Ivy Bridge 22nm Review: Intel Core i7-3770K and i5-3570K - BeHardware
Written by Guillaume Louel and Marc Prieur
Published on April 23, 2012
The Tick-Tock strategy
A little more than fifteen months after the launch of the Sandy Bridge architecture, Intel has now launched the somewhat delayed new generation of desktop processors, the Ivy Bridges, also designed for socket 1155. These processors have the challenging task of eclipsing the Sandy Bridges, which have been attracting plaudits since their launch in January 2011, in spite of the major delay linked to a bug on the SATA ports of the Intel Series 6 B2 chipsets. Have they managed to do so?
The Tick-Tock strategy
Launching a brand new processor architecture on a new fabrication process is a complex task and can cause serious delays due to the accumulation of problems on each side that can be difficult to locate. With a view to removing uncertainties, Intel has been working with a different development strategy over the last few years.
Every two years, therefore, Intel launches a new processor architecture on a fabrication process that has already proved itself. Thus, at the beginning of 2011, Intel launched Sandy Bridge, manufactured at 32nm on a mature process. This architecture has now been in production for more than a year. This is what’s known as the ‘Tock’. Also, every two years, the alternate years, a new manufacuring process is introduced and combined with the architecture launched the previous year. This is what’s known as the ‘Tick’ and this is what we have this year with the Ivy Bridge processors, engraved at this new 22nm process.
What we have here however isn’t simply a die shrink as we’re moving from 995 million transistors on Sandy Bridge to 1.4 billion on Ivy Bridge. Intel has however changed how it counts transistors. Without giving any more detail, Intel says that Ivy Bridge has 20% more transistors than its predecessor. Die size has gone down from 216 mm2 to 160 mm2.
As you can see on this scale illustration comparing the Sandy Bridge (above) and Ivy Bridge (below) dies, the additional transistors have mainly been added to the GPU. More than just a Tick, Intel is calling this a Tick+. As the integrated graphics part has historically been the weak link in Intel processors, adding transistors here seems, on paper, to make sense.
Ivy Bridge has therefore broadly speaking been designed along the lines of its predecessor, namely:
- Socket LGA 1155, with backwards compatiblility (via a BIOS update) with the motherboards launched in 2011 (P67/H67/Z68)
- A dual channel DDR3 controller as well as a 16-lane PCI Express
- A built-in graphics controller
- Three cache levels, the last (LLC) of which is shared with the graphics core
Several changes have also been made:
- Architectural changes to increase the IPC
- The memory controller officially supports DDR3-1600
- The PCI Express controller is now a Gen 3 type
- The IGP now offers DirectX 11 support and gives higher performance
- Support (under certain conditions) for three screens
At the same time as launching these processors, Intel also launched some new chipsets at the beginning of April, the Series 7s which support USB 3.0 natively. Along with the motherboards, these chipsets also support PCI Express 3.0 though some manufacturers were already marketing previous gen models that were announced as compatible. For more information on this, we refer you to our Z77 motherboard review.
22nm and Tri-gate
The arrival of this new process is thus the first major innovation of these Ivy Bridge processors. As usual Intel is first to the market with chips made at this manufacturing process. Making the engraving process finer improves what is called transistor 'performance'.
To recap, transistors are like switches. The gate serves as the switch button and instead of using your finger to turn it on or off, voltage is applied to the gate, which then allows, or doesn’t allow, the current to enter the transistor. On the visual below, you can see the voltage running in the channel (the current that the transistor allows through, like the current that lights up a light bulb) as a function of the voltage applied to the gate. At the bottom on the left, the transistor is switched off and at the top on the right it’s on.
When we talk about transistor performance we are in fact talking about the quantity of voltage you have to apply to get it to change state (inactive/active). With the Intel 22nm engraving, shown here in blue, this translates into a steeper gradient on the curve. The engineers then have several options. On the line on the right, there’s no change to the voltage required at the gate to turn the transistor on. The residual voltage in the channel (current leakage) does however drop, something that’s particularly important to keep down for mobile processor solutions.
On the line on the left however, the engineers can decide to maintain the same level of current leakage as previously. The result is that the voltage required for activation at the gate is reduced. This means you need to supply less voltage to the transistors and speed up activation time, making the transistor faster. With each new fab process, the engineers must therefore balance transistor performance against current leakage according to their needs.
The new 22nm Intel solution doesn't however simply innovate with respect to its engraving fineness but also the actual shape of the transistors. In a pretty radical move, Intel has become the first processor manufacturer to launch processors with transistors using a gate made on more than one plane and surrounding an elevated channel.
This is what’s know as FinFET technology, or as Intel calls it, Tri-Gate. Note that several fins can be used together to improve transistor performance.
Here Intel gives an example of the gains that can be obtained with Tri-Gate and which characterise this type of transistor. In effect, in this graph which shows transistor gate delay as a function of the voltage applied to the gate (the lower the voltage, the faster the transistor), we can see a huge performance gain at low voltages, which is very promising in terms of being able to create chips running at high clocks at the same time as maintaining low voltages. In a desktop processor this can mean that the chip will have a good propensity to undervolting.
Of course, the spec of transistors alone is only a small part of the equation. The variability of the fab process counts for a lot and we’ll check to see in practice what the gains announced translate to in terms of undervolting, overclocking and energy consumption.
In keeping with all 'Ticks', Ivy Bridge architecture is broadly based on its predecessor, Sandy Bridge. We therefore refer you to our previous article on this for the details. Today we’re going to concentrate mainly on the differences and innovations introduced with Ivy Bridge.
Numerous points in common
Seen from the top, the technical choices made by Intel for Sandy Bridge have been confirmed for Ivy Bridge. The first of these was the inclusion of what was historically known as the northbridge on the processor die.
Traditionally part of motherboards, this part of the chipset contained the memory controllers, PCI Express and, if there was one, the IGP. It is implemented on a single die in Ivy Bridge, with two or four CPU cores depending on the die model, a level 3 cache, the LLC, that can be up to 8 MB in size, a graphics core and the uncore part, which includes the DDR3 memory controller, control of displays, the southbridge link (via a DMI bus, which is the equivalent of a PCIe x4 bus), as well as the PCI Express x16 controller. All these blocks are linked by an internal ring type bus, which enables the sharing of the LLC between the x86 cores and the graphics core.
There are relatively few changes inside the cores themselves. There's no new AVX-style extension to the instruction set such as was introduced with Sandy Bridge (AVX2 will be ushered in with Haswell next year) but there have nevertheless been a few small changes.
First of all Intel has added a few instructions to convert 32-bit single precision floating point type data into a compressed Float16 format (1 sign bit, 5 exponent bits, 10 significand bits). These instructions (VCVTPH2PS and VCVTPS2PH) are available in 128-bit and 256-bit SSE/AVX vector variants. Note also in passing the introduction of new instructions that allow FS/GS segments, which are usually reserved for the operating system, to be read.
