can’t even wish for onboard Power and Reset buttons in this market segment, so we would settle with the Reset switch found on the backpanel. There we can also find a PS/2 port for keyboard and mouse, 8 USB ports (4 x USB 3.0 and 4 x USB 2.0), an eSATA port, one RJ45 port (Intel Gigabit Lan) and also the connectors of the onboard sound card.
Processor’s VRM is based on an Intersil ISL6379 controller that can drive up to 6 phases, but through the use of 6 phase doublers we reach a total of 12 phases. Those are based on Fairchild Ultra Dual-n MOSFET that integrates 2 MOSFET transistors in the same package.
GIGABYTE X99 Gaming 5
The last motherboard presented today comes from GIGABYTE and is dedicated to the gaming segment, thus the features of this board are meant to lure consumers of this market segment. The box is of medium dimensions with an attractive design based on shades of black and red.
In the package we can find everything needed, personally I’ve been pleasantly suprised by the I/O panel illuminated in red, and also by the textile sleeved SATA cables. On the other hand I couldn’t understand the purpose of the 8-pin EPS adapter since even if I connect three 8-pin EPS connectors there’s still only one on the motherboard that can sustain the same electrical current.
X99 Gaming 5 can host up to 4 video cards and is shipped with all the needed bridges. The sound board was also modified to please the gamers, using a Creative Sound Core 3D DSP that runs through an OPAMP Burr Brown OPA2134 that can be replaced if we’d like another timbre. For a better channel separations there we used separate layers in the PCB for each of the channels, while the tension filtering is done with audio-grade Nichicon capacitors.
With regard to SATA connectivity, X99 Gaming 5 is restricted to the 10 SATA 3 ports made possible by the X99 PCH, out of which 2 of them could be sacrificed in case we’d need a SATA Express port. We have at our disposal 2 M.2 ports, out of which one runs strictly in PCI-E mode designated for use with an Wi-Fi card that unfortunately does not come in the package.
On the backpanel we can find 4 USB 2.0 painted in yellow to highlight that the tension supplied by them is adequate for a pretentious DAC, 6 USB 3.0 ports (the white one can be used for quick BIOS flashing), PS/2 ports for keyboard and mouse, one RJ45 port (Killer E2200) and also the golden connectors of the onboard sound card.
Processor’s VRM is simple and solid, built with 6 phases based on PowIRstage IR3556M ICs (60A) from International Rectifier and solid Copper Bussman coils. The controller is also made by International Rectifier, in the form of IR3580.
Corsair Vengeance LPX 16GB DDR4-2666 CL15
Of course, we couldn’t get to work without taking a look at some brand new DDR4 kits. The first one comes from Corsair and it is a member of the Vengeance LPX family, dedicated to the mainstream market. From my point of view the new LPX series looks much better then the older Vengeance models, and the heatsinks are more reliable and well attached to the IC’s.
The package was a bit ruffed up by the not so reliable UPC guy, but fortunately no harm came to the modules. Corsair Vengeance LPX 16GB DDR4-2666 CL15 are certified to work at 2666 MHz with 15-17-17-35 timings for 1.2v. This kit is built around SK Hynix IC’s, manufactured on 20nm (MFR) and placed in single-sided configuration (4GB modules).
Corsair Dominator Platinum 32GB DDR4-2666 CL15
The second kit that got our attention is also from Corsair, but this time we are talking about a premium kit from the enthusiast Dominator series. The specs are identical but the modules have double density (8GB per stick with a total of 32GB for the whole kit).
This kit is also based on 20nm Sk Hynix IC’s but the double-sided configuration means that we will see better performance due to interleaving. As far as the built quality is concerned everything is top-notch, as you would expect from a Dominator kit. The kit is also compatible with Corsair Link and the lightning color can be configured as you please.
Crucial 32GB DDR4-2133 CL15
The third kit we are looking at today comes from Crucial and is a member of the Value series, which we can easily notice due to the lack of heatsinks. The specs are also value, the 4x8GB modules being certified for DDR4-2133 CL15 with 1.2v.
Many of you may ask why we took the time to talk about a value kit. Well, let me put it like this – at the moment there are only two manufacturers for DDR4 IC’s – SK Hynix which are implemented by most manufacturers and Micron, which can also be found on Crucial kits at the moment. If you remember the golden days of DDR2, you will have a nice pleasant nostalgia sensation when you see the new DDR4 Micron D9RGQ IC’s. Now, all we can wait for is the Samsung based DDR4 kits.
