Pentium 4 FSB 800 MHz and New Flagship i875P
On April 14 Intel plans to announce a new Pentium 4 processor clocked
at 3.0GHz though a model running at 3.06GHz was released over half a year
ago. The real difference is the FSB speed which makes 800 MHz. So, a year
after the Pentium 4 shifted to the 533MHz bus, this line has made one more
expected step increasing the FSB's rate by 50%.
In due time we will get faster processors with FSB 800 MHz. But what is more
important for most users who can't afford top-range processors is that lower-level
models also equipped with the 800MHz bus are expected in May. Both this Pentium
4 3.0 GHz and other upcoming models will support Hyper-Threading
which brings certain benefit (or loss) by making two logic processors out of one physical.
In the whole line of Stepping D1 with multiple VID different processors can have
different voltages in the range of 1.475 to 1.525 V (for those based on 0.13
micron technology), and heat dissipation of this particular Pentium 4 3GHz makes
81.9 W. This is noticeably more than the future lower models are going to have
(under 70 W).
The 800MHz processors require new chipsets to run because currently there's
no chipsets that officially support this FSB frequency (though overclockers can
try to reach 200(800) MHz FSB with their old boards, and you will see an example
below). But even an overclocked system will be just partly
balanced, because most chipsets before the epoch of FSB 800 MHz supported
only single-channel DDR memory, while the speediest DDR400 had only 3.2
GB/s. The bandwidth of the 800MHz bus of the new Pentium 4 is twice higher
(6.4 GB/s), and there are other devices which need this band. Logically,
new chipsets from Intel and other companies will support dual-channel DDR
memory which doubles this figure (up to 6.4 GB/s with DDR400 used). Today
we will test one of such chipsets.
Intel 875P aka Canterwood
By analogy with last year's line which supported FSB 533 MHz, i875P is
the most efficient model in the expected series. On the other hand, this
chipset doesn't have principal architectural differences from upcoming
Springdales. Take a look at the chipset's diagram:
Numerous new features certainly need to be described. Since i875P is positioned
as a substitute for i850E, i.e. as a high-performance solution, it supports only
533/800 MHz FBS (Willamette based processors do not support the new chipsets anyway).
i875P works with DDR333 or DDR400 memory in the dual-channel mode; the latter
memory type is supported only by processors equipped with the 800MHz bus. But
system can actually work with FSB 400 MHz with DDR266 memory. By the way, the
combination of 800/333 will force the memory work at 160(320) MHz - Intel says
that it improves the sync modes, and therefore, boosts performance. The memory
controller has conventional functions (for example, like that of iE7205);
it utilizes the dual-channel mode with 2 (or 4 in pairs) memory modules of the
same size and chip organization, and symmetrical contents of slots relating to
The chipset, being a desktop hi-end solution, supports memory with ECC enabled.
Besides, it features the Intel PAT (Performance Acceleration Technology) to make
up for the delay caused by error correction. It accelerates a data flow through
MCH (unnecessary in this case sync buffers inside the chipset are not used, data
are transferred directly) at 800/400 (FSB/memory). When the frequencies coincide
the syncronization is minimal, and the fast silicon (with higher signal speed)
used in i875P allows to refuse unnecessary buffers while "normal" chipsets
still need them. According to Intel, the benefit achieved (in comparison with
similar chipsets aka Springdale) makes 2-5%. It's not much, but it's something
(though only with the 800MHz bus and dual-channel access to DDR400). Besides,
it doesn't require any drivers or fooling around in BIOS.
Beside the support of the dual-channel memory controller, fastest FSB
(for desktop CPUs), and AGP 8x the north bridge (MCH) of the Canterwood
sports a CSA bus (Communications Streaming Architecture). This solution
will ensure high-speed connection of various devices to system resources,
as well as access with a guaranteed bandwidth of a dedicated channel. Today
there is only one solution for this bus - Gigabit Ethernet Intel PRO/1000
CT chip (Kenai II CSA), and no information is available on such chips from
other companies. Such an expensive chip will hardly attract many board
makers, and the CSA might remain unneeded.
It's obvious why the engineers had to make a dedicated bus for the 1Gbit
network adapter (in fact, CSA was created for this task), which, at the
same time, goes from the north bridge: it's because of the lack of throughput
in the previous solutions. As you know, the maximum throughput of a PCI
bus, which is used for data transfer in case if an external network card,
is 133 MB/s (actually, it will be less if we account for service data).