There is however a digital random number generator. This is known as a digital generator as the chip includes a source of entropy (the purely random part that certain enchryption tools simulate in the generation of keys requiring you to move your mouse around in all directions). Here Intel gives a speed of 2 to 3 Gb/s, which should provide good performance for the applications that require it. All this is contained in a functional block that is accessed with an instruction (RDRAND) that will thus be able to supply a random 16, 32 or 64-bit number on demand (conforming to ANSI X9.82, NIST SP800-90 and NIST FIPS 140-2/3 level 2).
Note finally a last change with respect to the instruction set, where there is improved performance for instructions that handle strings – optimisation of REP MOVSB and REP STOSB with improved speeds for memory blocks of more than 64 bytes. Intel says that it would like to delete the algorithms specific to each processor that are found in the libraries used by compilers or runtimes. This is an interesting step for the future (these algorithms that are optimised by the processor are often only partially optimised, creating performance differences that could be avoided – we refer you to our report on the subject), if indeed it is pursued going forward and if AMD also goes down this route. Up until now the REP MOVSB/STOSB instructions haven’t been the preferred route for these operations on K10 and later processors.
In addition to these changes in the ISA itself, Intel has also introduced some small optimisations in the pipeline to improve the IPC. Intel has thus improved the performance of divisions instructions and added improvements to detect and delete useless MOVs. Some changes with an impact on the IPC are not however necessarily included in the pipeline itself but directly in the uncore.
Improvements to the uncore
When it comes to latency, we noted progress both in respect of the LLC cache and the memory. Thus at 4.5 GHz, we noted 4.3ns for Sandy Bridge against 3.4 for Ivy Bridge.
With DDR3-1600 9-9-9 memory, the latency measured via AIDA64 drops from 45.1ns to 39.3ns.
Apart from its speed, LLC caching has also been reworked with the introduction of a process known as Adaptive Fill Policy, which impacts particularly on how the IGP and x86 codes share the common resource of the LLC. Intel says that it has worked on Sandy Bridge heuristics to optimise Ivy Bridge performance. This could reduce the effects of collisions that we noted at launch of Sandy Bridge when applications that tax both the x86 cores and the IGP are being used. We’ll check this in practice.
The LRU algorithm, which ages data in the cache, has become a bit more flexible, moving to two bits to improve granularity. The prefetch memory unit also now has a throttling mechanism which stops it hogging memory bandwidth when there are already too many memory accesses.
Note finally two additional changes in the uncore. The first concerns power-gating at processor level for the VccP used for DDR I/Os. When the highest C-states are being used (C3 package and higher), DDR I/Os are reduced. Intel hopes to obtain an energy economy of 100mW when the machine is at idle, which could be important on mobile platforms. Intel has made two changes with respect to the memory controller. The first is support for low voltage DDR3 on mobile Ivy Bridge versions and the second is “official" support for DDR3-1600 memory with two memory bars per channel. Remember of course, in our last review we used four bar DDR3 2133 motherboards with a Sandy Bridge processor without any issues.
In orange, you can see the changes made to the original controller (green). The changes in blue are to the buffers.
The last change concerns support for PCI Express 3.0, which has been added. In spite of what you might think, we haven't got the same implementation here as used by Intel for SNB-E (the X79 platform). Instead, it's based on the original Sandy Bridge one. Only the buffers developed for SNB-E have been used for the new implementation, which, according to the Intel engineers, should significantly improve performance in comparison to the SNB-E PCI Express 3.0 platform. If you remember, the raw performance figures that we measured at the time weren’t particularly great.
Variable TDP technology on mobile versions
Among the more unusual changes, variable TDPs have been introduced for mobile machines. In addition to a nominal TDP, there are also two distinct settings, TDP down and TDP up. The first can for example only be activated when the machine is plugged into a power source while the second represents a maximum energy economy mode. Intel hasn’t given many details on implementation yet and seems to have been introduced above all to give a bit more flexibility than Speedstep, which was the technology used for this up until now. Implementation of the TDP up/down modes will be at the discretion of partners and we’ll have to wait for the arrival of the first Ivy Bridge mobile CPUs to find out more.
A DirectX 11 GPU
While the improvements on the CPU side mainly concern implementation details, the changes made to the Ivy Bridge graphics core are much more significant. To recap, the Sandy Bridge family processors used one of two graphics cores, the HD 3000 or HD 2000, both of which were limited to DirectX 10.1. For 2012, Intel is finally giving official DirectX 11 support, which is great news.
In terms of the architecture, the Intel GPU has been broadly designed along the lines of the previous version with a few adaptations. Intel is above all putting the accent on architecture flexibility so that it can be adapted as well as possible to reductions in the number of shader unit blocks (Execution Units). While the HD 2000 and 3000 had six and twelve blocks respectively, the HD 4000 has sixteen EUs and the 2500 six.
DirectX 11 support means that there’s a tessellation unit as well as compute shader support, requiring several changes to be made to the EU blocks: the addition of a local shared memory as well as more flexible memory accesses thanks to scatter/gather operations. Moreover, Intel has optimised the EUs and increased yield by making the most of their processing unit parallelism.
There have also been some other small changes. There’s accelerated processing performance for the Geometry Shaders and the Stream-Out, while the setup part is able to eject the triangles that are out of the field of vision of the camera more rapidly. The texture units have also become more efficient and Intel has announced better quality anisotropic filtering. The last noteworthy architecture change comes with the introduction of a level 3 cache within the IGP so as to limit demands on the LLC. The size of this cache hasn’t been given.
Like SandyBridge, Ivy Bridge also has a block within its IGP dedicated to video compression and decompression. When it comes to video decoding, Intel hasn’t announced any changes and we still have full accelerated decoding of the MPEG2, VC1 and AVC formats.
For encoding, the QuickSync unit is back with full hardware encoding of the AVC format as well as partial MPEG 2 and VC1 encoding. Although MVC (3D version of AVC standard) seems to have featured at IDF, we haven't been able to find any trace of it in the documentation supplied by Intel. Either it isn’t supported or it will be implemented in a forthcoming version of MediaSDK. In effect, to benefit from QuickSync accelerated video encoding, applications developers must use an Intel library (MediaSDK). QuickSync has been announced as being faster though nothing has been said on any improvements to quality. We’ll check all this further on in the test.
Three screens, or almost
Note finally one other change to the display interface (FDI). Intel is now giving us two independent links based on DisplayPort and integrated into the processor itself. They can be partitioned into three independent outs. In theory, then, Ivy Bridge can support three screens. To go with this option however, compatibility with Sandy Bridge processors has to be sacrificed and as we saw in our Z77 motherboard review, none of the models on the market offer the simultaneous three screen option. It should however be available on mobile platforms on condition that two screens use a DP connection (which can be the screen that comes with the machine if it's connected via the eDP standard).
Ivy Bridge platform and range
Ivy Bridge platform and range
The spring launch only concerns the quad-core Ivy Bridge versions with the dual core models coming out this summer. No fewer than six processors for laptops have been launched with nine desktop models. They came on sale at the end of April. Here's a breakdown of the specs of the releases:
The maximum TDP of the processors has dropped from 95 Watts on Sandy Bridge to 77 Watts on Ivy Bridge, which was expected with the move to 22nm. In passing, it should be noted that these processors come in boxes with 95W written on them. This isn't the TDP of the processor but rather the platform required to support it - only 65W or 95W platforms exist.