|CPU||Intel Core i7 5960X||Intel Core i7 4960X||Intel Core i7 4790K|
|Motherboard||ASUS Rampage V Extreme||ASUS Rampage IV Black Edition||ASRock Z97 OC Formula|
|VGA||Galaxy GTX780Ti HOF v2|
|Cooling||Enermax LiqMax 120X|
|RAM||Corsair Vengeance LPX 4x4GB 2666MHz CL15||Corsair Dominator Platinum 4x4GB 2400MHz CL9|
|SSD||Intel SSD 730 240GB|
|Case||Dimastech Easy BenchTable v2.5|
|Temp ambient||27 oC|
|Ram clocks||DDR4 2133MHz||DDR3 2133MHz|
|Timings||12-12-12-28 1T||9-11-11-27 1T|
To assess Core i7 5960X’s performance I’ve used as comparison the high-end of the Ivy Bridge-E platform (Core i7 4960X) as well as the fastest processor that uses the refreshed Haswell core, aka “Devil Canyon” (Core i7 4790K). If the first mentioned processor is delighting us with 8 cores running at a mere frequency of 3.33GHz (3.5GHz with Turbo Boost), the aforementioned comes with 6 cores using a last-gen architecture (3.7GHz with all cores active, 4GHz on just one core), while Core i7 4790K has “only” 4 cores but gains ground due its generous frequencies (4.2GHz on all cores, 4.4GHz on just one).
For Core i7 5960X I’ve used the ASUS Rampage V Extreme motherboard, for 4960X I’ve went for the excellent ASUS Rampage IV Black Edition while for 4790K I’ve chosen the Z97 OC Formula from ASRock. For the DDR3 platform I’ve used the Corsair Dominator Platinum 4x4GB DDR3 2400C9 memory running at 2133MHz with 9-11-11-27-1T latencies, whilst for the DDR4 platform I’ve used Corsair Vengeance LPX 16GB 2666MHz, in order to maintain an identical quantity as with the other platforms. The frequency was also set to 2133MHz to have a comparison as close as possible, with latencies being set to the minimum allowed by this kit (12-12-12-28 1T).
Further on, Galaxy GTX780Ti HOF V2 took its well deserved place on the testing platform, being the fastest single-GPU video card on the market, its frequency going up to 1200MHz. The cooler used for the 3 processors was the already known Enermax LiqMax 120X, power supply was the same Seasonic P1200 as before whilst the Intel SSD 730 240GB happily made our very fast testing platform complete. The operating system used was Windows 7 SP1 Ultimate x64, with all updates installed. Also, we’ve installed the latest drivers for our motherboards, as well the newest NVIDIA driver, Forceware 340.52.
The methodology specific to each test will be detailed, where it applies, in the chapter dedicated to that specific test. As always, we’ve been trying to obtain a full set of tests that will answer the questions of a range of users as wide as possible. Therefore, we have synthetic tests as well as real applications, like a series of benchmarks dedicated to overclockers. In gaming scenarios we’ve used two resolutions, one chosen to isolate the processor – 1024×768, noAA, and one chosen to fit a real life scenario, 1920×1200, 8xAA (4x if limited by the game), with maximum details, which is how most owners of high-end platforms play.
Obviously, we have in tests both single and multi-threaded applications, applications that are used by millions of PC users around the world (such as IrfanView), but also applications targeted towards professionals (3D Studio MAX), games that proved themselves over the time as benchmarks, but also some of the newest releases, all of these chosen to help you get a full picture of the products we’re testing today, for your usage scenario.
All processors were tested at their default frequencies (with all supported functions active, such as Intel Turbo Boost, EIST etc). Also, I’ve included in this article a short air overclocking study, to see how both Core i7 5960X and DDR4 perform when overclocked.
We start with a real-life application, heavily optimized for multi-threading, namely Handbrake, which reached version 0.9.9. To measure the performance of the 3 processors tested today in a customary scenario, I’ve converted an 1GB .VOB file using the Normal preset, and timed the operation.
5960X obtains better results, but we can’t really say it started strong. Probably the load is not enough to showcase the power it possesses and a 4K encoding would have been better suited for this.
x264 HD Benchmark 5.0.1
On the other hand, x264 HD is a much better optimized application for multi-core processing, which is why 5960X scores significantly better than its opponents.
Adobe Photoshop CS6
Having reached version CS6, Adobe Photoshop can’t miss from our suite of real life applications, being probably the most used photo editing software, by both professionals and home-users. To measure the performance of the processors tested today I’ve used a photo in TIFF format, with a resolution of 3720 x 3796 and a file size of 40MB. I’ve applied the Radial Blur filter, using “Amount – 100%”, “Blur Method – Spin” and “Quality – Best”. After that I’ve applied the Extrude filter, using the defaults. To eliminate the impact of the storage device used (even though it is a SSD), after applying the filter I’ve used the “Undo” and “Apply Last Filter” commands and timed the operation. The result from the table is the sum of the time needed to apply the Blur filter and the time needed to apply the Extrude filter.
