[an error occurred while processing this directive]

Single-channel chipsets for Pentium 4: SiS648FX and VIA PT800


The release of the single-channel i848P chipset and its unexpectedly bright characteristics that put Intel's spring chipsets into an awkward position (they had two channels!) was a real success. In contrast to the i865P, which was past-oriented and had vague prospects, this chipset uses all features of Intel's modern south bridges and has no crippled north bridge components. One memory channel is locked but if 2GB RAM is enough for you, the only disadvantage of the i848P is its speed.

What about the competitors? They have no dual-channel solutions for new processors (with a 800MHz bus) to offer (the SiS655FX/TX and VIA PT880 chipsets are ready but we haven't seen any boards for them). At the same time, single-channel cheaper chipsets (the price was always an advantage of VIA's and SiS's products) were announced quite a long time ago, and today we are going to test first boards on the SiS648FX and VIA PT800. But first comes information on the chipset architectures. 

SiS648FX

Its predecessor got a very high score in its time: at that moment it was the most real competitor for the i850E+RDRAM(PC1066) tandem, and SiS648 based board with the DDR400 were cheaper than the top Intel+Rambus combination. Unfortunately, it had no Hyper-Threading support: the release of the B revision was postponed, and without such support the chipset wasn't of much interest among top solutions. Then the company released the dual-channel SiS655 (let me leave aside the SiS R658) which however didn't break new records. And after the announcement of the new Pentium 4 line with the FSB 800 MHz and a series of dual-channel chipsets (from Intel as well) we got no news from SiS.

Today we will test the first production mainboard on the SiS648FX we received. The product didn't get any vital features, that is why the company just added FX to its name. What's changed in the SiS648FX compared to the SiS648? Have a look at its block diagram.




The picture taken from the SiS site is not adequate :). First of all, the SiS648FX has the official DDR400 support (in contrast to its predecessor), secondly, the 800MHz FSB support is not shown at all. There are no more differences between the chipsets. The official couple of the new product is the SiS963L south bridge, while the SiS648 is coupled with the full version of this bridge - SiS963 (more expensive due to the integrated FireWire). This situation perfectly illustrates the changes: if the SiS648 was developed to be a leader in speed and functionality, the SiS648FX was released for the low-end sector. Remember that all  latest south bridges from SiS are pin compatible and have the same north bridge interface so that you can easily replace them.




Here are brief specs of the chipset (the south bridge features recommended are given for the SiS963L):

There's neither Serial ATA, nor FireWire. However, there's one more interesting feature we'd like to attract your attention to - HyperStreaming. This is SiS proprietary technology integrated in its all latest chipsets. It's obvious that SiS also stepped on the way of pushing its products using beautiful logos and esoteric terms. The matter is that all components of the HyperStreaming technology have been supported in SIS chipsets for over two years already and only now they are united under a single name.

The idea of this technology is that the memory controller in the chipset north bridge can operate in several modes.

  1. One stream - minimal access delays. The data given by SiS are obtained under the specific RTOS (Real-Time OS) and are as useful for an ordinary user as the scores of the Pentium 4 in the Digger. In our tests we will compare the SiS648FX with its competitors under the Windows.
  2. Several streams - pipelined and parallel operation which accelerates their combined processing (in particular, the number of virtual read/write channels is optimized). The company shows the advantage over the competitors' solutions  in the IOMeter test in operation with one or several hard drives. We won't use this test but several streams will be enabled anyway (for example, in games).
  3. Priority stream - guaranteed bandwidth for streams that need isochronism (for example, in case of networking or audio). Such functionality is also provided by other companies (in particular, by NVIDIA). It's difficult to test this feature.
  4. Always - intellectual transaction management (alternation of request so that one open memory page is accessed). It should be effective for all tests sensitive to the memory speed, which are quite a number in our suite.

Well, there's actually nothing revolutionary, let's see what will be the real scores. 

VIA PT800

This chipset is the standard extension of the VIA's line for Pentium 4 processors. The line is renamed from P4X to PT for the unification purposes (with the KT series for Athlon XP) and got the ending 800 (though it's not clear what it indicates). Its predecessor P4X400 was reviewed here (note that the boards on the new stepping of that chipset do support the DDR400, though it doesn't add any speed).




Here are brief specs of this chipset (VT8237 south bridge):




Compared to the P4X400, it supports Intel's new processors, has better memory parameters (size increased, ECC support) and uses a new south bridge which has more USB (2.0) ports and supports Serial ATA hard drives. If we compare the VIA PT800 to its direct competitors (single-channel chipsets for 4), this will be the most advanced chipset in functionality. Then what about its speed? 

