Introduction

Less than ten weeks after the official launch of the first 32-nm processors from Intel, the chip giant sends a special copy in the race. Because code-named “Gulftown” nothing more than the first six-core processor, which will be presented for desktop operation behind. With the start of the 16 March 2010, the market leader wants to show what you can do from the 32-nm technology. The Clarkdale processors as core i3 and core i5 has already shown that this production is certainly impressive. But with the Intel Core i7 extreme Edition 980X is not bigger, but properly geklotzt.

Six real cores, which thanks to Hyper-Threading be twelve virtual and a clock of 3.33 GHz, which can draw on the Turbo mode depending on the load behavior and here and there therefore gladly once 3.6 GHz available, are only the cornerstones of the processor. Because there is also a larger L3 cache rich 12 MB and the CPU can suffer some tweaking of the Westmere architecture and access such as on the new AES instructions. For all’ this will not increase the TDP, it remains at 130 Watts.

We will show more precisely what has done else everything in terms of technical changes and specifications themselves, on the next page. This is our detailed benchmark test followed as usual, followed by various special tests, in which we determine also the potential of the 1,000-euro processor in terms of overclocking.

Overview

The new flagship of the House of Intel 1.17 billion transistors is difficult. This high number of transistors, which as usual on the fast cache – cache – for the most part, packed in a 248 mm² large Intel the. This is the the six-core processor, significantly less than the a “Lynnfield”-quad-core CPUs, which puts it in 296 mm². Also the same-socket “Bloomfield” are still unbeaten with a die size of 263 mm2. Purely arithmetically, the newcomer is almost exactly the three times of the “Clarkdale”-two-core processors, which have a die size of 81 mm square. The factor of three can be seen almost 1: 1 for the transistors, because a Clarkdale brings it to 384 million, the Bloomfield as is well known to 1.17 billion. Simply put the Bill “Gulftown Clarkdale = 3 x” is so true.

< div >

Mouseover shows the processor labeled details

But it as almost always is not that easy in life then but also again. Because Intel puts the new six-core processor on the end platform around the socket LGA1366, during Lynnfield and Clarkdale trust the new mainstream socket LGA1156. 210 Area promises greater socket connection of Triple-channel DDR3 memory. This will work but as with maximum 1.066 raised in Bloomfield, DDR3 1333, continue to each track is missing from the new flagship. This seems to be the only blemish but only once.

Purely from the platform her it should be no problems with the six-core processor according to Intel, and also various Board. Here, updated BIOS versions were ready accordingly for several X 58 motherboards for some months. Therefore, our 18-month-old ASUS P6T Deluxe had no problems with the detection of Bloomfield with current BIOS.

The first Hexapeptides or six-core processor launches with rich 3.33 GHz. It is just as fast as the previous flagship on the basis of the same base of the beat. But the 32-nm newcomer comes with some additions. Conditionally by the cores, the entire L2 cache increases to 1.5 MByte, the L3 cache even for a whopping 12 MByte. So to continue the line of the first 32-nm CPUs based on the “Clarkdale” per core also 512 KB L2 cache and two cores have 4 MB L3 cache. The QPI clock of the new flagship is also the 6.4 GHz the set to the highest tact, Intel currently offers on the processors. The feature range includes all current SSE instructions, virtualization also dominate the Bloomfield. Features such as Turbo, hyper-threading and the AES-NI, we lose a few words in the next section.

&Lt; div > Photo Gallery “Intel’s Hexa-core ‘Gulftown’ processor” (18 pictures)

Hyper-Threading, Turbo, AES & co

Also the latest addition to Intel relies on the Turbo mode, as well as support for Hyper-Threading. The Turbo acts with an increase of maximum 266 MHz in the presence of only a core, all cores can permanently work higher stroke depending on the design of the cooling solution with 133 MHz. This has the effect that the processor load almost permanently works with 3.46 GHz, and are even 3.6 GHz available in specific applications that are tailored to only a core. Which admittedly rather minimal differences, which accounts for a Turbo enabled or disabled, we indicate in our benchmark course.

