iXBT Labs - Computer Hardware in Detail






Intel 965 Express Chipsets with Official Support for Core 2 Duo/Extreme

With the launch of i965 and i975, the 900 chipset series from Intel has nearly come to an end. This series was initially so revolutional, that its discussion even eclipsed the eternal "A vs. I" dispute for some time. However, i915 successors (i925X is not worth mentioning, as it's architecturally undistinguishable from the i915P) added so insignificant functional changes that according to the old Intel nomenclature, i965 chipsets may be called something like i915EP. Indeed, i925XE introduced support for the 1066 MHz FSB, i945/955 — DDR2-667. Of course, a much more important reason to launch the i945/955 was to create a platform for dual-core processors. The i965 had a similar reason to appear. However, let's proceed right to the details.

Intel P965/G965 Express

A noticeable difference of the new products is a modified naming procedure - letters are now put before the numeric index (however, the reasons for it are not clear). This flow chart shows the key characteristics of the G965 (P965 differs only by the lack of integrated video):

Here is a brief list of the key features of the 965-series chipsets:

  • Support for all "old" Celeron and Pentium processors (including dual-core processors) with 533/800/1066 MHz FSB, as well as all Core 2 Duo/Extreme processors
  • Dual-channel DDR2-533/667/800 memory controller, up to 4 DIMM modules with the total size of up to 8 GB (without ECC), Fast Memory Access and Flex Memory technologies
  • Graphics interface PCI Express x16
  • Integrated graphics core GMA X3000 (only G965) supporting Clear Video technology
  • DMI bus (~2 GB/s bandwidth) to the ICH8/R Southbridge
  • Up to six PCIEx1 ports
  • Up to four PCI slots
  • 4/6 (4 in ICH8, 6 in ICH8R) Serial ATA II ports for 4/6 SATA300 devices (SATA-II, the second generation of the standard) supporting AHCI mode and NCQ (ICH8R only)
  • Support RAID (ICH8R only) 0, 1, 0+1 (10) and 5 with Matrix RAID function (the same array of drives can be used in several RAID modes — for example, two drives may form RAID 0 and RAID 1, each array will have its own part of the drives)
  • 10 x USB 2.0 devices that can be disabled individually
  • Gigabit Ethernet MAC controller and a special interface (LCI/GLCI) for a PHY controller (i82566 for Gigabit Ethernet, i82562 for Fast Ethernet)
  • High Definition Audio (7.1)
  • Binding for low-speed and outdated periphery, other things.

Let's examine some issues in detail

CPU support. Compared to the previous chipsets, all i965 models initially support Core 2 Duo/Extreme processors. However, a chipset does not require some special functions for this support, Conroe processors may work on motherboards with other chipsets as well — for example, Intel's Core 2 documentation initially mentioned the i975X. Here lies a nasty reef: the first i975X-based products do not support Conroe, and the problem is not in BIOS versions, but in hardware incompatibilities of the platform. (Most motherboard manufacturers have launched a new revision or even a new model of a motherboard on i975X for Core 2 processors.) It's also quite possible that different i945 modifications (not i955X — for marketing reasons) will be officially validated for Conroe. But it hasn't been done so far, so the best choice for a non-professional will be to buy a i965 motherboard, as all these models are initially designed for Core 2. (Specialists are recommended to watch for BIOS updates of i945-based motherboards.) By the way, i965 chipsets do not officially support quad-core processors (Kentsfield core).

Memory support. There was added 400(800) MHz memory clock and Fast Memory Access technology. Besides, maximum memory size was increased: "junior" chipsets (i915/945) used to support 4 GB maximum. The first change is a purely marketing issue — according to our tests, even bandwidth of dual-channel DDR2-667 memory is excessive for all modern Intel processors (even DDR2-533 bandwidth is excessive for most processors with 800 MHz FSB). DDR2-533 and DDR2-667 got some chances only with extremely low timings. By the way, memory clock is automatically set to 533 MHz for processors with a 533 MHz FSB (mostly Celeron D). Flex Memory appeared even in the i915. It consists in dual-channel memory addressing, when the slots of both channels are populated with different memory sizes: interleaving, which helps to get faster access to memory, is done for the memory volume of the doubled memory size of the smaller module (couple of modules), the rest of the larger memory module (a couple of modules) is addressed linearly.