However the low energy consumption S versions remain at 65 Watts and the Ts at 45 Watts with an i7 included exclusively in this range. And just to make things more complicated, the Core i7-3770K and 3770 don't have the same base clock, unlike the 2600K and 2600.
Comparing the main specifications of the Sandy Bridge and Ivy Bridge Core i7s and i5s, the i7-3770K is priced at the same level as the i7-2700K and the i5-3570K at the same level of the i5-2550K. Unfortunately there aren’t any equivalents to the more affordable i5-2500K and i7-2600K in the new range.
Note that the Turbo on four core with Ivy Bridges is set at up to 200 MHz compared to 100 MHz on Sandy Bridge, which should give the Ivy Bridge models a small additional 2-3% advantage at equivalent base clocks.
The good news on the platform side is that the Core i5 and i7 Ivy Bridge models will work on any LGA 1155 motherboard that is based on H61, H67, P67 and Z68 type Intel Series 6 chipsets on the condition that the motherboard manufacturer supplies an up to date bios. Only the Q65, Q67 and B75 motherboards aren't compatible. At the time of writing, ASUSTeK and Gigabyte are already offering bios’ for their entire ranges, while updates from ASRock and MSI are gradually filtering through. So as to avoid any nasty surprises, we naturally recommend you to update the bios with a Sandy Bridge processor before installing an Ivy Bridge.
As is its wont, Intel has launched a new chipset lane for the Ivy Bridges, Z77 Express, Z75 Express and H77 Express motherboards on the general consumer ranges. We have already published a Z77 Express motherboard review, the main addition to the Series 6 being native USB 3.0 support.
Intel’s segmentation means that the processor can only be overclocked on the Z75 and Z77, while only the Z77 gives the option of three PCI-Express 3.0 ports. Only H77 and Z77 motherboards support Smart Response (HDD cache via SSD), Rapid Start (rapid secure standby with hibernation file on the SSD) and Smart Connect (regular exits from standby to retrive emails).
Of course, Z77, Z75 and H77 boards also have the advantage of having a bios that automatically works with Ivy Bridge. These boards also offer PCI Express 3.0 compatibility systematically, (variable on older generation boards). Outside of this, there’s no performance advantage – Ivy Bridge CPU and GPU performance will be equally fast whatever the chipset!
HD Graphics 4000 and 2500: energy consumption and 3D
HD Graphics 4000 and 2500Before moving onto the purely CPU tests, we’re going to start with an evaluation of GPU performance. Before carrying out this test, Intel supplied us with two relatively similar processors, the Core i5 3570K and Core i5 3550, which are equipped with the HD 4000 and HD 2500 IGPs respectively. The base clocks of these processors differ by 100 MHz, but in our GPU tests we aren’t limited by the CPUs because of the still relatively low IGP performance.
First we wanted to look at IGP energy consumption levels as given by the excellent application hwinfo. These are the internal CPU readings, which may explain why they‘re so optimistic, particularly at idle! We measured energy consumption in load in Furmark.
Intel says that it has worked hard to optimise the energy consumption of its IGP and this shows. In spite of the additional units, energy consumption has only increased by 3 Watts in load in comparison to the HD 3000 used in the 2700K. Energy consumption levels on the HD 2500 are very modest.
Lets move on to performance!
Next we wanted to measure 3D performance. To do so we used three titles, F1 2011 (in DX9 and DX11 modes), Battlefield 3 (DX10) and Civilization V (DX10). In addition to comparing Ivy Bridge to the previous generation, we also added the AMD A8-3850 APU. Its processor side performance level is lower but as the integrated GPUs give modest performance across the board, we weren’t limited by processor performance. As the Intel drivers refused to authorise the 1280x720 and 1920x1080 modes, we used 1366x768 and 1920x1200 instead.
We used the following platforms:
- Intel Core i7 2700K (HD 3000), Core i5 3550 (HD 2500), Core i5 3570K (HD 4000), Asus P8Z77-V Pro, 8 GB of DDR3 1600
- AMD A8-3850 (HD 6550D), Asrock A75 Pro4, 8 GB of DDR3 1600
For information, we also measured the performance of a Radeon HD 6570 (GDDR5) on the first platform. Will Intel’s efforts on the HD 4000 enable it to overtake the current Llano, a few stone’s throws from Trinity?
We put the graphics settings on this Codemasters game on the Medium preset. We took readings in DX9 and DX11 modes. To recap, the HD 3000 doesn’t support DirectX 11!
[ DirectX 9 ] [ DirectX 11 ]
At low resolution, the HD 4000 and HD 3000 perform at the same level. Once the resolution is higher however, there's a bigger gap, with a gain of 44% for the HD 4000. In spite of everything, the AMD APU is still a long way out front in this test.
With DirectX 11, we gain several frames per second, mainly at low res on the Intel DX11 IGPs. The 3570K then closes the gap a little to the AMD APU, which does however remain in the lead. All the IGPs are still a long way behind the modest HD 6570.
We used the game's built-in benchmark which runs in DirectX 10/11 mode. All the graphics options were set at minimum.
With this title, the HD 4000 struggles to add anything significant to what you get with the HD 3000, with a difference of just 18% at 1920 x 1200. Compared to the A8-3850 the level of performance is really too low! The AMD APU gave double what the HD 4000 scored and once again the HD 2500 is very average.
We used the game’s low profile in DirectX 10.
The first thing to note is that all the IGPs suffer here and none really makes the game playable. The HD 4000 however scores a small victory by equalling the AMD A8-3850 APU.
Overall the performance improvements given by the HD 4000 are welcome but aren't enough to compete with AMD's Llano APU architecture, which is disappointing coming just a few weeks before the announcement of its successor, Trinity. The gains could well be welcomed on mobile configurations however.
HD Graphics: CPU vs IGP, QuickSync
CPU vs IGP
We tried to measure the impact of sharing resources between the graphics core and the x86 cores by measuring combined IGP and CPU performance.
To recap, we measure the score obtained in Cinebench at the same time as measuring the number of frames per second displayed in Tom Clancy’s H.A.W.X.
We measured Cinebench performance, varying the number of threads from 1 to 4 for the Core i5s as well as the 8-thread mode with HyperThreading for the Core i7 2700K:
The first thing to note is that Intel seems to have corrected some of the issues… in its drivers! With this same scenario and Core i7 Sandy Bridge, we saw much more pronounced falls, close to the slideshow (see our original test). Here performance only falls in 8-thread mode and in still respectable proportions. We therefore think that there has been a development in the graphics drivers to mitigate the problem.
While this has had some impact, there continues to be a performance limitation and it’s still there on Ivy Bridge where there’s a performance reduction of around 24% with four threads. This performance difference is a little more marked on Sandy Bridge with HyperThreading where you lose around 37%. Note that here the Core i5 with the HD 2500 is only very slightly affected and doesn't engender the same traffic for the processor's memory resources.
Overall then, Intel has partially corrected the problem, even if it’s still there to a lesser degree and hasn't been resolved on our test Core i5 equipped with the HD 4000.