Hyper-Threading is a still still often bemused technology that promises a lot of efficiencies in a six-core processor with then a total of twelve threads however. According to Intel are for Hyper-Threading only about five percent more transistors on the complete the necessary here and since flows in performance increases by more than 30 percent in theoretical tests and always up to 20 percent in real-world applications. Also, we show separately to the differences between enabled and disabled, Hyper-Threading in each test.

Finally, there is still the AES function, which is becoming with the Bloomfield on the socket LGA1366. Beforehand there was in this connection much confusion, but the Gulftown based on Westmere architecture in 32 nm production comes with the new features. Overall, there are seven new instructions, which had presented Intel with Westmere. Six of them are grouped under the term “AES New Instructions” (AES-NI). Namely the functions AESENC, AESENCLAST, AESDEC, AESDECLAST, AESKEYGENASSIST and AESIMC, as an additional improvement takes PCLMULQDQ. Intel provides details about these features in more technically large documents on your own website.

The AES feature can be very quickly prove thanks to SiSoft Sandra. Because this benchmark also on the number of cores and threads scale, an increase is at the end of the six-core processor to the dual-core model to see 200 percent. The old flagship core i7-975 extreme Edition with same frequency is formally sawn with a 1,700 percent increase.

Too many threads!

The following image can be a run on the one hand the mouth together water, but it can also lead to problems, to the one previously not thought. Because too many threads can also encounter resistance. Not applications are meant but, no, it the problem of children has the years: the games. You having a hard time still in many area, to support more than two cores, these will be tolerated at least slowly without performance loss. But more than four threads are offered, which can cause again problems, which is why last year the Hyper-Threading technology has been demonized by Intel of players. But this has improved with Windows 7, Hyper-Threading now no longer automatically leads to a loss of performance – of poorly programmed games apart. Exactly we looked closer one of these not as optimized games.

Already in our recent tests, it was noticed that armed assault 2 (ArmA II) very nervously reacts to additional cores/threads. While the Lynnfield processors with four real and four virtual cores had many problems, “Clarkdales” moved with two plus two cores very well from the affair. Thus, it was clear that this game are allergic not reacted hyper-threading, but on too many threads. More than four threads take the game off course, the performance goes down. You keep this in mind, is clear, that is with the “Gulftown” virtually the nemesis of the game at the start. And it was what had to come: the game failed six cores with a crash shortly after the start the service.

It is known to help but we know as a resourceful tester. Using msconfig is a CPU from the core i7-980X simply with only four cores and eight threads. And behold, with this constellation the game works as before. One can easily operate ArmA II without Hyper-Threading and thus also “only” with six real cores, needs to be mentioned probably hardly separately at this point. By the way the version 5 + 5 not ran also, with eight virtual cores give conclusion simply.

As we operate in our testing field with a slightly older version of ArmA II, we have us caught the last patch and hoped for an improvement. And in fact: the current version of 1.05.62017 works. However the problem of armed assault 2 shows us, that there are still problems. Whether the desktop world is ready for six cores with Hyper-Threading, depends immensely by the applications and also from the games.

A new boxed cooler

Intel powered the core i7 extreme Edition 980X a new cooler. This has nothing with the previous boxed solutions which are trimmed on price/performance, more or less common, because the manufacturer now offers a large Tower cooler with supposedly much potential. The name “DBX-B” sounds only once not very exciting, but the differences from the previous boxed fan are ubiquitous. Four heatpipes, a large surface area and for the first time a connector to a rear panel are the stark opposite to the hitherto offered. The tightly integrated cooling fans, blue lights in operation, will be in the idle around 800 rpm not louder than 20 dB(A). If the temperatures rise, the fan on up to 1,800 RPM turns up and to then 35 DB(a) einpegeln. The customer has the choice by means of a small switch, whether he the “performance”- or wants to go to the “quiet” mode.

&Lt; div > “Intel Core i7-980X with new boxed cooler” Photo Gallery (20 pictures)

Because we use one of the most powerful solutions in our processor test for two years with the Noctua NH-U12P with NF-P12 fan, the new CPU cooler must prove decided against this. As this here is no cooler test and should be, we point to only the temperature under full load in the default state of the processor – in the Quiet-, as well as in performance mode. First of all, there is but slight installation problems. As to properly mount the new Intel cooler, you need a very long screwdriver, to combine the screws through the cooler through with the back panel. Since there is this very long screwdriver for small Phillips screws but rather rarely, standing with a normal budget model quickly on the hose. Scope of Noctua’s flagship, which contains an extra long screwdriver creates workaround here. This hurdle has been overcome, sits the DBX-B stuck – like all cooler bolted before him.