We are more interested in the new Fast Memory Access technology, which should theoretically provide better performance. In fact, it's a combination of technologies for the i965 memory controller, which allows to detect overlapped commands (for example, reading from the same memory page) and then to rearrange their execution so that overlapped commands are executed one after another. Moreover, traditionally lower-priority writes are now scheduled for the time, when the read queue should be empty. As a result, the process of writing into memory hinders reading even to a lesser extent. Note that Fast Memory Access is used in the i965 for all interfaces (FSB, DMI, PCI Express x16, integrated video), so not only the data exchange rate with a processor is raised.

PCI Express x16. The only moment here worth commenting on has to do with two video cards operating in tandem. Indeed, the i955X was officially certified for ATI CrossFire. And the i975X also allowed to split the graphics interface x16 into x8+x8 (i955X could work in x16+x4 mode, where the x4 supported peripheral ports of the Southbridge). Even before the i965 was announced, it became known that it wouldn't get CrossFire certification. That is it wouldn't allow to use two ATI video cards in tandem on the level of drivers (like it was done by NVIDIA for all "non-native" chipsets). Now that ATI is bought by Intel's competitor (AMD), the i965 has zero chances for future official CrossFire support.

GMA X3000 integrated video. That's the fifth generation of modern integrated video from Intel, faster (at least theoretically) and supporting new 3D standards. We are planning a separate article about the G965 integrated video, because now we can only list the specifications: as the drivers are not ready yet, the G965 is not present on the market as a sterling product, though the chipsets are already shipped to manufacturers and the motherboards on this chipset are ready. (As for us, we can only confirm that we failed to run any game from our test procedure using the July version of the GMA X3000 drivers.) Clear Video technology is intended for hardware acceleration and improvement (deinterlacing + color correction) of video playback (including HD). It should also offer digital interfaces (including HDMI) for video output. Intel notes that the X3000 is optimized for Windows Vista Aero experience. So, a well rigged computer based on the G965 with integrated video is capable of supporting the Microsoft Windows Vista Premium logo program.

ICH8/R Southbridge. There are no global changes in the new Southbridge, no new industrial interfaces. But nevertheless, it features some important differences from its predecessors. The number of PCI Express x1 ports in all ICH8 modifications is increased to 6 (ICH7 used to have only 4), which makes no difference to a regular user — there are enough interfaces for integrated controllers anyway, and separate devices (which are not integrated into a motherboard — sound cards, TV tuners, etc) are practically not available in stores. The number of SATA-II ports in the ICH8R was increased to six, but now there can be only four PCI Bus Master devices. Intel's southbridges have finally caught up with NVIDIA products in terms of USB support — 10 ports are supported now. There is a theoretically useful feature for system administrators - enable/disable ports individually. Besides, up to recently NVIDIA has been the only company, which chipsets used Gigabit Ethernet MAC controller (even two of them in the nForce 500 series) — now Intel also has such functionality. But due to a specific interface, only Intel products can be used as a PHY controller (with the corresponding effect on costs). (This modification of Gigabit Ethernet reduces the number of available PCIEx1 ports by one, as their interfaces outputs are combined.)

No support for PATA(IDE). However, the ICH8 has one more peculiarity that raises most questions. The fact is that this is the first chipset without Parallel ATA support. Yes, ICH6 started the tendency by leaving only one channel in the PATA controller (for two devices), some chipset manufacturers gradually adopted it. And now Intel delivers coup de grace. But are users ready for it? Intel is used to being a locomotive of the industry. But its actions very often look like running before a train. Out of doubt, SATA interface is more convenient from the "physical" point of view: cables are more convenient, PCB layout is simpler, functionality and performance are higher. If we could just do without diskettes to install Windows! However, most users are quite content with SATA controllers in Native IDE mode, when the OS can be installed without pressing F6. There are no problems with SATA drives either. They have been available in stores for a long time. They even offer some advantages over PATA models.

