Intel has been sitting pretty as the producer of the fastest CPUs in the world ever since it launched Core 2 back in mid 2006. No matter what AMD has done to catch up, Intel has always been one step ahead, leaving AMD to compete on value alone. This was no less true when Intel launched its Core i7 900 (Bloomfield) CPUs last November (blimey, that’s nearly a year ago now) and opened up the performance gap even further. However, this time Intel made what could be interpreted as a few mistakes.
For a start, these new CPUs used a new socket so required a new motherboard. In itself this wasn’t too big of a deal as every now and then such changes are required. However, the X58 chipset that these boards were based on was very expensive so combined with a minimum £200 for the CPUs themselves, it made for a very costly upgrade. Add in the fact that the vast majority of people simply didn’t need such powerful hardware and we found little reason to actually recommend investing in this platform, especially as AMD offered a much more affordable and smooth upgrade path.
Of course, this wasn’t really a major problem for Intel as it could rely on its Penryn generation of Core 2 CPUs that were still going strong in the value and mainstream market. However, we knew this platform was having its last hurrah as no more CPUs would be released for it, so again it left space for AMD to easily compete.
Now, though, Intel has finally launched new processors based on the architectural improvements present in Core i7 900 but with a few changes that bring the cost of the chips down to a more sensible level. Perhaps more importantly, these CPUs will be accompanied by a new chipset that is set to cost considerably less than X58, making the overall system cost lower as well. So, if you’ve been waiting to upgrade your ailing Core 2 Duo system and haven’t been tempted away by AMD or couldn’t stomach forking out £350 just for a CPU and motherboard then this looks set to be the platform for you.
These new CPUs, collectively codenamed Lynnfield, will come under the Core i7 800 and Core i5 700 brand names while the chipset is to be known as P55. Specifically, today sees the launch of three new CPUs; Core i7 870, Core i7 860 and Core i5 750 and these will have nominal rated speeds of 2.93GHz, 2.8GHz, and 2.66GHz respectively. Pricing will no doubt fluctuate considerably over the next couple of months but official launch prices are £339, £174, and £122 (converted directly from US dollar prices). Sadly, current prices in the UK are quite a bit higher at £439, £226, and £159. All told, these aren’t exactly bargain basement prices but they’re a damn sight more affordable than Bloomfield chips. The Core i5 750, in particular looks set to be a steal, assuming it overclocks reasonably well.
As for motherboard prices, well they’re not quite as cheap as we hoped they might be. We’ve spoken to a number of manufacturers and it seems like the lowest price models will be around the £100 mark while top end models will push £200. These are still much more competitive prices than X58 boards, which started at £150 and went well passed £300, but if you were hoping for an ultra budget model then you’ll have to wait for prices to drop.
So that’s the broad picture but before we get to testing and drawing any conclusions, let’s take a look at some of the technological changes Intel has made with Lynnfield.
As mentioned previously, these two new CPU ranges are based on the same Nehalem microarchitecture as Core i7 900 (which in turn was very similar to that of the Core microarchitecture) so the actual parts that do the calculations are identical and the fundamentals of the rest of the CPU are the same.
Essentially, through improvements in micro-op handling, the addition of SSE4.2 instructions, and an increase in the number of execution units, Nehalem offered a significant increase in performance at the micro-op level over the Core microarchitecture that came before it. In other words, whether you’re dealing with single or multi-threaded applications, Nehalem can run them more efficiently (i.e. faster) than Core.
Other changes included a move to a three level cache system, improved power management, and the introduction of Turbo Mode and all these features have been carried over from Bloomfield to Lynnfield. Simultaneous Multithreading (two threads can run on each core), QPI, and an integrated memory controller (IMC) were all also introduced with Nehalem but it’s these features that have either had the chop or been modified in the move from Bloomfield to Lynnfield.