Performance

Testbed:

Software:

OS and drivers:

Test applications:

Board ASUS P4P800S-E Deluxe DFI 648FX-ALE ABIT VI7
Links ASUS P4P800S-E Deluxe DFI 648FX-ALE ABIT VI7
Chipset Intel 848P (RG82848P + FW82801ER) SiS648FX (SiS648FX + SiS963L) VIA PT800 (PT800 + VT8237)
CPU support Socket 478, Intel Pentium 4, Intel Celeron
Memory slots 3 DDR 3 DDR 2 DDR
Expansion slots AGP/ 5 PCI/ ASUS Wi-Fi AGP/ 5 PCI AGP/ 5 PCI
I/O ports 1 FDD, 2 COM, 1 LPT, 2 PS/2 1 FDD, 2 COM, 1 LPT, 2 PS/2 1 FDD, 1 COM, 1 LPT, 2 PS/2
USB 4 USB 2.0 + 2 connectors of 2 USB 2.0 2 USB 2.0 + 2 connectors of 2 USB 2.0 4 USB 2.0 + 2 connectors of 2 USB 2.0
FireWire 1 port + 1 connector of 1 port (VIA VT6307) - -
Integrated ATA controller ATA100 + SATA RAID ATA133 ATA133 + SATA RAID
External ATA controller - - -
Sound Analog Devices AD1985 AC'97 codec C-Media CMI9739A AC'97 codec VIA VT1616 AC'97 codec
LAN controller Intel 82547EI (CSA Gigabit Ethernet) integrated Fast Ethernet integrated Fast Ethernet
I/O controller Winbond W83627THF-A Winbond W83697HF Winbond W83697HF
BIOS 4 Mbit AMI BIOS v2.51 2 Mbit Phoenix AwardBIOS v6.00PG 4 Mbit Phoenix AwardBIOS v6.00PG
Form-factor, dimensions ATX, 30.5x24.5 cm ATX, 30.5x20.5 cm ATX, 30.5x19.5 cm
Average current price (quantity of supplies) $140(4) $67(18) $87(29)

Test results

First come the low-level memory controller tests, especially with the Hyper-Threading enabled... or rather the HyperStreaming in the SiS chipset.

The latency is not low at all in the SiS648FX (though SiS doesn't provide comparison exactly with the i848P and PT800). Judging by the data obtained the Intel chipset can take the lead (with a traditionally high write speed) and VIA's new solution will take the middle position. Now let's run our favorite benchmarks to reveal the real difference in the way the memory controllers work.

Teh best benchmark is the one that compresses a sizable volume of data with some archiver of a complicated algorithm and a large dictionary that uses tens and hundreds of megabytes of memory. The SiS falls behind by 16%, while the PT800 yields less than 3% to the leader.

In case of MPEG4 encoding we don't have such difference: SiS648FX has caught up with the VIA and even outscored it (i848P beats both by 3%).

Remember that not all applications are dependent on the memory speed, for example, MP3 encoding depends exceptionally on the CPU clock.

In case of one of the best external rendering modules used for a scene in the 3ds max (Brazil Rendering System) the SiS648FX falls a little behind its competitor (about 3%).

The Photoshop scores look curious: the VIA outscores the SiS by 3%, but the Intel 848P unexpectedly falls into the last place losing 1.5% to the SiS648FX. I don't see any reasons for that. However, that was the only artifact in its operation.

You can see that all the scores in the games look equal. The SiS remains an outsider, the VIA PT800 outdoes it by 5-7%, and the i848P by 6-10%. 

Conclusion

All three single-channel chipsets for the modern Pentium 4 (I don't take into account ALi's products) have identical features which are actually defined by the north bridge. There's nothing to improve until the transition to the PCI Express but, and the difference is only in the memory size and type (but not the clock speed). The VIA PT800 looks the best with its ECC support, which is a rare thing for budget chipsets (plus, the i848P has the CSA bus). South bridges can differ, but I also want to distinguish VIA's product VT8237: the ICH5 doesn't support RAID modes for Serial ATA (though there's ICH5R version), and the SiS963L doesn't support Serial ATA at all.

The i848P is the fastest runner in this pack, though the VIA PT800 yields only 3%, but remember that Intel's chipset was represented with the best board among all, while for the VIA only one model was released so far. The SiS648FX loses 3 to 10 % and sometimes even more. Again, there's only one board based on this chipset, but the gap is too great to be ignored. Taking into account scarce functionality and low performance of the SiS648FX among all three chipsets, it will have to have its price cut down to remain competitive. However, the SiS has the newer dual-channel SiS655FX (which adds the FSB 800 MHz to the SiS655, which is neither outstanding nor bad) and its successor is coupled with the SiS655TX (which supports Advanced HyperStreaming). Well, the company does have the dual-channel solutions to offer, and you might remember that the ordinary dual-channel chipsets from Intel (without PAT) are quite close to the i848P in speed, i.e. the market segmentation by the number of channels is rather conditional. 
 
 

Sergey Pikalov (peek@ixbt.com)
Dmitry Mayorov (destrax@ixbt.com) 
[an error occurred while processing this directive]