But there is still a big problem with DVD drives. Their SATA modifications are extremely rare. Besides, limited as they are in their choice, users have to pay through their nose "to join the civilization" (without getting any other advantages). As far as we know, drive manufacturers are sluggish about this issue, because the yield from making mass products is very low. And it threatens to sputter out when they have to install an additional bridge or to overhaul the interface controller. Perhaps Intel decided "to press through" the production of SATA models — Intel's chipsets make up a considerable part of the market. It will become clear in the nearest future whether this plan is a success. As for now, buyers of i965-based motherboards have to face this problem. They will be helped by motherboard manufacturers (including Intel), who will add an additional controller. Unfortunately, not all IDE controllers work well with ATAPI devices (CD/DVD drives). When you have to connect a PATA hard drive to it, the results may be unpredictable. All potential buyers of i965-based motherboards are recommended to inquire about the operation of a given model in forums.

Q965. In conclusion of the theoretical part, we'll say a few words about Q965. That's not the first Intel's chipset, initially designed for a business segment — along with basic models in the chipset series (like P and G modifications), the company also manufactures several junior models (PL, GL, etc) with curtailed functionality, but much cheaper and popular among system integrators. In this case, Q965 (and its curtailed modification - Q963) are manufactured for the vPro business platform. So it's based on the G965, but uses only "serious" features of integrated video, while Clear Video is locked (perhaps, on the driver level). In return, the Q965 uses a modified Southbridge — ICH8DO (Digital Office) supporting Active Management Technology. AMT is a set of technologies that facilitate computer administration: remote control and system restore, gathering information on workstations, network isolation of an infected computer, etc. As part of the VIIV platform, the G965 uses ICH8DH Southbridge (Digital Home) supporting Quick Resume Technology, etc. However, this information is irrelevant for the absolute majority of motherboards in stores, as you will see only motherboards based on P/G965 with ICH8/R on the shelves.

Performance tests

Testbed configuration:

  • CPU: Intel Pentium 4 660J (3.6 GHz, 800 MHz bus, Prescott-2M core), Socket 775
  • Motherboards:

  • Memory:
    • 2x512 MB PC2-5400(DDR2-667) DDR2 SDRAM DIMM Corsair (CM2X512-5400C4)
    • 2x512 MB PC2-8000(DDR2-1000) DDR2 SDRAM DIMM Corsair (CM2X512-8000UL)

  • Video card: ATI Radeon X800 XT 256 MB
  • HDD: Seagate Barracuda 7200.7 (SATA), 7200 rpm


  • OS and drivers:
    • Windows XP Professional SP2
    • DirectX 9.0c
    • Intel Chipset Drivers
    • Intel Chipset Drivers
    • ATI Catalyst 5.2

  • Test applications:
    • RightMark Memory Analyzer 3.62
    • 7-Zip 4.10b
    • WinRAR 3.41
    • DivX 5.2.1 Pro codec
    • XviD 1.0.2 (29.08.2004) codec
    • SPECviewperf 8.01
    • Doom 3 (v1.0.1282)
    • FarCry (v1.1.3.1337)
    • Unreal Tournament 2004 (v3339)

Test results

We decided to compare the representative of the new chipset generation with the fastest motherboard from the older generation. Formally, this comparison is not well posed, because the i955X is positioned higher than the i965 (on the level of the i975X). But according to our tests, the difference between the i945 and the i955 is quite insignificant. Besides, it looks like the new model has an improved memory controller. We cannot compare integrated video results yet (for the above mentioned reason) — perhaps, we'll write a separate shootout.