For today’s test, Photoshop had 2GB of memory allocated, and the CUDA acceleration was disabled. In the end I’ve obtained an abstract image, perfect to frame for display in the lobby of a pretentious offices building.
Also the case with Photoshop, things are crystal clear – the application is optimised for parallel processing, therefore 5960X scores much better than 4790K and 4960X even though its frequency is significantly lower.
If the Photoshop test is targeted towards those that actively use this application, or even professionals, the test I’ve realised using IrfanView is addressed to a much wider audience. For this test I’ve used the latest IrfanView version and a set of 50 photos with a resolution of 3000 x 2000, of approximately 2MB each. I’ve created a test set using the Batch Processing feature, which I ran on each processor. Thus, using the aforementioned feature I’ve resized the 3000 pixels wide pictures to 1280, adjusted colors, changed orientation to horizontal and rotated the images to the left. Also, I’ve used the Sharpen filter with +15 value, +2 Contrast and +5 Blur. Last but not least, I’ve added a watermark of 200×200 pixels on each image. In the end, from 50 images summing up ~100MB I’ve obtained a set of 50 images ready for web, summing 6MB.
Of course, I’ve chosen a wider range of tests to load as much as possible the processors used in this test, chances being that you will never use as many effects in a single batch. However, I consider that our test set showcases what most users that use simpler applications to touch up their pictures do, applications such as IrfanView, ACDSee, XnView etc. I’ve used IrfanView because of its Batch Processing feature that allows saving settings, and also because it is one of the most used tools for digital images management.
Oh well, IrfanView is a frequency dependent application and not of core count, therefore 5960X places last, having the lowest frequency.
3D Studio MAX 2015
To test the performance in 3D rendering, in a real scenario, I’ve requested the help of a passionate member of lab501’s forum, which created a special scene that I’ve rendered using the 3D Studio Max renderer also used in production, namely Mental Ray. The scene as well as the benchmark’s usage instructions can be found here; a lower time represents a better result.
In 3DStudio Max 2014, Haswell-E brutally replies, showing us what it’s capable of when using a truly optimized software dedicated to professionals.
WinRAR is memory bandwidth dependent and it is the first test in which DDR4 in quad-channel shows its teeth – 5960X decisively wins.
Same thing applies to 7Zip, which scales very good with core count.
When talking about single-core performance, 4790K is the clear winner due its noticeably higher frequency. Table turns in the multi-core test where 5960X easily leads the pack.
In Cinebench the situation is attributed to the number of threads used, just as with POV Ray. After all…. it’s still raytracing we’re talking about.
PCMark 7 declares 4790K the winner, which proves itself faster in the multimedia and office tasks. We are talking about a daily tasks simulation after all, and in this case we are not using the full potential of the 8 cores / 16 threads.
AIDA64 4.6 Cache and Memory
This is our opinion regarding memory bandwidth of the newest processors from last year. Even though the figures don’t indicate, I maintain my opinion. Of course, at launch time, when we are talking about DDR4 2133 things don’t seem to be much better than IB-E. Keep in mind though that in the memory manufacturers roadmaps there are DDR4-4xxx kits….We can thereby understand that this first battle between the matured DDR3 and the newly arrived DDR4 is not representative for the future of this technology.
AIDA64 4.6 CPU
The maturity of the DDR3 memory and the increased frequency can’t help the 4790K or 4960X to keep pace with the 8 cores of 5960X.
Obviously, WPrime is dependent of threads count, therefore 5960X wins, closely followed by 4960X.
In SuperPI 1M frequency has its say thus 5960X places last. Keep in mind though that it only reaches 3.5GHz on a core, opposed to the 4.4GHz in the case of 4790K…
As expected, same situation in PiFast….
He, he, he… I think our friend Jean Paul (creator of the chess tests) will be very happy when he sees this results.
3DMark 2011 marks 5960X as winner at the end of the benchmark due its better scores in the CPU tests but 4790K proves itself faster when talking strictly about 3D performance.
3DMark Fire Strike
…and the same thing is happening also in 3DMark Fire Strike.
Yes, I know, no one plays anymore FarCry 2, true that. Just as no one plays at 1024 x 768. However beside the real-life situations (and we have plenty in the gaming chapter, as you will see), we must also use a known reference system, which proved itself reliable over the course of time, and FarCry 2 is one of the most representative DX10 games from this perspective. Also, beside the real-life situation (1920 x 1200, 8xAA), we must also isolate the processor, and for that purpose we are using 1024×768, noAA. In real-life situations, the difference between the 3 processors is minimal…
Alien vs Predator
In AvP we’ve obtained the same FPS in 1920×1200, and in 1024×768 the differences are pretty marginal, although 4790K has the advantage once again.