Starting with the IMC, Lynnfield CPUs will still have one, and it still interfaces with DDR3 memory, but it has been reduced from triple channel to dual channel. Assuming you’re using 1,333MHz memory this means a reduction in memory bandwidth from 31.2GB/s to 20.8GB/s, which is quite a drastic reduction that brings it below the theoretical level of AMD’s latest systems. Nevertheless, it is still markedly more than could be easily achieved with the fastest Penryn systems, which still compete with AMDs current fastest systems in terms of overall performance, so clearly memory bandwidth isn’t the be all and end all.
The next big change requires us to look at the whole system architecture to explain. With Bloomfield, as well as bringing the memory controller onboard, Intel introduced a new interconnect between the CPU and other parts of the system. Called QuickPath Interconnect (QPI), it talks to the X58 IOH ”northbridge” which in turn interfaces with PCI-Express graphics and the rest of the IO subsystem through the ICH ”southbridge” (hard drives, USB, audio, etc). This link is very fast, providing up to 25.6GB/s of bandwidth, which means it can support, amongst other things, 36 lanes of PCI-Express attached to the X58 IOH. However, this amount of bandwidth is seldom going to be used by most systems so for Lynnfield Intel decided to take a different approach.
Instead of QPI, Lynnfield integrates 16 lanes of PCI-Express graphics communication onto the CPU itself and all other communication is done via a DMI link, which can deliver up to 2GB/s. This may sound like a drastic drop but everything running through the DMI was running through the DMI on Bloomfield so there should be no performance drop here. The only really concern is graphics performance.
While 16 lanes is perfectly adequate for any single graphics card (even the dual-GPU monsters like ATI’s HD 4890 X2 and nVidia’s GTX 295) it is going to limit the potential for running multiple cards in CrossFire or SLI. While these modes will still be supported by means of a 2×8 configuration, there may be situations where multiple high-end graphics cards are bandwidth limited. This is a question we shall return to shortly in another article.
Ah, encoding MP3s. No matter how many advances there have been in audio formats and even with the proliferation of online music services, encoding our audio CDs into MP3s is still a stalwart of a computer’s regular tasks. So, let’s see how these new chips holds up.
”’Singe-track, single-thread encoding”’
It’s the way it always used to be done. One track after another until the album’s finished. For this test we encode 11 tracks using the LAME encoder.
A slight improvement over single-threaded encoding, this method will take advantage of multiple cores for each track it is currently encoding. This test encodes the same 11 tracks and uses the multi-threaded version of LAME.
Finally, we have the quickest method; encoding as many mp3s at once as there are cores. We used DBPowerAmp’s benchmark to load as many cores, or virtual cores in the case of i7, as possible.
”’Photoshop Batch Processing”’
For this test we run all the auto-adjustments available in Photoshop on a large collection of images and time how long it takes to complete. It’s an old test that isn’t multi-threaded but it is incredibly reliable at demonstrating differences in single-threaded performance.
”’WinRAR File Compression”’
Though zipping and unzipping files is a fairly quick and easy task, the frequency with which a heavy computer user has to do it means saving five seconds each time you unzip something is a real boon. We therefore feel it makes for an interesting test of system performance.
This test is split into two parts. The first times how long it takes to compress a large video file and the second records how long it takes to compress the folder of photos used in the above Photoshop test. We then repeat this test using the multithreaded version of WinRAR.
While many people will associate 3D performance with a graphics card, for CAD applications a lot of the grunt work is still performed on CPUs so this is something we’ve put to the test. We use two industry standard tests, called POV-Ray and Cinebench. Both programs can be freely downloaded and the benchmark begins with just a single click so they’re a really easy comparison tool if you fancy seeing how your system shapes up.
Cinebench has both a single-threaded and multi-threaded test while POV-Ray is just multi-threaded. Slightly disappointingly, Cinebench only returns a score, rather than a time to completion, so the numbers in isolation don’t really mean anything. However, when compared they adequately show up any difference in performance.
Perhaps the next most obvious task for a powerful CPU in a desktop PC is video encoding. Whether it’s for editing your family videos or re-encoding your favourite videos to an iPod-friendly format, video encoding is becoming more and more common and so we’ve used two scenarios for testing video encoding performance.