It can limit 1Gbit network adapters, especially because of other PCI devices.
Besides, a data flow from the PCI bus must go through various "matching"
processes at the level of interhub connection controllers (for Intel's
chipsets) in both hubs. If a disc subsystem is active, the channel which
transfers data from the network controller gets narrower, and even integrated
Gigabit Ethernet chips get bounded since they are attached to the south
bridge. That is why it's better to connect the network controller directly
to the north bridge (closer to the memory) via the bus of excessive throughput
of 266 MB/s.
The south bridge also sports some new traits. Now it supports up to
8 USB (2.0) ports, and incorporates a Serial ATA controller supporting
a RAID array (in ICH5R south bridge). Other features are standard: AC'97
v2.2 (up to 6 audio channels), 6 PCI 2.3 BusMaster devices, 2 [Parallel]
ATA100 channels etc. VIA has announced a south bridge marked VT8237 with
even a bit better characteristics, but its actual production will commence
Now let's have a look at the integrated SATA controller. The good news
is that no special drivers are needed for operation of SATA devices in
any OSes. But in OSes not native for IDE (which do not support an arbitrary
number of ATA controllers and need IRQ14 and IRQ15 for two logic IDE channels
- all Microsoft's OSes except Windows 2000/XP) only the compatibility mode
can be used. In this mode two IDE channels (a maximum of 4 devices) disable
either one of PATA channels or both SATA ports, that is why devices connected
to them won't be seen.
In case of Windows XP (Windows 2000 is still studied) the ICH5R south
bridge allows for a RAID0 array but only on two hard drives connected via
SATA ports. Besides, Intel Application Accelerator RAID Edition initiates
the process of creation of a RAID array in the background mode after the
second hard drive is connected (with the OS installed on the first one)
It doesn't need OS reinstallation and flows almost unnoticeably for a user.
It's still studied whether RAID1 can be realized. If the outcome is positive,
you will need just to update the software, like in case of technology certification
for Windows 2000. Unfortunately, we didn't have enough Serial ATA hard
drives at the time of writing this review, and we couldn't test this component
of the south bridge.
Finally, Intel's new line of chipsets needs only a 4-layer PCB design.
It lowers the costs for mainboard production in comparison with the first
desktop dual-channel Intel E7205. It was achieved at the expense of some
optimizations in outputs and in MCH (the MCH chip turned at 45°, auto control
of DRAM channels frequency syncronization, terminating resistors on die
Intel Pentium 4 3.0 GHz (200x15), HT enabled, Socket 478
Intel Pentium 4 3.06 GHz (133x23), HT enabled, Socket 478
i875P based mainboards:
2x256 MB PC3200(DDR400) DDR SDRAM DIMM TwinMOS, CL 2 (also used as DDR333)
512 MB PC4200 RDRAM RIMM Samsung
Video card: NVIDIA GeForce FX 5800 Ultra
Hard drive: IBM IC35L040AVER07-0, 7200 rpm
OS and drivers:
Windows XP Professional SP1
SiS IDE 2.03
SiS AGP 1.15
Intel chipset software installation utility 5.00.1007
Intel Application Accelerator 2.3
Intel Application Accelerator 188.8.131.526 (для i875P)
NVIDIA Detonator XP 42.82 (VSync=Off)
CPU RightMark 2.0 RC3
Futuremark 3DMark03 Pro
VirtualDub 1.5.1 + DivX codec 5.02 Pro
discreet 3ds max 5.1
MadOnion 3DMark2001 SE build 330
Gray Matter Studios & Nerve Software Return to Castle Wolfenstein v1.1
Croteam/GodGames Serious Sam: The Second Encounter v1.07
Here are brief characteristics of the boards tested today for the first
||ASUS P4C800 Deluxe Gold
||ASUS P4PE Black Pearl
||ASUS P4C800 Deluxe Gold
||ASUS P4PE Black Pearl
||Intel 875P (RG82004MC + FW82801EB)
||Intel 875P (RG82004MC + FW82801ER)
||Intel 845PE (RG82845PE + FW82801DB)
||Socket 478, Intel Pentium 4 (support HT), Intel Celeron
||AGP Pro/ 5 PCI
||AGP/ 5 PCI
||AGP/ 5 PCI
||AGP/ 6 PCI
||1 FDD, 2 COM, 1 LPT, 2 PS/2
||4 USB 2.0 + 2 connectors with 2 USB 2.0
||6 USB 2.0 + 1 connector with 2 USB 2.0
||2 USB 2.0 + 2 connectors with 2 USB 2.0
||4 USB 2.0 + 1 connector with 2 USB 2.0
||1 + 1 on backplate (not supplied)
|Integrated IDE controller
||ATA100 + SATA
||ATA100 + SATA RAID
|External IDE controller
||Promise PDC20378 (ATA133+SATA RAID)
