Today Intel announced its new i845PE and i845GE chipsets, supporting DDR333, and we decided to date the comparison of all current Pentium 4 chipsets to this announcement. Actually already in May, after the release of new Pentium 4 533 MHz FSB processors we reviewed all existing chipsets, supporting that FSB. This review can be considered an update, especially as i845PE and i845GE are not a subject for a separate review.
i845PE and i845GE
Already in May it was clear that i845E and i845G won't officially support
DDR333, as this standard was adopted too late these products were developed without
it in mind. In May it was also clear that updated chipset version would be released
in fall, supporting DDR333 under any conditions. i845G + PC2700 testing proved
the significant boost, provided by this memory, and possible appeal of such a
combination for users not willing to use RDRAM for some reason (though our tests
have repeatedly proved that i850E provides the best performance today). At that
our researches proved great instability of i845G-based boards operating with DDR333
(due to this even today we are forced to use the same Chaintech 9BIF0, as in May),
so today's chipset couple is attractive at least by simultaneous performance and
second important advantage of i845PE/GE is their prospect. As you know, already
in November Intel is to announce the first Hyper-Threading-enabled
Pentium 4. According to preliminary data, i850E and i845E chipsets are potentially
capable of supporting this technology, and i845G/GL is not (Hyper-Threading will
be featured only in Pentium 4 with 533MHz FSB, so there's no sense to speak of
older chipsets). Though it seems impossible to use already released i850E and
i845E with such processors, as the new models (3.06 GHz would be the lowest in
the line) require some technological changes for the support. At the same time
i845PE/GE boards should be initially designed for these features.
Finally, speaking of announced chipsets, we should state that i845PE, like its i845E predecessor, is a discrete chipset, and i845GE - integrated, featuring the same graphics, as used in i845G. Well, actually there's a slight difference between these integrated products: i845GE has higher graphics core clock (266MHz against 200MHz of i845G). A minor increase it is (speaking of absolute performance expected), and graphics memory, being an allocated buffer, has the same clock as well. Nevertheless, it's still interesting to look at the performance increase.
This chipset, despite its marking resembles older variants, is a significantly
improved SiS model for Pentium 4. It's one of the first (it's hard to be more
precise, as Taiwanese announcements do not always indicates the readiness and
release) AGP 8x chipsets and the first that we saw actually supporting DDR400.
Naturally, the latest requires its conditions, but at last our samples worked
with DDR400 modules (you should get the complete list of certified modules from
manufacturer website before the purchase, though SiS has its own certification
program), caused a noticeable performance boost, and that's a very important
indication of actual required clock support. For the sake of accuracy, note that
chipset specs doesn't feature DDR400 support, but this is caused by absence of
a standard (to mention compatibility with).
Chipset Northbridge doesn't offer any more new technologies, Pentium 4 with
533MHz FSB is implied. As for the support of future Hyper-Threading-enabled Pentium
4, it's not that clear, but it seems SiS will release an updated Northbridge revision
for such CPUs. Another thing that's worth a word is SiS648 MuTIOL 1G protocol
for data transfers to and from SiS963 Southbridge, featuring 1 GB/s throughput.
The double increase is the result of using QDR instead of DDR. Can't say that
previous 533 MB/s were not enough, but this innovation is not to spoil the product
for sure, and the ability of SiS to technologically "train" in advance
is surely a plus.
Southbridge, supporting various interfaces is not revolutionary, but still provides
all features of a modern chipset, including FireWire(IEEE 1394a), almost absent
in competing solutions, up to 3 ports using PHY external controllers, but without
Serial ATA. Other Southbridge features include: two ATA66/100/133 IDE controllers;
USB 2.0 (1 EHCI), up to 6 ports; Fast Ethernet MAC supporting 10Base-T/100Base-TX;
6 PCI 2.2 slots; AC'97 2.2 controller supporting 6-channel codecs.
The marking indicates that VIA planned to implement DDR400 support. However, this was unsuccessful, as our samples refused to work with DDR400, and later the company itself admitted there were problems. So, it's turns out to be just a copy of already reviewed P4X333. VIA plans to take a rather usual step: to release P4X400A, actually supporting PC3200. And now we have to wait and test P4X400 boards for the sport of it. To test with DDR333.