The first is our age old test of re-encoding a portion of DVD quality MPEG-2 video into the open source Xvid format. This test is a bit long in the tooth as the version of Xvid we use is quite old so not multi-threaded (not to mention the fact that the source footage isn’t even HD) but in being so it gives us a very reliable indication of single-core performance.
The next scenario takes a section of 1080p footage, again in MPEG-2 format, and re-encodes it into the h.264 format. This is a much more up to date test as the source material is Full HD and the h.264 codec is about the most advanced video format currently available. We also use the open source converter, AutoMKV, for this task, which fully supports multi-threading.
It’s arguable that for many of you, gaming performance will be of paramount interest when it comes to buying new parts for your PC. However, the simple fact of the matter is, if you have a powerful gaming PC, CPU performance is not going to be your bottleneck. So as long as you have a reasonably nippy dual, triple, or quad-core CPU, you’ll be fine. It’s for this reason we didn’t go to town with our game testing and have looked at just one game, Crysis.
We use scripting to run through a custom time demo, with frame rates recorded along the way. Each setting is run three times and an average taken to ensure a consistent and fair result. We used a ‘low’ graphical setting to demonstrate the effect a fast CPU can have when graphics card performance isn’t a bottleneck and then we used a ‘high’ setting to show the limited effect a fast CPU has when your graphics card is the bottleneck.
Though Intel is launching three CPUs today, it only provided us with two for testing: Core i7 870 and Core i5 750. We will assess the piggy in the middle, the Core i7 860 in the near future.
Our testing consists of mostly ”real world” benchmarks including games, photo editing, mp3 encoding, and video transcoding along with a few canned benchmarks in the form of PCMark Vantage and the 3D rendering tests Cinebench and POVray.
- Intel Core i7 870
- Intel Core i5 750
- Intel Core i7 965
- Intel Core i7 920
- Intel Core 2 QX9770
- Intel Core 2 Q9550
- AMD Phenom II 965 BE
”’Common System Components”’
- AMD ATI Radeon 4870 X2 Graphics Card
- 2 x 1GB Qimonda IMSH1GU03A1F1C-10F PC3-8500 DDR3 RAM
- Western Digital Raptor X 150GB Hard Drive
- Pioneer BDC-S02BK Blu-ray Drive
- Windows Vista Home Premium 32-bit
”’Core i7 870 and Core i5 750 (Lynnfield) Test System”’
- Intel DP55KG Motherboard
”’Core i7 900 (Bloomfield) Test System”’
- Intel DX58SO ”SmackOver” Motherboard
”’Core 2 test Sytem”’
- Asus P5E3 Motherboard
”’AMD Phenom Test System”’
For testing, Intel provided its DP55KG Extreme Series motherboard. It supports two graphics cards, though in accordance with the limit of 16 PCI-Express lanes the second slot is only x8 in length, which is quite an odd sight.
One of the best things about P55 is that it’s a one chip solution, so motherboards based on it have much more space than we’re used to seeing. This is something we saw when Asus gave us a preview of its P7P55 EVO and it’s the same with this Intel board. Indeed, there’s so much free space that Intel has found room to integrate a fancy illuminated skull, which even flashes red when the hard drive is accessed – very funky!
Other obvious things to note are the four memory slots – evidence of the return to a dual-channel memory controller. There’s also a distinct lack of any extraneous cooling with just a small heatsink on the PCH southbridge and a few little ones on the power circuitry round the CPU.
(centre)”’With so few extra chips and little cooling required, P55 boards are set to look quite spartan even when packed with feaures.”’(/centre)
While I’m not going to draw an absolute verdict on this particular board today, it looks like a perfectly serviceable example and, as you’ll see later, it puts in a pretty good performance when overclocking. Furthermore, all this space saving means we are bound to see some great boards in MicroATX or even MiniITX form factors, which is quite an exciting proposition.