The performance mode is not mentioned by Intel in the specifications. Doing the fan to just turns up 2,600 rpm, where he is clearly audible. A quiet work is quite simply no longer possible under full revolutions in this setting, because even the 1,800 rpm not quietly occur an untrained radiator tester. The cooling performance is evident without a doubt, but she will be bought with a higher volume. The old boxed cooler for the same socket while also works with up to 2,100 rpm, achieved not the volume of the new model however by far. The Noctua works with maximum 1,300 rpm, which turns out as an almost inaudible. However, he achieved lower temperatures than the new boxed model DBX-B Intel already fair quiet mode.

Anyone who acquires an expensive processor, should more spend a euro for a good cooler. We know this wisdom not only since the present day; our many cooler tests prove it each month, several times. The new CPU cooler Intel DBX-B comes only with the core i7 extreme Edition 980X. There are no plans to commercially offer it separately or to ask for more CPUs so far.

Test system

Our test system has seen almost 60 processors last year; Time to weed out older models and start a new beginning with a new platform. Here we have used also many suggestions of the reader, which we received in the last few months. The biggest change to the previous year is therefore the graphics solution, which as of now a Radeon HD 5870 of MSI is used. This card is characterised by a very low energy requirement in the idle, if it is necessary, but performance fed up.

In addition to the necessary components for each platform, such as different mainboards, the power supply with “80Plus-Silber” certification remains the same. Also used the “cooler master stacker 832 RC” continues with its two integrated 120 mm fans, to allow a possible fair and realistic comparison between the counterparty and at the same time to come across any thermal problems of counterparties.

As the operating system, put we looks like the other software used in the new edition on Windows 7 in the 64-bit variant, benchmarks will be broken down into the section. There are following all other details of the test system:

• Complete breakdown of the test system and the components used.
  • Processor
    • Hexa-core
      • Intel
        • Intel Core i7-980X extreme Edition – 3.33 GHz, 1.5 MB L2 cache, 12 MB L3 cache, QPI 6.4 GHz (Gulftown B1)
    • Quad-core
      • Intel
        • Intel Core i7-975 extreme Edition – 3.33 GHz, 1 MB L2 cache, 8 MB L3 cache, QPI 6.4 GHz (Bloomfield D0)
        • Intel Core i7-960 – 3.20 GHz, 1 MB L2 cache, 8 MB L3 cache, QPI 4.8 GHz (Bloomfield D0)
        • Intel Core i7-870 – 2.93 GHz, 1 MB L2 cache, 8 MB L3 cache (Lynnfield B1)
        • Intel Core i5-750 – 2.66 GHz, 1 MB L2 cache, 8 MB L3 cache (Lynnfield B1)
        • Intel Core 2 Quad Q9550 – 2.83 GHz, 12 MB L2 cache, FSB1333 (Yorkfield C1)
        • Intel Core 2 Quad Q8200 – 2.33 GHz, 4 MB L2 cache, FSB1333 (Yorkfield r0)
        • Intel Core 2 Quad Q6600 – 2.40 GHz, 8 MB L2 cache, FSB800 (Kentsfield G0)
      • AMD
        • AMD Phenom II X 4 965 Black Edition – 3.40 GHz, 2 MB L2 cache, 6 MB L3 cache, HT 2.0 GHz (Deneb C3)
        • AMD Phenom II X 4 925-2.80 GHz, 2 MB L2 cache, 6 MB L3 cache, HT 2.0 GHz (Deneb C2)
        • AMD Athlon II X 4 620-2.60 GHz, 2 MB L2 cache, HT 2.0 GHz (Rana C2)
    • Triple-core
      • AMD
        • AMD Phenom II X 3 720 Black Edition – 2.80 GHz, 1.5 MB L2 cache, 6 MB L3 cache, HT 2.0 GHz (Heka C2)
        • AMD Athlon II X 3 435-2.90 GHz, 1.5 MB L2 cache, HT 2.0 GHz (Rana C2)
    • Dual-core
      • AMD
        • AMD Phenom II X 2 550 Black Edition – 3.10 GHz, 1 MB L2 cache, 6 MB L3 cache, HT 2.0 GHz (Callisto C2)
        • AMD Athlon II X 2 250-3.00 GHz, 2 MB L2 cache, HT 2.0 GHz (Regor C2)