Brief comparatives of boards tested:
Initially we planned a larger testing, but it turned out that the difference between i845G/GE(PE), SiS645DX/648, and VIA P4X333/400 using the same memory type is very small and gets within natural variation limits. So, we decided to remove unnecessary information and compare the fastest Pentium 4 chipsets bundled with the fastest memory they support. Though in most cases such solutions show less absolute performance than i850E+RDRAM (especially PC1066/4200 RDRAM), they can be used foe some reasons, not directly related to performance.
Let's traditionally begin with low-level memory tests.
Some i845PE writing speed falloff is a feature of a specific board, not affecting,
according to Wstream results, the general performance. All, except VIA chipset,
have similar marks, and naturally DDR400 SiS648 is the first.
You can see that computing tests are almost not affected by chipset performance,
as only CPU clock and cache are important (permanent in our testing).
MPEG4 encoding is the first real-world problem, enabling the competitors to
show themselves. Three Intel and SiS chipsets are very close, P4X lags by about
5%, and SiS648 with faster memory gets 4% advantage.
4MB WinAce archiving is traditionally nice test of memory operations performance,
and strangely SiS648 lags by about 8%, and VIA chipset more usually lags by another
5%. Intel memory controller is pretty good. Note that, using DDR400, SiS chipset
expectedly easily wins over i845 couple by more than 7%.
Synthetic SYSmark clearly shows the minority of chipset technical designs
difference in the office app fields - just faster (by theoretical 20%) memory
helps SiS648 to outrun 4%.
Hi-res 3Dmark results depends on graphics cards most, but VIA chipset was
a bit (by 3%) behind. It's also interesting to compare integrated graphics of
Intel chipsets (not with GeForce4 Ti4600, but between two of them) - as you can
see, there's almost no difference.
In Return To Castle Wolfenstein, based on Quake3 and thus more sensible to
memory subsystem, graphics card performance is not the only limiting factor already
at 800x600 and higher. In such conditions, SiS648 + DDR400 is faster by about
5%, and DDR333 keeps up (but lags a bit) with others. i845Gx integrated graphics
looks rather good even on the top NVIDIA product background, and again there's
no effect of i845GE higher chip clock. However already at 1024x768 with higher
settings everything starts to resemble 3Dmark, and VIA chipset loses.
Finally, in Serious Sam 2, that requires more 3D-performance, the difference is minor already at 800x600 (P4X400 still lags, perhaps, it has lower AGP clock), and i845Gx graphics shows its little use for current 3D games.
The situation with Intel integrated graphics and its similarity forced us to conduct an additional test that proved the equality. Though everything turned out to be simpler, as Intel graphics drivers 11.3, available at the moment of testing, couldn't recognize new chipset (naming it "i845G/GL Video Accelerator" in Windows) and correctly work with it at higher clock. So the test of higher graphics core clock is the subject of the future.
Actually we couldn't take any interesting conclusions out of the closer look at mentioned chipsets. All of them (except VIA product) continue their lines, and new features is either currently useless (i.e. AGP 8x), or just (though it's good) makes the ready solution cheaper, as any peripheral controllers are widely available, so nothing prevents from implementing them onboard.
The only usable innovations, currently supported by only SiS648, is DDR400(PC3200) memory, that is not (and will not be) standardized but actually exists and boosts the performance. So, if you wish to build a max-performance Pentium 4 system without RDRAM, pay attention to this bundle. Even in case of unavailability or higher price of DDR400, using SiS648 with DDR333 let it be the first in almost all tests.
It seems that the only serious minus of SiS chipset is its supposed incompatibility with future Hyper-Threading-enabled Pentium 4. Of course, you can wait for a newer revision or even an updated chipset, but you can also purchase Hyper-Threading-ready i845GE and i845PE boards (depending on your needs). These chipsets were the best with actually popular PC2700 and, moreover, their only minus is traditionally high price.
Pity, but we can't say anything good about VIA P4X400. Mentioned problems
should have made the company produce better chipsets or not to do anything. And
now we have a partially inoperable product, that always lags a bit, has cloudy
prospects and doesn't have any unique interesting features. We can recommend such
boards only to fans of VIA or to those willing to save money (but why purchase
new Pentium 4 then?).
Write a comment below. No registration needed!