As Ivy Bridge includes a new version of QuickSync, we measured performance with Cyberlink's Media Espresso with a version optimised for the Ivy Bridge processors supplied to us by Intel. The application allows us to use a choice of hardware decoding (from the source video), hardware encoding (from the destination video) or a combination of the two. This last mode is of course the one to go for.
Before our presentation of the results we should say that the three modes are not identical in the application, something we covered previously in this test.
Ivy Bridge does indeed give a significant performance gain on Sandy Bridge with encoding time, already very fast, down by 27%. If we look at encoding performance alone, the performance levels offered by the different solutions are basically identical. In effect, MediaSDK is so fast here that it’s the CPU decoding (not multithreaded) that can become a limiting factor. When we turn hardware decoding on however, strangely it’s the Core i7 2700k that takes the lead.
Of course, the massive gain in speed comes from the combination of hardware decoding and encoding units and their capacity to work hand in hand. There’s still the question of whether the encoding quality is the same with all three Intel IGPs.
The three files produced are different and the results are not the identical!
We compared the encodings with MediaEspresso on our three platforms. Let’s start with our rapid scene taken from the film Inception. A compatible HTML5 browser is required to use our image comparison tool.
Click here to display the tool in a new tab
There were a few little noteworthy surprises. Let’s start with the less obvious of them. Visually the 3550 and 3570 encodings seem similar. There are however some slight nuances, notably in terms of the colours: the HD 3550 encoding is slightly lighter than the HD 3570K version. This makes us think that Intel applies different filters on the image before encoding with its MediaSDK. It couldn’t however be said that one of the two is more faithful to the original colours than the other. Next, the 2700k image comparison shows up the same types of issues, namely loads of artefacts in the form of coloured squares. The encoding algorithm seems to be the same but it looks as if the change is during the decoding phase. Given the quality of results on this image however, we can’t decently declare any winner.
Click here to display our image comparison tool in a new tab
On our second test image, taken from a scene with very little movement, the differences are more marked. Visually you can see that in none of the three cases does the encoded image have the same dynamic as the source image, which is brighter. This apart, the Ivy Bridge version has a clear advantage here, preserving a bit more detail.
Overall the MediaSDK video encoding for Ivy Bridge seems, at least on these scenes, to be equivalent or a bit better than on the one for Sandy Bridge. The overall quality however remains very average and the faults previously noted in our report on video compression haven’t really been ironed out.
Core i5-3570K and 3770K, DZ77GA-70K and protocol
The Core i5-3570K and 3770KFor the purely processor tests we used a Core i5-3570K and a Core i7-3770K.
An Ivy Bridge on the left and a Sandy Bridge on the right
These LGA 1155 processors look pretty similar to their Sandy Bridge equivalents apart from the resistors at the back of the packaging. In load our i5-3570K has a VID of 1.16V, against 1.19V for the i7-3770K. The real voltage reported on the CPU-Z screenshot is lower because of the standard Vdrop phenomenon.
[ Core i5-3570K ] [ Core i7-3770K ]
The DZ77GA-70KWe used a Z77 Intel motherboard, the DZ77GA-70K. We generally use Intel motherboards as they often have a more mature bios on launch and are exempt from features designed to ‘cheat’ on performance benches, as is sometimes the case on third party boards.
We didn’t do all that well here in terms of the maturity of the bios as, although the new mouse-operated UEFI interface is nice to have, it’s still rather problematic for in-depth use. When it comes to the Turbo for example it doesn’t correctly recognise processors other than the 3570K or 3770K (the Turbo has to be set manually). Also high memory ratios such as DDR3-2133 don't work on Sandy Bridge and the Turbo’s thermal envelope can't be modified in the bios (you have to use the Intel XTU software).
Beyond these slight software inconveniencies which we hope will be rapidly resolved, the motherboard did pretty well, with the exception of its rather long boot time. This is a shame as from the hardware point of view the motherboard is very well equipped with:
- A Gigabit port (Z77 + PHY Intel WG82576V)
- A Gigabit port(Intel WG82574L)
- 8 USB 3 ports (4 internal and 4 at the back via Z77 + 2 hub Genesys Logic GL3520)
- 4 SATA 6G (Intel Z77 + Marvell 88SE9172)
- 1 eSATA 6G (Marvell 88SE9172)
- FireWire (TI TSB43AB22A)
- Codec HD Audio (Realtek ALC898)
- WiFi 802.11n module / Bluetooth 2.1 USB
- Buttons to switch it on / reboot
- An LED code display for the boot phase
Its power stage is made up of 8 phases for the CPU cores (VCC) and there are some LEDs on the PCB to display the number of active phases. In terms of extension slots there are of course four DDR3 DIMMs capable of holding up to 32 GB of DDR3, two PCI-Express x16 ports linked to the CPU (Gen2 on Sandy Bridge and Gen3 on Ivy Bridge) and running at x16/x0 or x8/x8, two PCI-E x1 Gen2 ports, two PCI ports and one PCI-E x4 Gen2 port. Intel has used a PLX PEX8606 switch to allow the use of all the Gen2 ports and additional chips with the eight Z77 Express lanes.
We used the test protocol elaborated for the AMD FX review. The changes are detailed on this page. Unless otherwise stated, the tests were carried out on the following platforms.
- ASUSTeK P5QC (LGA775)
- Intel DP55KG (LGA1156)
- Intel DP67BG (LGA1155 Sandy Bridge)
- Intel DZ77GA-70K (LGA1155 Ivy Bridge)
- ASUS M5A99X EVO (AM3+)
- 2x4 GB DDR3-1066 7-7-7 (Q6600)
- 2x4 GB DDR3-1333 7-7-7 (Q9650)
- 2x4 GB DDR3-1600 9-9-9
- GeForce GTX 580 + GeForce 280.26
- SSD Intel X25-M 160 GB + SSD Intel 320 120 GB
- Corsair AX650 Gold power supply
We did of course check to see if performances on the DP67BG and DZ77GA-70K motherboards were the same with the same processsor and memory settings. They were.
Energy consumption and efficiency
Energy consumption and efficiency For the energy consumption analysis we tried to use a test which is more or less representative for all architectures of what we get in applications in terms of performance and energy consumption. In the end we opted for Fritz Chess Benchmark once again. In addition this application has the advantage of allowing us to fix the number of threads to be used.
The energy consumption readings therefore shouldn't be taken as absolute maximum values but rather as typical of a heavy load - applications specialised in processor stress such as Prime95 can consume up to 20% more. All energy economy features, including those on motherboards such as the ASUS EPU, were turned on for this test, as long as they didn't have a negative impact on performance.
Remember we give two types of readings, the first at the 220V wall socket using a wattmeter for the whole test configuration and the second at the ATX12V via a clip-on ammeter. This reading allows us more or less to isolate the energy consumption of the processor but it isn’t unfortunately exactly comparable from one platform to another as in some cases a small part of the CPU energy consumption comes from the standard ATX 24 pin socket.
Note that in comparison to the Intel DP67BG motherboard used previously for LGA 1155, the Intel DZ77GA-70K is very uneconomical in its current state, at idle first of all with an energy consumption increase of 15 Watts at the socket and 4.8 Watts at the ATX12V, which seems to be because of the numerous additional chips and the poor efficiency of its power stage at idle. Note, when at idle, going by the LEDs on the motherboard, one phase remains on and the others aren’t completely off either as their LEDs flicker. However optimal yield at idle is supposed to be obtained with a single active phase.