We'll traditionally start with low-level analysis of memory potential in our RightMark Memory Analyzer. Interestingly, the maximal data read rate in the i955X is a tad higher. This chipset is also a tad faster in equal conditions. At the same time we can see that the i965 gets faster as the memory clock grows. Though in terms of bandwidth, there must be no difference starting from dual-channel DDR2-533.

But the P965 is still a tad faster in writing, all modes tested demonstrating the same results. Maximum memory write rate is expectedly the same in all chipsets (when the effect of a CPU cache is barred) - it's higher by 1% in the Gigabyte's model, which FSB clock is raised by this value.

Memory latency tests seem to illustrate a defeat of the P965, but let's have a closer look at the results. The new chipset is outperformed by the i955X not only in similar modes and even in all operating modes — just compare DDR2-533@3-3-3 with DDR2-667@4-4-4. Considering the cycle time, memory timings in the first case are 11.25ns, in the second case — 12ns, which must provide lower latency of DDR2-533 memory. Indeed, that's the case with the i955X. The i965 demonstrates a different picture: DDR2-533@3-3-3 is outperformed by DDR2-667@4-4-4 as well as DDR2-800@5-5-5 (12.5ns timings). In fact, this diagram resembles the one with read rates: the higher the memory clock, the higher the results. That may be the effect of the optimization of the new memory controller for DDR2-800.

However, there is another factor that may explain such results. Let's not forget that with all its advantages, RMMA is a purely synthetic test. Its code does not resemble typical user applications. Considering "advanced intelligence" of the i965 memory controller, we can easily imagine that attempts to optimize execution of the "wrong" code in this test results in growing overheads. You must admit that any user will prefer a product that demonstrates low results in synthetic tests, but leads in real applications. Well, let's see whether the memory controller optimization will reveal itself in our assortment of tests from the "real world".

It's very interesting to compare archiving results: DDR2-533 mode is actually not very good in the P965 (low results everywhere), the i955X is victorious under the same conditions. But DDR2-800 modes with extremely low timings bring victory to the new product (note however that we didn't test the i955X in DDR2-667@3-3-3 mode).

As usual, video encoding speed, measured according to our open test procedure, is practically the same in all cases. A little advantage of the P965 here can be explained by a slightly raised FSB/CPU clock.

Professional 3D applications and games demonstrate the same picture, we have already seen it many times. Here are the main points: the P965 is slightly outperformed by the i955X under the same conditions; using memory with a higher clock (up to DDR2-800) has a positive effect on the i965 performance; a combination of these factors makes the P965 outperform the i955X (which was not tested in DDR2-667@3-3-3). Well, it's time to draw a bottom line.


The new series of desktop chipsets from Intel offers no outstanding features, which will definitely make users buy P/G965-based motherboards. This conclusion can be shaken only by the tests of the G965 integrated video, which are impossible so far, because the drivers are not ready yet. Perhaps the most likely reason to buy an i965-based model these days is because it's absolutely ready for Core 2 Duo/Extreme processors. But if you wait a little, you'll be able to choose a similar (but cheaper) model on i945 or competing chipsets. ICH8 Southbridge raises most doubts as for rationality of choosing the i965 chipset — cutting out the built-in PATA interface seems too high a price for increasing the number of PCIEx1 and USB ports (and, by the way, reducing support for PCI devices). Intel's built-in gigabit controller is also an attractive feature, but it's not unique. Besides, the cost of this solution will certainly be higher than for those from Marvell or Realtek. And all the features of the built-in controller are revealed only in ICH8DO.

Speaking of performance, we can note two points. Positive: the i965 is no slower than the fastest of its predecessors (i955X), it's probably even faster. Negative: the new chipset succeeds only with fast low-latency memory; in case of DDR2-533, it is sometimes outperformed by the i955X by indecent 10%. However, we may rush our performance conclusions: we compare the first P965-based model with the fastest motherboard we ever tested. Well, in future we shall compare performance of several representatives of the new chipset generation and bear out or reject this conclusion.

Sergei Pikalov (peek@ixbt.com)
September 14, 2006

Write a comment below. No registration needed!

Article navigation:

blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook

Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.