As mentioned earlier, one of the more regrettable aspects of these new CPUs and motherboards is that they require a completely new socket. Instead of LGA775 as used by Core 2 or LGA 1366 used by Core i7 900, Core i7 800 and Core i5 700 will use LGA 1156. As you’d imagine, the number refers to the number of contacts on the CPU, and pins on the motherboard. Contrary to what you might expect this doesn’t mean LGA1156 is any bigger than LGA 775, indeed it’s exactly the same, though LGA 1366 is markedly larger.
”’(centre)LGA1366, LGA1156, LGA775(/centre)”’
Along with the new socket is a new type of retention mechanism. Like LGA775 and LGA 1366, it consists of a metal plate that folds over the CPU and a sprung arm that holds it in position. However, instead of having hinges on opposite sides of the chip, LGA 1156 has the arm and plate hinges on the same side. This means that one fluid motion unlocks the arm and opens the plate. It’s definitely an improvement, though of course is of little real significance unless, like us, you’re swapping CPUs every other day.
Perhaps even more annoying than the change in socket is the change in cooler mount, so you’ll have to factor in an extra £20 or so for a compatible cooler when you buy one of these systems. Some cooler manufacturers may be supplying replacement brackets that conform to the new fitting but they’ll be few and far between.
”’(centre)LGA1366, LGA1156, LGA775(/centre)”’
Retail chips will be shipping with a very similar cooler to previous Intel chips – complete with the rubbish plastic retention clips – but it has a lower profile. We tested this cooler out, indeed we started out benchmarking with it, but very quickly realised that it was holding back performance. As Lynnfield is temperature-aware, it was throttling performance because the cooler struggled to blast away enough heat. We continued our testing with a Thermalright MUX-120 (provided by Intel, so it’s obviously well aware its coolers aren’t up to snuff) and had no further problems, so we suggest you also skip the retail version of this chip, get the OEM one without a cooler, and go for a decent aftermarket cooler.
One of the most interesting additions to Nehalem was Simultaneous Multi-Threading (SMT) or Hyper-Threading as Intel calls it. This is a technique that enables a single physical core of a CPU to act as though it were two, thus improving multi-threaded performance. While the gain isn’t as good as it would be if you added another whole core, it does provide a noticeable boost in performance and uses a lot less silicon.
Now, it’s with Hyper-Threading that Intel has made the whole Core i* number scheme rather complicated. Instead of calling Bloomfield, Core i7, and Lynnfield, Core i5, (and thus making everyone’s lives easier as we wouldn’t have to constantly be referring to internal codenames) Intel has given Lynnfield CPUs with Hyper-Threading the Core i7 800 name and those without Hyper-Threading the Core i5 700 name. Thus we now have to retrospectively call Bloomfield CPUs Core i7 900.
The last thing to note about Lynnfield on an architectural level is its improved Turbo Mode. Turbo Mode was introduced with Bloomfield and it refers to the CPUs ability to dynamically overclock itself, by adjusting its multiplier, depending on workload. When enabled, the CPU will constantly try and work at its maximum speed which, if all four cores are under load, will likely be the CPU’s reference speed. However, if only one or two cores are under load then these cores can be made to run faster, via overclocking, without taking the CPU higher than its thermal envelope (TDP). In other words, while your CPU may be nominally rated at 3GHz, when only one core is working it could be running at 3.33GHz, say.
Now, Turbo Mode was present in Bloomfield but its impact was minimal because it could only increase the multipler by 1 (which when you’re starting with a multiplier of 20, is quite a small change). However, with Lynnfield, Intel has really opened the taps on Turbo Mode, allowing the multiplier to be upped by 4. Specifically, the three CPUs that are being launched today are clocked at 2.93GHz, 2.8GHz, and 2.66GHz but have max single-core Turbo Mode speeds of 3.6GHz, 3.46GHz, and 3.33GHz respectively – not bad for a free performance boost.
One last thing to note about Core i5 CPUs is that they don’t have Virtualization Technology for Directed I/O (VT-d), a recent enhancement to Intel’s hardware virtualisation support. This should be of little concern to the vast, vast majority of home users though.