In load there’s also quite a difference, notably due to a lower Vdrop impact, with the Core i7-2600K 17 Watts higher at the socket and 9.6 Watts higher at the ATX12V and the i5-2500K 16 and 7.2 Watts higher. We have therefore given the values obtained on each of the platforms, the other Sandy Bridges only being measured on P67 and the Ivy Bridges only on Z77, which allows us to compare the Ivy Bridges between each other .
[ 220V socket ] [ ATX12V ]
The energy consumption taken at idle is similar between the Core i5-2500K and i7-2600K models and the i5-3570K and i7-3770K, with however something rather strange happening linked to the DZ77GA-70K motherboard. If we look at the numbers obtained on this board, we can see that there’s a smaller gain with a single thread: with the i5s we’re down from 25.2 to 24W at the ATX12V and from 28.8 to 25.2W with the i7s.
When the processor is occupied at 100%, the gap grows with the i5s dropping from 60 Watts on Sandy Bridge to 51.6 Watts on Ivy Bridge and the i7s dropping from 73.2 to 64.8 Watts. This represents respective energy consumption reductions of 14% and 11.5%.
We therefore chose to use two different methods to isolate the processor energy consumption:
- Energy consumption at the ATX12V
- 90% of the difference in energy consumption between load and idle at the socket
We took this at 90% so as to exclude power supply yield. Note that while the first reading favours processors that draw a small proportion of power from the standard ATX socket, the second favours those with high energy consumption at idle. Unfortunately no method is perfect.
[ 220V socket ] [ ATX12V ]
With the DZ77GA-70K the energy efficiency of processors on the ATX12V has gone down, which was to be expected given the fact that the energy consumption is higher than with the DP67BG. At maximum load and with an identical motherboard, Ivy Bridge gives a gain of about 20% (20.8% and 19.2% on i5 and i7 at 90% of the difference at the socket, 23.7% and 19.9% at the ATX12V), but it should be noted that Fritz doesn't really benefit from the Ivy Bridge IPC improvements as we'll see in the tests at equal clocks.
TemperatureBefore moving on to overclocking we wanted to take a look at the temperature levels on the Ivy Bridges. In effect, while energy consumption is down, the temperature readings for the cores on the various monitoring tools are significantly up. To recap, Intel processors come with a Digital Thermal Sensor (DTS) for each core which reports to a register the temperature difference with the maximum temperature supported, expressed as TJmax. These sensors are less reliable as temperatures stray from the TJmax and even if they have been improved over successive generations, they’re still at a maximum error margin of +/- 5°C with an objective of +/- 2°C, which explains in part the sometimes significant differences in readings between the various cores. We’d like to thank Martin at HWiNFO for his help on this subject.
On Ivy Bridge Desktop the maximum temperature (Tjmax) has been increased to 105 °C by Intel, against 98°C on Sandy Bridge Desktop. Up until now, such maximum termperatures were above all found on mobile or server chips. Once this temperature is reached, mechanisms intervene to reduce the clock in a process known as throttling, so as to protect the processor. In practice we measured these mechanisms as kicking in at around 95°C on Sandy Bridge and 102°C on Ivy Bridge after switching off the fans on overclocked processors. These temperatures are of course higher than many users allow themselves to go out of fear of damaging their CPUs in spite of Intel's official specs. Processors shut themselves down completely at 120-130°C to avoid permanent damage.
[ Before Throttling ] [ After Throttling ]
Here are the readings taken on different processors in load in Prime95 with a room temperature of 25°C in load in Prime95 with a Noctua NH-U12P SE2 (tests outside casing). We give the average of four readings which allows us to reduce the impact of any margin of error:
- Core i5-2500K: 48°C (delta T of 23°C, 50°C margin vs Tjmax)
- Core i7-2600K: 51°C (delta T of 26°C, 47°C margin vs Tjmax)
- Core i7-2700K: 53°C (delta T of 28°C, 45°C margin vs Tjmax)
- Core i5-3570K: 56°C (delta T of 31°C, 49°C margin vs Tjmax)
- Core i7-3770K: 59°C (delta T of 34°C, 46°C margin vs Tjmax)
In spite of the reduction in energy consumption, temperatures are significantly up, with an 8°C increase on the i5 and 6 to 8°C on the i7. The main culprit would seem to be obvious, the increase in the number of Watts to be dissipated per mm².
In load in Fritz, the readings we took for the cores alone in HW Monitor were 61.2 Watts on the i7-2600K and 49.8 Watts on the i7-3770K but because of the new 22nm engraving the 3770K Ivy Bridge only takes up around 45mm² as against around 70mm² for the Sandy Bridge, which represents a reduction in area of 36%. This is a bigger difference than from die to die, which is 26% (160mm² vs 216mm²) because of the increase in IGP size. All this gives us the following:
- Sandy Bridge: 0.87 Watts per mm²
- Ivy Bridge: 1.11 Watts per mm²
Of course the cooling is carried out on larger areas: the die is bigger than the cores alone and is in contact with the CPU IHS (metal shell), which itself is in contact with the processor radiator, but with losses in thermal conduction each time. There also seems to have been a regression here as Pt1t, which has taken an Ivy Bridge apart, notes that the contact between the die and the IHS is made by thermal paste instead of a metallic joint in indium, which offers better thermal conduction. The negative impact of this change seems marginal in practice as testing without the IHS didn’t show any notable temperature gain.
Nevertheless, in taking the maximum temperature supported on the Ivy Bridge desktop models up to 105°C, Intel has compensated for the increase in temperature and in the end we’re left with an almost identical margin for Sandy Bridge and Ivy Bridge with the same cooler, which means the cooling requirements for Ivy Bridge haven’t had to be improved. Those who like to keep their CPUs cool will however have to shell out for a more effective cooler.
Overclocking and undervolting
Overclocking and undervoltingIvy Bridge has the same overclocking mechanism as Sandy Bridge. Overclocking by the bus is very limited (around 5 to 7%) and only the K processors have a free multiplier, which is what is required for a good increase in clock.
On Ivy Bridge the maximum multiplier is up to x63 from 59 on Sandy Bridge, which gives a maximum clock of 6.3 GHz without overclocking the bus, which will only interest those who are adept at overclocking via LN2. The IGP also has a higher overclocking margin, which is now x60 instead of x57, giving a maximum of 3 GHz (steps of 50 MHz).
As with Sandy Bridge, it’s the Turbo multipliers that are modified for overclocking and the thermal envelope has to be increased at the same time to avoid seeing clocks drop back down in load to the base clock. These multipliers can now be modified in real time in Windows.
The memory has also received its lot of innovations. In addition to official support for DDR3-1600, Ivy Bridge now allows you to go up to DDR3-2667, as against DDR3-2133 on Sandy Bridge. In practice, ASUS for example has taken things further and offers up to a ratio of DDR3-3200.
We now have granularity in steps of 200 or 267 MHz, but strangely on its own DZ77GA-70K motherboard only the following ratio settings were accessible: DDR3-1067, 1333, 1600, 1867, 2133, 2400, 2667. ASUS however does make the 200 MHz steps available.