||AC'97 codec Analog Devices AD1985
||AC'97 codec Analog Devices AD1980
|Integrated network controller
||3COM Marvell 940-MV00 (Gigabit Ethernet)
||Intel 82547EI (Kenai II CSA - CSA Gigabit Ethernet)
||4 Mbit AMI BIOS v2.51
||4 Mbit Intel BIOS v6.00
||3 Mbit Phoenix AwardBIOS v6.00
||4 Mbit Award BIOS v6.0
||ATX, 30.5x24.5 cm
||ATX, 30.5x24.5 cm
||ATX, 30.5x21.5 cm
||ATX, 30.5x23 cm
|Set of accessories
|Adjustment and overclocking settings
Today we will test performance of the i875P vs. memory controller's mode and
compare the chipsets including tests of the processors close in frequency on the
old and new buses. In the second part we will use the i875P based board which
officially supports both FSB speeds, and one of numerous boards announced this
week which is built on i845PE and officially supports 800 MHz FSB and DDR400 memory.
The specs of i845PE doesn't provide for such mode; this is ordinary overclocking:
the bus speed is raised up to 200 MHz, and the memory can be jacked up to 400
MHz. Such approach is possible on most boards which support overclocking. The
recently announced boards have this mode declared officially. In both cases with
the 3.06GHz CPU (on 533MHz bus) the memory can work only as DDR333 (2xDDR333 of
i875P). Unfortunately, the SiS655 based board that we had in our lab refused to
start on the 3GHz processor, but we will try it on another model as soon as possible
to find out whether SiS' latest chipset is ready for the new processors.
Test results for i875P
The memory controller of i875P was tested in 4 modes: single- and dual-channel
ones for DDR333 and DDR400. The tests were carried out on Intel's board as it
arrived earlier, but ASUS's model performed faster, and it was honored to represent
the chipset in the overall tests. The scores of ASUS P4C800 for the dual-channel
DDR400 are given in this section to estimate its advantage over Intel's solution.
The speed growth in the chain DDR333->DDR400->2xDDR333->2xDDR400 is linear
(it ideally extends to 2xDDR400 on ASUSTeK's board), and the final difference
between the two chain ends makes 10-20% in the real applications and up to 70%
in synthetic ones. The only exception is that in WinRAR DDR400 outscores 2xDDR333:
I'm not sure about the real reason, but usually all queer results in this test
(and in all archiving tests) are accounted for by the limited memory write speed.
The gain is pretty good, the priorities in selection of memory for i875P are obvious,
but it's more important to test the fast memory on the new chipset in comparison
with the older market players.
Let's start with the CPU RightMark.
The block of equations solving doesn't benefit from the increased FSB speed,
but it clearly shows the CPU difference of 67 MHz (it's even a bit greater as
the clock speed on the mainboards deviates from the rated value), and the Pentium
4 3.06 GHz takes the palm.
The rendering block makes for the lower CPU speed with the higher memory exchange
rate in i875P, but its pre-caching is still too effective, and the performance
doesn't depend much on memory, and doesn't depend on FSB at all.
The effect of the accelerated CPU bus and memory access will be estimated with
SPEC CPU2000 which tests performance of a system in several popular algorithms
used in scientific calculations. Here we will show only the overall scores.
Both tests depend on the memory, and the i845PE becomes an outsider even when
overclocked. This effect looks worse in SPECfp which always noticeably reacts
to acceleration of FSB and memory, while in SPECfp this looks like a formal defeat.
The most important conclusion is that i850E, after the tough and equal fight with
SiS655, loses to i875P! So, if you are concerned much about speed in calculations
you should go with Intel's new top chipset and processors with 800MHz bus support
(for desktop PC). However, the advantage of the faster FSB makes just 2-3% in
SPECint and 5-6% in SPECfp (but the memory acceleration should be accounted for
as well). If you compare the finished i875P-based top solution and the previous
i850E-based leader, the gap will be more tangible: 5% and 9% respectively.
By the way, the above situation is similar to CPU Marks of the updated game
benchmark 3DMark03 (the 3D section is omitted as it's used mostly for video accelerators).
