Intel Core i7 and Core i5 LGA1156 - BeHardware
Written by Marc Prieur
Published on September 8, 2009
A new platform
10 months after launching on the LGA1366 platform, Intel has decided to roll out its Nehalem architecture on the new LGA1156 platform, supposed taking it mainstream. What changes does it bring and what do they mean in practice?
A new platform
The first difference comes with the socket, down from 1366 to 1156 contact points as a result of various modifications that we’ll go into further down. Whenever a new socket comes out, it seems a new fan is also called for and it doesn’t seem unreasonable to wonder at the rationale for all this. If you measure the difference between the mounting holes in recent Intel sockets, you find:
- LGA775: 72 mm
- LGA1156: 75 mm
- LGA1366: 80 mm
We have to say, we’re somewhat mystified as the fact that the LGA1156 socket is compatible with neither of the others suits nobody except cooler designers – although even some of these, like Noctua, will be offering free adaptors to their users.
The other innovation, harder to spot, is the integration of the Northbridge in the processor, which means the die goes from 263 to 296mm˛ and the number of transistors from 731 to 774 million. Socket LGA1366 constituted a first step for Intel with the integration of a triple-channel DDR3 memory controller and LGA1156 continues on this route with a double-channel DDR3 memory controller. In practice, the absence of the third channel won’t be missed as it only offered very limited, even inexistant gains because the increased bandwidth wasn’t always big enough to make up for higher latency.
Integration of the PCI-E controller onto the processor itself is a real improvement, especially as it is linked internally to the memory controller by a QPI type bus. It supports 16 PCI-E 2.0 lanes and can supply 1 x16 slot or 2 x8 slots in a multi GPU configuration. On LGA1366 the PCI-Express was supported by the X58 with a total of 36 lanes. When you think that a multi GPU configuration isn’t much affected whether working at 2x8 or 2x16, what some users will regret is the absence of a 3rd or 4th x8 slot.
As with the X58, the P55 allows CrossFire configurations as well as SLI configurations as long as the motherboard is compatible, in other words a card for which the manufacturer has paid NVIDIA for a license…
The complete integration of the northbridge within the processor leads to a greater simplification in the design of motherboards as the P55 Express chipset designed for the LGA1156 is in fact just a “Super ICH10” with a couple of extra PCI-Express lanes to make a total of 8. These still function at 2.5 GT/s (as against 5 GT/s for those in the CPU) or 250 MB/s per lane in each direction.
As the Northbridge is completely integrated on the processor, connection to the chipset via a QPI bus is no longer necessary and this simplifies motherboard design. Instead you’ll find the usual DMI bus linking the MCH and ICH but still limited to 2 GB/s which is a bit low if only because 8x PCI-E total 4 GB/s.
New processors3 new processors have been launched on the LGA1156 platform:
- Core i7-870, 2.93 GHz, $562 (same as the i7-950)
- Core i7-860, 2.80 GHz, $284 (same as the i7-920)
- Core i5-750, 2.66 GHz, $196
All three are quad cores and as in the LGA1366 versions, there is 64 KB of L1 cache, 256 KB of L2 cache for each of the four cores and 8 MB of shared L3, all engraved at 45nm. We didn’t note any difference in performance in terms of these caches with equivalent clocks. DDR3-1333 that already worked with LGA1366 processors is now officially supported.
While the two Core i7s have hyperthreading as with the LGA1366 processors, the Core i5 does not. The gains brought by this technology are more than satisfying, with for example a 23% gain in Cinema 4D, 20% in Anno 1404 and MinGW, 17% in 3ds max 2010, 12% in MainConcept Reference or 11% in After Effects CS4. In Nuendo 4 however there is a negative impact of 6% and we’re down 3% in Crysis and 18% in Arma 2, instances in which you’ll have to disactivate hyperthreading or better still, set the affinity by hand on the physical cores.
Note also the differences in clocks of the uncore part (memory controller / L3 cache) and Turbo mode functioning:
To recap, Turbo mode automatically increases the clock on Intel processors depending on the load, within the limits of TDP. This means if you’re using only one of the cores of an i5-750, it can clock up to 3.2 GHz as long as you stay within the power consumption limit of 95W. You can see that Intel has pushed the Turbo boost even further on these new processors, so much so that the base clock no longer means very much.
In practice in our test protocol we measured an overall gain of 6% linked to Turbo for the Core i5-750 and 4.3% for the Core i7-920, with a high of 11% for the i5-750 in WinRAR and Crysis and 7% for the i7-920 in Crysis.