On the Intel motherboard we overclocked four processors cooled by a Noctua NH-U12P SE2 with a room temperature of 25°C (tests outside casing):
- Core i7-2600K (old, January 2011)
- Core i7-2700K (recent)
- Core i5-3570K
- Core i7-3770K
We started with our Sandy Bridges, on which we set the Vdrop on Low instead of High by default in the manual settings (except for the setting allowing us to obtain 0.96V). We limited ourselves to 1.42V on the sensor, in steps of 0.05V – for some time Intel hasn‘t really been giving information on the maximum voltage supported on its processors. At 32nm, you’re advised not to exceed 1.4V and at 22nm you'd expect the threshold to be even lower (1.3V?).
The Core i7-2600K is fairly reluctant when it comes to major clock increases, as you'd expect. For 3.5 GHz we used a voltage of 1.06V, then 1.21V for 4 GHz and 1.35V for 4.5 and 4.6 GHz. Although we went up to 1.42V, we weren’t able to stabilise it at 4.7 GHz. At 4.5 GHz the temperature of the cores was 67°C, compared to 51°C by default, an increase of 16°C.
The Core i7-2700K was more supple in spite of a similar undervolting with 1.06V at 3.5 GHz. Only 1.16V was required for 4 GHz, 1.31V for 4.5 / 4.6 GHz, 1.36V for 4.7 / 4.8 GHz and 1.42V for 4.9 GHz. At 4.5 GHz the temperature was 66°C against 51°C by default, which is an increase of 15°C.
The first Ivy Bridge tested, the i5-3570K supports undervolting with 0.96V at 3.5 GHz. We weren’t able to go any lower on the DZ77GA-70K motherboard. We managed 4 GHz at just 1.01V but we had to go up to 1.26V to clock 4.5 GHz and 1.36V to clock 4.6 GHz. At this clock, the temperature was 72°C against 56°C by default, which is an increase of 16°C.
Like the i5, the i7-3770K clocks 3.5 GHz at 0.96V. To clock up to 4 GHz however we had to increase the voltage to 1.06V and, as with the i5, it clocked 4.5 GHz at 1.26V. To get another 100 MHz we had to increase voltage by 0.05V and then another 0.05V to clock 4.7 GHz. At 4.5 GHz the temperature of the cores was 76°C, compared to 59°C by default, which is an increase of 17°C.
In terms of clocks, these first Ivy Bridges don’t seem to be all that powerful. Of course, overclocking isn’t an exact science and it’s difficult to generalise on such a small sample. Nevertheless our results seem to be in accord with the first tests from other people in possession of Ivy Bridge processors: while you can get close to and even reach 5 GHz with a recent Sandy Bridge, you have to aim at something more like 4.5 GHz with Ivy Bridge.
In comparison to energy consumption at the ATX12V at the base clock, we managed to stablise the processor at clocks of 3.5 GHz, 4.0 and 4.5 GHz with the following increases or decreases in energy consumption:
- i7-2600K: -26% / +9% / +64%
- i7-2700K: -26% / +0% / +50%
- i5-3570K: -29% / -14% / +61%
- i7-3770K: -30% / -9% / +61%
The current Ivy Bridges seem at ease with a clock of around 4 GHz as this overclocking can be obtained at lower energy consumption than the initial level. At 4.5 GHz, we are however too close to their maximum and the increase in voltage leads to a significant increase in energy consumption.
Performance at equal clocks, DDR3-2133, PCI-Express 3.0
Performance at equal clocksBefore moving on to the tests of the models on the market we naturally wanted to check performance at equal clocks so as to measure the IPC gains announced in practice. To do so we overclocked a Core i7-2600K and a Core i7-3770K to 4.5 GHz on a DZ77GA-70K motherboard. They come with 2x4 GB of DDR3-1600 9-9-9.
There’s an average gain of 3.7% at equal clocks. Some applications are significantly faster, with for example a 7.7% difference in Adobe Lightroom, 6.8% in x264 and 6.1% in Bibble. In contrast, there’s only a 0.6% gain in Fritz Chess, 1.3% in Visual Studio 2010 and 1.7% in WinRAR.
In games the average gain is 2.8%, but Crysis 2 is apparently limited by the GPU. If we discount Crysis 2, the average goes up to 3.2% with just 1.2% better in F1 2011 but 5.9% in Starcraft II. As you’ll have understood, as a ‘Tick’ at equal clocks Ivy Bridge doesn’t produce anything dramatic in terms of IPC. Sandy Bridge was itself already very good in terms of efficiency!
What about DDR3-2133?For this test G.Skill supplied us with a 16 GB kit of DDR3-2400 RipjawZ certified at 11-11-11 at 1.65v.
Unfortunately, this mode wouldn’t run on the DZ77GA-70K motherboard, but we did run our test suite on Ivy Bridge, still at 4.5 GHz with DDR3 2133 10-10-10 so as to show you the gains this gives compared to standard DDR3-1600 9-9-9.
We started with a memory latency reading in AIDA64 and the multithreaded bandwidth in the RightMark Multi-Threaded Memory Test:
- Latency: 33.7ns instead of 39.3ns
- Reads: 30.67 GB/s in place of 23.52 GB/s
- Writes: 29.14 GB/s in place of 22.37 GB/s
What does this translate to in practice?
The average gain with applications was 3.7% … the same as that provided by Ivy Bridge over Sandy Bridge! Three applications pulled this figure upwards significantly: 7-zip with +13.8%, Lightroom with +11.4% and WinRAR with a gain of 7.1%. In eight other cases the gain was under 2%.
The average gain in games was 5.7% and there was even a gain in Crysis 2, which is rather surprising given that it was limited by the GPU. Starcraft 2 responds best to the memory clock increase with a gain of 7.7%.
With RAM being very affordable at the moment, acquiring a faster kit shouldn’t therefore be completely excluded. A DDR3-2133 kit is around 50% more expensive than a DDR3-1600 one, but for 8 GB the difference is only around €25!
PCI Express 3.0
Our last test on this page concerns PCI Express 3.0 performance. We compared the performance obtained with PCI Express 3.0 x16 and PCI Express 2.0 x16 with a Radeon HD 7950 to take the readings, in non-paged pool mode to push the interface and controller to a maximum.
A good surprise as theoretical performance was doubled going from 2.0 to 3.0. There’s a particularly marked gain from the CPU to the GPU where we were at 9.2 GB/s on Sandby Bridge-E (see our test here). Performances between Sandy Bridge and Ivy Bridge were the same in 2.0 x16 mode. Note however that such bandwidth isn't really exploited as yet, as noted in this article.
3D rendering: Mental Ray and V-Ray
3d Studio Max 2011 - Mental Ray
We now move on to the practical tests, firstly with a 3D rendering in 3d Studio Max 2011 using the Mental Ray rendering engine on an Evermotion scene. We rendered the scence at 600*375 so as to keep test times reasonably low.