          All Athlon II and Phenom II processors have been, unless otherwise indicated, with C & Q and measured C1E disabled (out). We had a comprehensive study on this subject at the beginning of the year 2009 in the program. The memory controller was operated on all AMD systems in the mode of “UnGanged”.

      • Intel
        • Intel Core i5-661 – 3.33 GHz, 512 KB L2 cache, 4 MB L3 cache (Clarkdale C2)
        • Intel Core i3-530 – 2.93 GHz, 512 KB L2 cache, 4 MB L3 cache (Clarkdale C2)
        • Intel Core 2 Duo E8600 – 3.33 GHz, 6 MB L2 cache, FSB1333 (Wolfdale E0)
        • Intel Core 2 Duo E8400 – 3.00 GHz, 6 MB L2 cache, FSB1333 (Wolfdale E0)
        • Intel Core 2 Duo E7400 – 2.80 GHz, 3 MB L2 cache, FSB800 (Wolfdale r0)
        • Intel Pentium G6950 – 2.80 GHz, 512 KB L2 cache, 3 MB L3 cache (Clarkdale C2)

          The integrated graphics core of Clarkdale processors is disabled for the period of pure processor test, since the Radeon HD 5870 used came. A look at the processor including graphics performance provides our separate analysis.

  • Motherboard
    • AMD
      • ASUS M4A79T Deluxe (AMD 790FX + SB750) – socket AM3 – revision 1.01G – BIOS release (October 28, 2009).
    • Intel
      • ASUS P6T Deluxe (Intel X 58 chipset) – socket LGA1366 – revision 1.02G – BIOS: 2004 (28 January 2010)
      • ASUS P7P55D (Intel P55 chipset) socket LGA1156 – revision 1.02G – BIOS: 1207 (December 7, 2009)
      • Asus P5E3 premium (Intel X 48 chipset) – LGA775 socket – revision 2.00G – BIOS 0803 (May 21st, 2009)
  • Memory
    • 2 x 2,048 MB DDR3-1333 OCZ Platinum (CL7-7-7-20-1T, 1.50 volts, dual-channel)
    • 3 x 2,048 MB DDR3-1066 OCZ Platinum (CL7-7-7-20-1T, 1.50 volts, Triple-channel)
    • 2 x 2,048 MB DDR3-1066 OCZ Platinum (CL7-7-7-20-1T, 1.50 volts, dual-channel)
  • Graphics card
    • MSI Radeon HD 5870, 40 nm, 1024 MB GDDR5, reference clock
  • Peripherals
    • Samsung HD501LJ, 500 GB (SATA-II hard disk)
    • MSI DR8-A (burner)
  • Power supply
    • Cooler master UCP 700 watts (Silver 80Plus)
  • CPU cooler
    • Noctua NH-U12P with NF-P12 fan
  • Driver versions
    • ATi catalyst 9.11
    • Intel chipset driver 9.1.1.1020
  • Software
    • Microsoft Windows 7 64-bit
    • Microsoft DirectX 11

Benchmarks

In addition to the completely new designed test system, we have subjected also the benchmark course an update. Some theoretical tests are still with the part of, because these are usually a step as any real application. There, new features that have found a place in processors be checked first in their performance. Because these theoretical tests but use not much users in front of the PC, we consult many practical applications. Various multimedia applications, but also everyday things like packaging files include today in the first place. The test is rounded off by some games. On this point, we have divided the test course. Once we evaluate the games low resolution at high details without quality-enhancing resources (AA/AF), the second part draws on the currently most used resolution of 1,680 x 1,050 pixels and represents the performance as it is in real use with at least four AA/AF of the day.

As in the past, each benchmark is performed several times, to be able to find any break in the performance. This happens in almost all applications unfortunately more common than expected, both the negative and the positive direction. The average of the five tests (without extreme outliers) is then applied in the single charts.