The uncore clocks are also affected and the difference with LGA1366 CPUs should be noted. On these, the multiplier was not restricted on any of the processors and had to be at least double that of the memory. So with a platform reference clock of 133 MHz, you had to clock the uncore at at least:
- 3.20 GHz for DDR3-1600
- 2.66 GHz for DDR3-1333
- 2.13 GHz for DDR3-1066
On LGA1156 it’s different as the uncore multiplier is fixed at x16 on the Core i5s and x18 on the Core i7s. Independently of this clock, you can anyway go with DDR3-800/1066/333 for the Core i5s and DDR3-800/1066/1333/1600 for the Core i7s.
To be completely thorough, we should note that according to our tests, the LGA1156 processors give slightly lower performance in certain situations than the LGA1366 processors when they are configured in exactly the same way, though we aren’t able to say why. With DDR3-1066 7-7-7 on both, a clock of 2.93 GHz with Turbo disactivated and the uncore at 2.40 GHz, we got a notable difference of 5% in 3ds studio max without HyperThreading and 1.8% with, 5% in both cases in Nuendo and 3% in MinGW and in Mame.
Here for comparison are the performances of a Core 2 Q9450 at 2.66 GHz and a Core i7 750 with Turbo disactivated:
The gains given by Nehalem architecture are there for all to see, even without Turbo and Hyperthreading, averaging 16.5%.
Energy consumption, overclocking, the test
The CPUsFor this test, Intel supplied us with two models:
- Core i7-870, 2.93 GHz, $562
- Core i5-750, 2.66 GHz, $196
From left to right: LGA775, LGA1156 and LGA1366. As you can see, an LGA1156 processor is the same size as an LGA775.
Energy consumptionEnergy consumption was measured in two places: processor consumption alone at the ATX12V socket and with a clip-on ammeter and total consumption with a power meter at the wall socket, always loading the machine with Prime 95. This means that other components such as the graphics card or the hard drive are in idle when these readings are taken.
In the case of the readings at the ATX12V, because of the different ways in which the CPUs are supplied, on a Phenom II the memory controller (not integrated on the Core 2s) which consumes between 10 and 15W in charge is taken into account, while on the Core i7 it isn’t (also the case for the L3), the uncore being supplied via the ATX standard.
The new Core i5/i7s consume more than the LGA1366s in idle, in spite of all the energy economy features being activated: perhaps integration of the PCI-Express controller is the difference here. In load, energy consumption is higher than on the LGA1366 platform, which can be explained by the fact that the Turbo mode goes further. Note, moreover the small difference between 4x and 8x Prime on the Corei7-870.
At the wall socket the LGA1156 platform gets a great score in idle, thanks to the single P55 southbridge on the motherboard. It does better than with the socket 775 platform which is truly remarkable. In load of course it’s another story given the energy demands of the Core i5s/i7s, but the Core i7-870 configuration is for example more economical than the Core i7-920 while the processor on its own consumes more.
OverclockingGiven the fact that they already consume a lot of energy, we limited overclocking to 1.3V in the bios (1.27-1.3V in practice) for each of our processors, which is about 0.1V up on the base voltage.
In both cases we managed a stable 3.8 GHz with a bus clock of 190 MHz for the i5 and 173 MHz for the i7. There is a decent margin for overclocking then but you’ll need to fit yourself out with the right cooling to go any further.
The testFor this test, we took advantage of the availability of the RTM version of Windows 7 to redesign our test protocol. The OS first then: we’re now using a 64-bit version of Windows 7, which means that all software available in 64-bit mode is tested in this mode.
We have taken the opportunity to update the software, which means 3ds max is now tested in version 2010, Min GW and WinRAR (3.8 up to 3.9) have been updated, as have After Effects (CS3 up to CS4) and Nuendo (4.2 up to 4.3). The VirtualDub/DiVX combos and AutoMKV/x264 have been replaced by Avidemux/x264 and MainConcept Reference/H.264, while the test files of virtually all the tests have changed or been modified (higher rendering resolution for example).
In terms of the games, we have decided to retain Crysis 1.2 and its ultra-heavy CPU test but to retire World In Conflict and replace it with more recent and demanding games: Arma 2, Grand Theft Auto IV and Anno 1404 join the protocol. So as to show up processor differences to a maximum, we set all graphics options to a max so as to load right up, at the same time as limiting resolution to 800*600 so as to eliminate any smoothing due to the power of the mono-GPU solution used on the test configuration.