In this first test, the gain between the i5-2500K and i5-2570K was 7.9% and between the i7-2600K and i7-3770K it was 7.7%. This comes mainly from the clock increase which makes a difference of 5.9 and 5.7% with the Turbo. As expected the six-core LGA 1366 and LGA 2011 dominate things here, while the AMD FX-8150 is behind the quad-cores with hyperthreading.
3d Studio Max 2011 - V-Ray 2.0
Still in 3d Studio Max 2011, we changed the engine for the more popular third party engine, V-Ray 2. We used another version of the same scene prepared by Evermotion for this engine, still with a 600*375 rendering. Rendering times are a good deal faster but of course we’re not carrying out a comparison of the engines themselves or the quality of the final files.
Using the V-Ray rendering engine gives similar results with a gain of 7.7% on the i7 and 7.6% on the i5. The Core i7 with six cores is a long way ahead of the field in this largely multithreaded test while the FX-8150 falls between the new LGA 1155 i5 and i7.
Compilation: Visual Studio and MinGW/GCC
Visual Studio 2010 SP1
We compiled the source code of the 3D Ogre engine in Visual Studio 2010 SP1.
This time there was a significant Ivy Bridge i5 gain of 7.2% with 6% on the i7. Once again the six Intel cores dominate while the FX-8150 comes mid-way between the new Ivy Bridges.
MinGW / GCC 4.5.2
The same source code was then compiled in MinGW / GCC 4.5.2.
The improvements introduced on Ivy Bridge have quite a positive impact in MinGW, with compilation 4.5% faster at the same clock. The gain in moving from Sandy Bridge to Ivy Bridge is thus higher than 10% (10% on the i5s and 11.2% on the i7s). The LGA 2011 six-cores have their lead significantly cut here, while the i5-3570K is closing on the AMD FX-8150.
Compression: 7-zip and WinRAR
7-zip has been added to our test protocol. In contrast to WinRAR, this application is highly multithreaded if its highest performance algorithm, LZMA2, is used. We measured the time required to compress a large volume of files.
The switch to Ivy Bridge brings less of a gain in 7-zip with an i5 gain of 6.5% and an improvement of 5.8% on the i7s. Highly multithreaded, the six-core Core i7s continue to dominate, while the AMD FX-8150 loses ground on the new LGA 1155 i7 range.
The same files were compressed in WinRAR using the ‘Best' RAR algorithm.
WinRAR doesn’t really use more than two cores for compression and the gains were quite modest: 6.1% on the i5s and 5.3% on the i7s. While the LGA 2011 processors are once again out front, they're only very slightly so and only on the processor with 15 MB of L3 cache. Although the AMD FX-8150 does well in intensively multithreaded tasks, here it loses quite a bit of ground.
Encoding: x264 and MainConcept H.264
StaxRip - x264 build 2085
For video encoding we retained the popular x264, here in build 2085. We used the StaxRip interface to transcode a 1080p file taken from the Avatar Blu-ray using two passes in fast mode with a bitrate of 10 Mbits /s. We’ve posted the times for both passes, the first being less multithreaded than the second and only really exploiting three or four cores.
[ Total ] [ 1st pass ] [ 2nd pass ]
x264 benefits hugely from Ivy Bridge, as you can see in the tests at equal clocks with a 6.8% gain. This gives us a performance increase of 13.7% on the i5s and 12% on the i7s. The i7-3770K is thus very close to the six-core Intels while the i5-3570K closes the gap on the i7-2600K and is faster than the FX-8150.
Looking at the time taken for each of the two passes, we can see that the Ivy Bridges are incontestably fastest for the first pass, which is the least multithreaded. The six-core processors are however faster on the second pass and the i5-3570K is this time behind the FX-8150.
MainConcept Reference 2.2 H264 Pro
We now move on to another H.264 codec, the MainConcept one. We used the MainConcept Reference H.264 interface to carry out the same type of transcoding as in x264. Note that the first pass is more multithreaded here and we have only given the overall score.
There were also big gains with this other H.264 codec, with an i5 improvement of 12.4% and an i7 gain of 10.4% when comparing the i5-3570K and 2500K and the i7-3770K and 2600K. The six-core Core i7s are still at the head of the pack, while the FX-8150 is only a short head in front of the i5-3570K.
Photo processing: Lightroom and Bibble
Adobe Lightroom 3.4
We have now introduced photo processing by lot to our protocol. We started by exporting a lot of 96 RAW photos from a 5D Mark II as JPEGs in Lightroom, applying various effects such as colour and lens correction or noise processing.
Lightroom enjoys a big performance gain with Ivy Bridge (11.6% with the Core i5s and 10.8% with the Core i7s). The Core i7-3770K thus outperforms the Core i7-990X (LGA 1366 six-core) but the LGA 2011 six-cores are still untouchable. The i5-3570K extends its lead over the FX-8150.
In Bibble we processed a lot of 48 RAW photos. Note that Bibble is slower than Lightroom but, as with the rendering engines, we didn’t carry out this test to compare the applications with each other - this would imply comparing the quality of results as a slower export may also be of higher quality.
In Bibble the gains are even more significant with an 11.7% improvement from the Core i5-2500K to the Core i5-3570K and 12.3% from the Core i7-2600K to the Core i7-3770K. Here however, the LGA 1366 six-cores are out of reach and the FX-8150 is once again down on the i5-3570K.
Chess AI: Houdini and Fritz
Houdini 2.0 Pro
We finished up our survey of applications with quite a particular choice, namely artificial intelligence algorithms designed for chess. We started with Houdini Pro 2, via the Arena 3 interface. Version 1.5 dominated the top of the chess engine classifications and Version 2 seems destined to do the same. We left the engine running until the 24th move at the beginning of a game and noted the speed in kilo nodes per second.
With just a slight gain at equal clocks there’s just a modest 7% improvement for the i5s in Houdini and 8.4% for the i7s. The FX-8150 was down on the i5-3570K once again.
Fritz Chess Benchmark 4.3
We then moved on to Fritz Chess Benchmarking from Chess Base. Once again the results are given in kilo nodes per second.
In Fritz the new Intel range gives even less of an advantage with an i5 gain of 6.4% and 6.3% with the i7s: at equal clocks we noted a gain of just 0.6%. Good news for the FX-8150 here, which just has the edge over the i5-3570K.
3D gaming: Crysis 2 and Arma II: OA
Crysis 2 v1.9
The 3D gaming part of this comparative begins with Crysis 2. We used the latest version 1.9 in DirectX 11 and measured the framerate obtained at 1920*1080 Ultra at a precise point in the game during a shoot-out.
As may be the case with respect to the graphics card, the graphics options, the game engine and the scene, the framerate is seriously limited here by the graphics card. There’s no significant gain with the new range over the previous one, with just 0.4% with the i5s and 0.6% with the i7s. The Intels fully dominate the AMD offer here.
Arma II: Operation Arrowhead v1.59
In Arma II: Operation Arrowhead we measured the framerate when crossing a village in the first solo mission, still at 1920*1080 and with all options pushed to a maximum, including visibility.
This time performances aren’t limited by the GPU and the Core i5-3570K is 7% faster than the i5-2500K, and the i7-3770K 6.3% faster than the i7-2600K. As with all other games, there’s no advantage in Arma II to using six-core processors and this allows Ivy Bridge to take charge, widening the gap with the AMD range.