Now we are going to give a complex estimate to the new processor as a component
of the system assigned by the chipset. Low-level memory tests.
i845PE with its single-channel controller and a low write speed might forget
about leadership in real applications. Even the high read speed with FSB 800 MHz
won't help it. Among the dual-channel models the best are SiS655 and i875P (with
the current BIOS version for ASUS P4T533 its performance in the low-level tests
has changed again, and the i850E loses the first position even in the write speed,
the best test for RDRAM). But the Canterwood takes the lead with the 800MHz FSB.
Well, Intel's new chipset has proved the right to be called the top player.
The memory subsystems will be compared with the applications which most of
all depend on memory - archivers.
Here the i845PE scores the best results with the new processor. The i875P outdoes
the rest by 10-12% with the older CPU, not to mention its victory with the FSB
800 MHz (it outruns the fastest models on 533MHz bus by 17-21%).
MPEG4 coding depends on the memory stream rate. The throughput of DDR400 is
not sufficient for i845PE. i875P coupled with 2xDDR333 shares the first place
with SiS655 coupled with 2xDDR400. Pentium 4 3.0GHz looks the best here thanks
to the Canterwood support, but its advantage over the summit of the 3.06GHz processor
is minimal (about 2%) and achieved mostly at the expense of the chipset.
Now the final rendering in 3ds max - it's obvious that the performance does
not depend on the memory controller and FSB (both in CPU RightMark and, for example,
in MP3 encoding with Lame).
The professional 3D graphics test SPECviewperf gets a push from the memory
acceleration, and, together with the increased FSB's speed, i845PE almost
catches up with the former leader i850E. i875P breaks away from the old generation
by 12-16%. A processor is not a determining factor here; it's more important to
deliver data to the video accelerator via the AGP bus as fast as possible.
And the synchronous dual-channel memory controller on the Canterwood helps it
In games i845PE coupled with DDR333 falls into the last position and catches
up the old dual-channel chipsets when bundled with DDR400 (and FSB 800 MHz). The
i875P is ahead. Its breakaway is especially noticeable when it's teamed up with
2xDDR400. The advantage is not that great as in the archivers: here it runs into
8-10% in last year's real games in the low resolution and graphics quality, and
it falls down several times in newer games and higher resolution and image quality.
Obviously, in games it is a graphics card which is a determining factor.
The most interesting data are brought into the table. Please, let
us know what you think about such a way of data representation.
|i845PE(FSB 800) vs. i850E(FSB 533)
|i875P(FSB 800) vs. i850E(FSB 533)
|i875P(FSB 800) vs. i845PE(FSB 533)
|i875P(FSB 800) vs. i845PE(FSB 800)
|P4 3.0 GHz vs. P4 3.06 GHz (on i845PE)
|P4 3.0 GHz vs. P4 3.06 GHz (on i875P)
The i875P chipset has some unique features like support of the integrated
Serial ATA controller or even SATA RAID, of new Pentium 4's FSB and a dual-channel
memory controller. It will be a good purchase for those who earlier wanted
i850E or iE7205. This chipset provides all possible support, and the only
downside is a lack of FireWire ports. PCs based on such chipset are not
going to be too expensive.
If you compare its performance with the maximum scores of Intel's chipsets
of the previous generation, the benefit will make 5-20%. But most readers
are not interested in the overclocking results, and in comparison with
i845PE+DDR333 (+CPU on 533MHz FSB) the gain i875P achieves is 10-35% (it
refers to all applications which markedly depend on memory). Although it
may be incorrect to compare mainstream and top-performance chipsets, all
functions and speed characteristics of Canterwood will hardly differ from
the Springdale series, and we can judge about performance of the May's
line by the i875P. Besides, the new chipset has nothing which would send
priced of finished systems up, and I hope to see i875P based systems with
a rational price tags (though they won't be too low due to a great number
of integrated features).
As to the gain from combining the Pentium 4 with the faster bus, it's
for you to decide whether it's much or little. Do remember that the speed
rises also at the expense of the chipsets. Some will be satisfied with
fps and seconds reached; others will be disappointed in the architecture
or some CPU makers because the performance does not double (at the expense
of the dual-channel memory) or at least rise 1.5 times (at the expense
of the bus); some others will be waiting for new chipsets and boards (instead
of the processors) which will bring richer functions at lower prices... As
new solutions will arrive in our lab, we will give you all ins and outs.
Write a comment below. No registration needed!