So that’s the theory out of the way. Now, let’s take a look at the chips and boards themselves.
Looking first at the Core i7 870, it puts in a quite astonishing performance, consistently keeping tabs with the Core i7 965 and more often than not beating the Core i7 920. Admittedly, this isn’t surprising considering its price but it’s still good to see that this apparently mainstream/budget platform has the headroom to compete with the best. In fact, once you consider the power saving of this platform and if you don’t plan to run crazy SLI setups it looks to be the clear choice over Core i7 900, though we’d obviously recommend you go for the much more sensibly priced Core i7 860.
Things aren’t quite so immediately positive for the Core i5 750 with its lack of Hyper-Threading particularly telling in multi-threaded tests. However, once you consider that its default clock speed is only 2.66GHz, it’s clear that Turbo Mode and the improvements inherent to Nehalem are working wonders as this £160 CPU has some of the best single-threaded performance figures in our tests. Most importantly, though, it betters anything at its price point. And again, once you consider its power saving enhancements it’s a clear choice for a bargain CPU.
While many of us are happy just to let our CPUs run at their default clock speed (because overclocking often requires disabling power saving features and CPUs are generally ‘fast enough’ now anyway) it’s always nice to get a bit of free performance where possible. That said, we always feel it’s of limited benefit to spend hours tweaking the living daylights out of a CPU to get the best from it when the amount of overclocking headroom varies between samples of the same CPU. With this in mind, we set out to see what we could come up with just by adjusting the base clock and leaving the CPU to control the rest of its functions, i.e. we left on Turbo Mode, all the power saving features, etc. This resulted in an overclock you might actually use everyday without tripling your energy bill.
With its higher initial clock speed and use of Hyper-Threading the Core i7 870 doesn’t have a lot of headroom for overclocking (without some exotic cooling) as it already generates a lot of heat when going full tilt. Nonetheless, we were able to get this CPU running at up to 3.92GHz in Turbo Mode. That equated to its Cinebench scores rising from 3804 to 4,096 in single-threaded mode and from 14,197 to 15,027 in multi-threaded mode – a percentage increase of eight per cent and six per cent respectively. While this wasn’t a huge increase, with power consumption only going up from 170/255 to 175/265 (idle/load), it’s a nice little overclock you could use everyday. Incidentally, trying to push things further caused power consumption and temperature to sky rocket so we wouldn’t envisage getting much more out of this CPU.
Conversely, the Core i5 750 was one of the best overclocking CPUs we’ve ever encountered. Again without changing anything but the base clock (up to 150MHz) we were able to get this CPU running at 3.8GHz in Turbo Mode. This made its Cinebench scores rise from 3,412/11,294 to 4,154/13,496 (single/multi), an increase of 22 and 19 per cent respectively. Best of all, power consumption only went up to from 168/240 to 180/255 (idle/load). Quite simply, this CPU is a bargain hunting overclocker’s dream!
Intel may have annoyed the computer buying public by releasing two whole new platforms within a year of each other (Intel does insist the two will coexist, by the way) but the simple fact of the matter is Lynnfield is a superb platform. Yes, for enthusiasts, Core i7 900 is still the platform of choice due to greater multi-graphics card support, triple channel memory, and the Core i7 920 being such a great overclocker. However, Core i7 800 provides much the same performance as Core i7 900 but costs less (at least Core i7 860 will) so unless you’re seriously considering multiple high-end graphics cards we’d stick with Lynnfield. As for the Core i5 700, it’s set to be the ”budget” CPU of choice with great performance out the box, some really great overclocking potential, and superb power consumption figures. As expected, AMD has dropped its prices with the Phenom II X4 965 Black Edition now topping the range at just £155 but we have to say, if you’re buying a new system, Core i5 is the way to go.
”’Core i7 870”’
”’Core i5 750”’
Unlike other sites, we thoroughly test every product we review. We use industry standard tests in order to compare features properly. We’ll always tell you what we find. We never, ever accept money to review a product. Tell us what you think - send your emails to the Editor.