3D gaming: Rise of Flight and F1 2011
Rise Of Flight v1.021b
We used Rise Of Flight, a First World War fighter plane simulator, at 1920*1080 at high graphics settings. In this test we launched a customised mission with a 32 vs 32 dogfight, with the framerate measured with the back-facing view of our 31 acolytes.
In Rise Of Flight performances were 8.6% up on the i7s and 6.2% up on the i5. The i7-3770K leads the i7-3960X by a short head showing that the presence of six cores isn’t really an advantage while AMD processors haven’t come to the party at all on this title.
We ran the brand new F1 2011 at 1920*1080 with settings pushed to a maximum. We measured the framerate at the start of the Monaco GP.
The new Ivy Bridge range gives a 7% gain in performance in F1 2011, but is a short head behind the i7-3960X, which leads the field. Once again, the AMD processors didn't do well. Note that in this game Hyperhtreading has a slightly negative impact on performance, which explains the almost inexistent difference in performance between the i5s and i7s in spite of the difference in clocks.
3D gaming: Total War Shogun 2, Starcraft II and Anno 1404
Total War: Shogun 2
For Total War: Shogun 2 we used the huge battle of the 'DX9 CPU' test modified for DX11 at 1920*1080 and with high graphics settings.
Once again moving from the i5-2500K to the i5-3570K gives you an extra 6.5% and from the i7-2600K to the i7-3770K an extra 7.6%. These new processors are placed in the top two and once again the AMD processors are down on the Intels.
Starcraft II v1.3.6
For Starcraft II a replay was generously donated by some of the French forum users (thanks!). This replay contains a very (very) full-on attack and we measured its framerate at a resolution of 1920*1080 with all graphics settings pushed to a max.
Of all the games, Starcraft 2 benefits most from Ivy Bridge, with a 5.9% performance gain over Sandy Bridge. Combined with the clock difference, you get a 10.8% improvement on the Ivy Bridge Core i5s and 11.1% on the Core i7s.
Anno 1404 v1.3
Lastly, in Anno 1404 we loaded a saved game with a city of 46,600 inhabitants that we partly visualize from a distance. The resolution was 1920*1080 and all graphics settings were pushed to a maximum.
In Anno 1404 the gain is lower with 6% for the i5-3570K over the i5-2500K and 5.5% for the i7-3770K over the i7-2600K.
Performance averagesAlthough individual app results are worth looking at, we also calculated a performance index based on all tests with the same weight for each test. We have included two averages, one that’s applied across all the tests with the exclusion of 3D games and the other specific to 3D games.
[ Standard ] [ By performance ]
With applications, the Core i7-3770K gives an 8.7% improvement over the i7-2600K, while the i5-3570K is 9.1% up on the i5-2500K. This doesn’t make a huge difference in terms of the overall classification, which is still dominated by the Intel six-cores in these mostly multithreaded tests. Note however that the i5-3570K is on average very close to the AMD FX-8150, further undermining the sales potential of this solution. The i7-3770K has a 23% advantage over the i5-3570K, mainly due to the absence of Hyperthreading.
[ Standard ] [ By performance ]
In 3D games, you get less of a gain with Ivy Bridge, with a 6.9% difference between the i7-3770K and i7-2600K and 6.4% between the i5-3570K and the Core i5-2500K but the new kids on the block are top of the class. As games don’t really exploit more than four cores, Hyperthreading doesn't give the i7-3770K any advantage and it only performs 2.6% better than the i7-350K, in spite of an additional 2 MB of cache and a clock that’s 2.9% higher. The AMD processors are relegated to the bottom half of the table here.
ConclusionWhen Intel launched Sandy Bridge in January 2011, it received a very favourable welcome but what was a ‘Tock’, which is to say a new architecture, was in fact also a sort of ‘Tick’, ie a new engraving process. The 32nm engraving launched a year earlier had in fact only been used on two and six-core processors and the combination of the new (ish) engraving process with the new architecture allowed Sandy Bridge to steal quite a march on its LGA 1156 predecessors launched in September 2009, which were in fact simply a general consumer roll-out of the LGA 1366 Core i7s from November 2008!
The situation isn’t the same for the arrival of Ivy Bridge which has the heavy task of outdoing Sandy Bridge with just some small micro-architecture improvements and the move to a new 22nm process with Tri-Gate. Intel has rather put the accent on the graphics part of Ivy Bridge, something we can expect to be the case for the next few generations. The new HD Graphics 4000 and 2500 IGPs do indeed offer significant 3D gains on the previous IGPs, though AMD still has the advantage here and should be able to extend it further with the forthcoming arrival of Trinity.
In spite of these gains, the level of performance is still insufficient for comfortable 3D gaming usage, unless you’re using these IGPs on laptops with screens with 1366*768 pixels that Intel itself hopes to see disappear in the medium term. On desktop PCs, 1920x1080 is now the norm and no HD Graphics solution allows you to use such a resolution correctly in 3D, which makes paying extra for a 4000 version rather pointless as HD Graphics 2500 will suffice for other usages.
In terms of the actual CPU part of the processor, there is of course less of a difference with Sandy Bridge. The gains at equal clocks are 3.4% on average with peaks at 7% in the most favourable cases such as H.264 encoding and photo processing, which is pretty good for what is only a ‘Tick’, especially as Sandy Bridge was already performing very well here.
The most problematic thing seems to be the lack of maturity of 22nm Tri-Gate. In terms of energy consumption first of all; while there is a notable reduction, it isn't as significant as expected and this results in higher temperatures of the cores because of the increased power dissipated per mm². We did however note that our test processors performed promisingly with respect to undervolting and that with Intel increasing the maximum temperature threshold to 105°C, no additional cooling is required.
The other issue is that there isn’t as much of a margin for overclocking as with the latest Sandy Bridge models. While a recent Sandy Bridge can be clocked close to if not up to 5 GHz, we had to settle for 4.6 or 4.7 GHz on our two samples of the new Ivy Bridge models and this seems to be representative. This completely annuls the IPC gain for anyone looking to push a CPU to its limits, especially given the slight pricing increase! We do however expect improvements here in the months to come, as is often the case.
The Sandy Bridge Core i5s and i7s are indeed a hard act to follow, especially when the accent is put on the IGP. No doubt a little too much was expected of the Ivy Bridge Core i5s and i7s. Overall the verdict is positive nevertheless as Intel is giving better performance at the same price with these new models, whether this be via the IPC gain or the Turbo, the more powerful IGP or the power efficiencies. It is however a shame that Intel hasn’t been more aggressive when it comes to pricing, particularly with the K processors. It's worth noting that the Sandy Bridges haven’t been reduced in price since their launch over a year ago! This is a first and due to the lack of competitiveness in the AMD offer.
At the end of the day then, the Ivy Bridge Core i5s and i7s do represent a better choice than their predecessors and constitute the best of what is currently available, but the difference is rather tiny and this evolution doesn't justify an upgrade in spite of the welcome compatibility of the new range (such as, say, moving up from a Sandy Bridge Core i5 to an Ivy Bridge Core i5).
Copyright © 1997-2015 BeHardware. All rights reserved.