The configurations are as follows:
- ASUSTeK P5QC (LGA775)
- Intel DP55KG (LGA1156)
- Intel DX58SO (LGA1366)
- ASUSTeK M4A79-T (AM3)
- 2x2 Go DDR3-1333 7-7-7
- GeForce GTX 280 + GeForce 190.62
- Raptor 74 GB + Raptor 150 GB
- Creative Audigy
- Windows 7 64 bit
3ds Max 2010 & C4D R11
3D Studio Max 20010
We begin with the famous image rendering software, now in its x64 and 2010 version. The test scene used comes from SPECapc for 3ds max 9 (space_flyby_mentalray) and uses the Mental Ray rendering engine.
Lack of Hyperthreading is costly for the Core i5-750, so much so that it is “only” on between the Core 2 Quad Q9550 and Q9650. The Corei7-870 is noticeably faster but remains behind the i7-950 which nevertheless costs the same.
Cinema 4D R11
The rendering software Maxon is well known in the overclocker community through Cinebench, which allows you to compare processor performance easily. Cinebench however uses version R10 of the Cinema 4D rendering machine, while version R11 doubles performance. We use this latest version in 64 bit mode with the scene from Cinebench R10 rendered at a higher resolution so as to prolong rendering time.
The i5-750 is almost as good as the QX9770 which isn’t bad, while the i7-870 is a little behind the i7-950.
MinGW & WinRAR 3.9
MinGW / GCC
This is an applied test with the compilation of MAME source code using GCC under the MinGW development environment. We are now using version 5.1.4 of MinGW and compile the source code of Mame 0.133.
The Core i5-750 does well compared to the Core 2s – it’s in front of the QX9770 – but remains behind the Phenom II and far behind the Core i7s that put HyperThreading to very good use here. The Core i7-870 is once again just behind the i7-950.
We are now using vesion 3.9 64 bit WinRAR, which introduces new multithreading optimisations for the compression of a group of files.
Here HyperThreading doesn’t really have an impact and the i5-750 gives very good performance in comparison to the other processors. Note that the LGA1366 platform is out front, even at lower clocks as is the case with the i7-920 which does almost as well as the i7-870!
H.264: Avidemux and MainConcept
Avidemux + x264
We have abandonned VirtualDub and DiVX encoding and replaced it with H.264 encoding for video processing. To start with, we use Avidemux version 2.5.1 to compress a 1920x1080 HD video file via the x264 codec at intermediary quality.
HyperThreading is a real bonus here but the Core i5 does pretty well all the same with a score between the QX9770 and the Q9650. The Core i7-870 is just behind the Core i7-950.
MainConcept Reference + H.264/AVC Pro
For this second H.264 encoding we use MainConcept Reference and its H.264/AVC Pro codec on “High”, still with the same video.
The Core i5-750 does better than any of the Core 2s and it should be noted that the Phenom II X4 965 is on a par here. The Core i7s go even further, with the 870 just behind the 950.
After Effects CS4, Nuendo 4.3
After Effects CS4
In our new protocol we use version CS4 of Adobe After Effects. We have gone for a composition using various effects to render 3D animation, with multi-processing activated so as to be able to make the most of the number of available cores.
Performance on the Phenom IIs and Core i7s are a good deal up on CS3. The Core i7s extend their advantage while the Phenom II gives a less catastrophic performance opposite the Core 2s. the Core i5-750 is virtually on an equal footing with the QX9770 and the 870 is, for a change, a little behind the 950.
A new music project using various native plugins as well as 2 HalionOne virtual instruments was exported as a wav file (thanks to Draculax). As Nuendo doesn’t react well to HT (negative imact of about 6%), we forced the affinity by hand.
Here the Core i5-750 is between the QX9770 and the Q9650, while the i7-870 is once again a little less rapid than the i7-950.
Crysis & Arma 2
With patch 1.2, Crysis has a very heavy CPU bench (to be found in the Bin32/Bin64 directory). The test was carried out at very high settings but at 800x600 so as to limit dependence on the graphics card. Affinity was forced on the physical processors, as Crysis is down by around 3% otherwise.
Here we’re faced with a GPU rather than CPU limitation, or at least a GPU/Drivers limitation, which comes into play as of the QX9770 and in spite of the low res. Unfortunately this smooths results at around 30 fps.