Just as with AvP, in Crysis 2 we’ve reached the same FPS in 1920×1200, and in 1024×768 the differences are insignificant.
Here we can also see the same, extremely close results in gaming conditions, with a small advantage for 4790K.
To determine the power consumption, I’ve used a power meter while running the benchmark used for AvP, 1920×1200, 4xAA, but also Prime 95 Blend.
The power consumption displayed in the table is representative of the whole system and it’s the highest reading recorded by the power meter during the given test. In other words, the average consumption is much lower, but we are interested in the peak, worst case scenario, the maximum stress that our power supply unit is going to take. Evidently, the maximum consumption of the system can be higher when running multiple stress-testers for each component (for example FUR for video card, Prime95 for processor, HDTune for HDD, all at the same time), but we are interested only in the maximum consumption in real scenarios, in our case gaming, but also the maximum consumption of the platform when fully loading the processor.
I must admit that I didn’t expect the Core i7 5960X to use less energy than Core i7 4960X , but if we spend a minute to think about it, it’s not all that fantastic. We are talking about the consumption of the whole system, and a quad-channel configuration based on DDR4 memory running with 1.2v can save a lot of energy when put against a system with DDR3 memory using 1.65v. More than that, we are dealing with a processor based on Haswell that has numerous energy saving optimisations that runs at much lower frequencies and is only using 1.017v in LOAD.
The result? In Prime95 Core i7 5960X manages to use 39W less than Core i7 4960X even though it has 2 extra cores…wow. In the gaming tests it manages to use only an extra 2W more than Core i7 4790K which has half the core count of the 5960X monster. Well to be fair, Core i7 4790K does run almost 900MHz higher and is powered with an extra 100mv, but it’s still a notable performance for the first desktop dedicated octa-core from Intel.
In terms of temperature reached during use, the graphics above are purely orientative, because each Intel processors generation report temperatures differently. The temperatures that can be seen above are not representative for what a normal system can reach, inside a case. When talking about a poorly ventilated case we could see higher temperatures, whilst in the case of a properly ventilated case we could see lower temperatures. To measure the temperature I’ve used RealTemp 3.70 and I’ve recorded the temperature reached under LOAD, after a session of Prime 95 64bit, respectively Alien vs Predator, using the same settings as in the power consumption tests. Ambient temperature was maintained around 27 oC over the course of the tests and there were no other fans on the benching table besides the NH-U12P and the video card’s fan.
Oh well, we can spot improvements in the temperatures department as well. The generous size of the die coupled with the soldered IHS makes the Core i7 5960X able to reach very good temperatures, with 10 Celsius degrees lower than 4790K, which is a simpler design with 4 cores and thermal grease between the IHS and die. It reaches temperatures with 3-4 degrees lower than even his predecessor, Core i7 4960X. Very impressive, but now let’s see how we stand at the overclocking chapter….we have headroom or what? 🙂
Core i7 5960X – overclocking
The first thing I wanted to see was the maximum frequency I could get that can be used 24/7. As I didn’t had much time on my hands and I intend to analyze Haswell-E overclocking in a separate article, I runned just the 64 bit version of Cinebench R15.
With a voltage of 1.37v (the default voltage is 1.017v) I got 4480MHz in full stability conditions, temperatures increasing pretty much as it can be seen in the screenshot. The memory increased up to 3000MHz with the timings 15-15-15-30 1T while the uncore was set at 3860MHz. Keep in mind that these are very early results obtained in a very limited timeframe. That being said 8 cores / 16 threads running close to 4500MHz is not too shabby.
Intel Core i7 5960X displays an incredible amount of power in multi-threaded optimised applications, although the frequency at which it runs is significantly lower than its 6 and 4 core counterparts. For the professionals that could benefit from the use of the massive parallelism, processor’s cost, of the X99 motherboards and that of the DDR4 memory is not big at all and it will quickly write-off. Technology enthusiasts have something new to dream about and can ask Santa Claus in a few months. To be honest, I was pleasantly surprised by the much lower power consumption as opposed to Core i7 4960X but also by the excellent temperatures reached in tests (10 Celsius degrees lower than Core i7 4790K).
Core i7 5960X is everything we could wish for but few of us would afford it, especially since the motherboards won’t be exactly cheap, and the DDR4 memory kits are pricier than their weight in gold, at least for now. If we don’t want to run exotic 4-way configurations, then Core i7 5820L comes with 28 PCI-Express 3.0 lanes, costs only a little more than Core i7 4790K and has 6 cores / 12 threads.
After the launch of Ivy Bridge-E where I was left with a bitter taste not only because Haswell was already on the market but also because the performance gain was not extraordinary, Haswell-E puts on display a strong representation from Intel, scoring high by performance, power consumption, temperatures but also good overclocking potential in proportion with its complexity.