• Theoretical tests
  • 3DMark Vantage 1.01
  • Cinebench R10
  • SiSoftware Sandra 2010.1.16.11
  • x 264 HD benchmark 3.01
  • WinRAR 3.91 (integrated test)
  • wPrime 2.00
• Applications
  • Autodesk 3ds Max 2010 with “SPECapc for 3ds Max 9″
  • Lame 3.98.2
  • MainConcept H. H.264/AVC Pro 1.6.1
  • Paint.NET 3.36 with PDNBench 3.20
  • PCMark Vantage 1.01
  • SPECjvm2008 (Java virtual machine benchmark)
  • TrueCrypt 6.3a
  • WinRAR 3.91 (real packing)
• Games
  • Anno 1404, full version, patch 1.01
  • Armed assault 2 (“ArmA II”), full version, patch 1.03
  • Call of duty: Modern warfare 2, full version, factory
  • Colin McRae: DiRT 2, full version, factory
  • Far cry 2, full version, factory, integrated benchmark
  • Resident Evil 5 benchmark version
• other required tools
  • CPU-Z 1.53.x beta
  • Core temp 0.99.5
  • Fraps 3.03
  • Prime95 25.11

Theoretical tests

3DMark Vantage

After some years on the knob has the long-serving 3DMark06 and affects not only the graphic now somewhat dusty but in addition, the CPU limit for fast graphics cards is becoming more and more apparent, it was high time for a successor. Accordingly the Finnish manufacturer Futuremark has brought the 3DMark Vantage on the market after a long wait, who was from the outset been programmed for the Direct3D 10 API. But this is already some months ago, we only take with the benchmark in the course, as it is the results for everyone at home quickly understand.

Download: 3DMark Vantage

Cinebench

The popular, from Germany, Maxon Cinema4D is represented within our benchmark show in the form of Cinebench R10. The software uses up to 16 processors to the raytracing and thus benefit from all available on the market desktop processors from AMD or Intel currently. In our test, we present the version Cinebench R10 published in mid-2007. The diagrams show the test as usual on the one hand with only one processor core, on the other hand also the multi-core test, which uses Hyper-Threading.

Download: Cinebench

SiSoft Sandra 2010

No matter whether it comes to motherboard, memory, hard drive, peripherals, cards, processor, network, interfaces, BIOS, Windows or DirectX, SiSoft Sandra has extensive answers at hand. For much of the hardware in the PC, there is also a benchmark tests, the PC test can on its performance compared to some reference computers. All’ these values are however almost exclusively purely theoretical nature and have little relation to the practice, but processors in their theoretical possibilities can be compare well. Also, programs like Sandra usually significantly more offer ways to test newly integrated features of the processors before they are integrated in months or years in real programs. We have arithmetic, cryptography and chosen by memory bandwidth from the extensive repertoire for the tests. It is used the version 16.11 of December 3, 2009.

Download: SiSoft Sandra

x 264 HD benchmark

The “x 264 HD benchmark” was developed by the colleagues from Tech ARP and is provided also for download. Several repeated testing shows the performance when you convert a high-quality video under x 264-aware. The tests will it split in two parts on which we publish both the first and the second part with the version 3.01.

wPrime

For several years, SuperPi among tests such as the Amen in church. Unfortunately the test is only very marginally meaningful due to his age for current processors, which is why he finds no attention with us in the future. As an alternative we offer wPrime. wPrime version 2.00 is end of September 2008 in the appear and provides the best results in our eyes, as all processor cores or threads on Windows full capacity and used.

Download: wPrime

WinRAR

Although 7-zip for free and in terms of compression is superior to many competitors, approaches the software in terms of distribution by far not on WinRAR which is available in the form of RAR for DOS and Windows 3.1. Now WinRAR is able in addition to rar also other formats like 7z, for example, to extract, to compress are however only rar and zip available. Since WinRAR 3.60, now also the popular Packer offers multi core support. It should be noted is the program that with each new version and faster processors, everything a small little faster and better is compressed, so that the lead of 7-zip has been significantly reduced. The program in the version 3.91 must bring the integrated test behind them in this section, the part for the practical application in a separate section follows.

Download: WinRAR