New to our test protocol, Arma 2 is configured with all settings at a max including max visibility (10 km), which brings the configurations to their knees. Resolution stays at 800x600 to avoid the graphics card impacting on performance. Note that you must disactivate HT or force the affinity for this game, as you’ll lose 15 to 20% of fps otherwise and it will suffer from rather disagreeable lags.
The first thing is that the Core 2s are generally struggling, significantly down on the Core i5s and i7s and even the Phenom II X4. The Phenom II X4 is not far behind the Core i5-750 and i7-920, with top prize going to the Core i7-975.
GTA IV & Anno 1404
Grand Theft Auto IV
GTA IV is also new to our protocol, chosen for its weight and multi-threading optimisations. Once again all the settings were pushed to a maximum, with the exception of the textures so as not to exceed available video memory, all at a res of 800x600.
Once again the Phenom II does well with a higher score than the QX9770, while the Core i5 and i7s stay in the lead, the i7-870 doing a little better than the i7-920.
Our last test game is Anno 1404, a strategy game tested at max settings but with resolution still at 800x600. We use a saved game with a city of 46,600 inhabitants that we partly visualize from a long distance.
The game is well adapted to Hyperthreading and this gives the Core i7s a big advantage with the i7-870 – rarely enough to be worth noting – doing better than the i7-950. Although not as fast, the Core i5-750 nevertheless does very well as it is a good deal faster than the QX9770.
Average performanceAlthough individual app results are worth looking at, we calculated a performance index based on all tests with the same weight for each test. An index of 100 was given to the Intel Core 2 Q9400.
The first thing to note is that the new test protocol has changed the hierarchy of previously tested processors. Our old protocol put the Phenom II X4 965 on a par with the Q9550. It is now between the A9650 and the QX9770! The Core i7s also do better with the new protocol because the difference between the i7-920 and the QX9770 has more than doubled, going from 7% to 17%.
Overall, the Core i5-750 does better than the Core 2 QX9770 which isn’t too bad given its pricing. The news is not as good for the Core i7-870 that remains behind the Core i7-950, which is priced at the same level. The Core i7-860 ought to be slightly up on the Core i7-920, with both costing the same.
ConclusionSince its launch in November 2008, the Core i7 has made a strong impression both in terms of performance and high pricing, as has the LGA1366 platform. By rolling the architecture out on the LGA1156 platform, Intel is now trying to retain just the positives.
Several aspects make the LGA1156 platform more attractive than LGA1366 was when it came out. Firstly, and this isn’t directly linked to Intel, DDR3 memory is now much cheaper with 4 GB of DDR3-1333 now going at under €100. What’s more, while the LGA1366 platform required quite complex motherboards, the design for LGA1156 cards has been simplified: with no QPI and a single chip chipset, these cards ought to be a good 50 euros or so cheaper.
The processors aren’t as affordable and this may be a little disappointing when you first look at the platform. If, for example, you compare with a Core i7-920, the Core i5-750, although $88 cheaper at $196, is also a long way down in terms of performance. But this would be to ignore the superiority over processors such as the Phenom II X4 965 ($245) and the Intel Core 2 Q9550/Q9650 ($266, $316).
The Core i7-860 ($284) is also fairly well placed as, for the same price, it offers performances that ought to be up on an i7-920. Unfortunately this isn’t the case for the Core i7-870 ($562), which is much too expensive in comparison to the 860 and costs the same as an i7-950 at the same time as giving slightly lower performance.
Where does all this leave the LGA1366 platform? In the short term, in no-man’s land: more expensive and more demanding in terms of energy consumption, the platform only gives better performance because of higher-clocked CPUs that are more expensive and not necessarily easy to overclock. Only those sold on Tri-SLI will find the platform’s use enduring as LGA1156 cannot support a 3rd PCI-Express port at decent speeds. In the longer term, Intel will launch the highly-priced Gulftown, a 32 nm hexa-core, on the platform in 2010. It should be compatible with current motherboards, so if you like the sound of this processor, stay with LGA1366.
Above and beyond the frustrating separation of platforms, LGA1156 looks like an apt successor to socket 775. All that’s missing is decently priced processors and there’s the rub. Intel won’t be offering anything other than the Lynnfield and its 774 million transistors engraved at 45 nm between now and the beginning of 2010, and there’s no reason why it should lower prices. We’ll then have to wait for 2010 and the arrival of the Core i3 (32 nm, 2 cores + hyperthreading) to move over to LGA1156 at an affordable price. In the meantime, no options other than AM3 and LGA775!
Copyright © 1997-2015 BeHardware. All rights reserved.