As we promised, today we've got the latest solution for the Socket A platform - the SiS748 chipset developed by Silicon Integrated Systems. But since it differs from its predecessors in a single parameter, the FSB 400 MHz support, we'll leave aside the discussion of the difference between the 333MHz 400MHz FSBs but make a brief overview of all SiS chipsets made for the AMD platform. It can be useful even for those who are not interested in today's hero :). So, here's the row of SiS discrete (without integrated graphics) chipsets for the Socket A platform as they were launched.

SiS733/735. This is a single-chip solution traditional for "old" SiS (north and south bridges combined). 200/266MHz FSB and DDR200/266 RAM supported (though PC100/133 could work as well). Proprietary (integrated) MuTIOL bus of 1.2 GB/s, ATA33/66/100, AGP 4x, 6 USB 1.x ports (2 host controllers), integrated 10/100 Mbit Ethernet (without PHY), AC'97 interface and ACR/CNR (actually unneeded). There was just one difference between the chipsets: the 733 didn't have the DDR SDRAM support, though the actual difference is in the marking. I can't say these chipsets were popular (I don't remember any boards based on the 733), but there were some solutions even from Chaintech, not to mention SiS partner ECS.

SiS745. This is SiS latest single-chip solution for the Socket A. It supports DDR333 and integrated FireWire controller compared to the 733/735. The other specs are the same: MuTIOL 1.2 GB/s, ATA33/66/100, AGP 4x, 6 USB 1.x ports, 10/100 Mbit Ethernet without PHY, AC'97 (v2.2 vs v 2.1 in the 733/735). It failed to become widely popular though it was the only chipset with the integrated FireWire (IEEE1394a) controller support at that time. There were boards only from ECS/PCChips.

SiS746. The chipset is based on the classical dual-bridge scheme. But it actually looks downgraded relative to the SiS745: it supports such unnecessary stuff as AGP 8x and UATA133, but the FireWire (IEEE1394a) controller is lost. Actually, there are several south bridge PCB compatible versions, including those with the FireWire controller, but today all SiS chipsets come with only the "L" version (without this controller). The bridge was divided so that the customers could integrate the functions they wanted. And the fact that the bridge version with the FireWire controller is not needed looks like a certain tendency...

SiS746FX/748. They are very similar to the SiS746: the 746FX just got the FSB 333 MHz and DDR400 support while the SiS748 supports the FSB 400 MHz in addition. Apparently, the SiS748 will be SiS last chipset for the Socket A platform.

Below you can also see two tables with SiS chipsets, with the north and south bridges going in chronological order (left to right). The table with the south bridges is not of much interest as there is only one bridge (SiS963L, and sometimes SiS962L) used in modern boards based on chipsets for AMD's processors of the 7th generation (Athlon/Duron). Stages when the chipsets stepped up are marked with red.

North bridges of SiS chipsets for Socket A platform:

As you can see, SiS first changed the memory clock speed leaving the FSB as is. But it wasn't an original approach - if you remember, AMD formerly denied a possibility of the Athlon XP with FSB 333 MHz, not to mention 400 MHz. As to the rest of the specs - 3 GB RAM instead of 1.5 and AGP 8x instead of 4x didn't look striking for  experienced users.

South bridges of SiS modern chipsets:

SiS was one of the first companies to provide support for the USB 2.0 and FireWire (the latter is also integrated only in NVIDIA's chipsets). 8 USB ports of the SiS964 can be considered a pure marketing trick, though SiS wasn't the only company that went mad on the ports. 1GB/s MuTIOL 1G looks impressive but I doubt that we will ever evaluate its contribution into system performance. In all other respects, the bridges look good. Those which have the FireWire look even more advanced than Intel's or VIA's ones. However, the 964 doesn't integrate FireWire probably because the SATA+RAID leaves no space for this controller :). But in general, there's only one south bridge for the Socket A.

Before we start the tests let me remind you once again that the SiS748 supports "single-CPU FSB" (400 MHz/Athlon XP 3200+). One of the competitors on the NVIDIA nForce2 Ultra 400 chipset will be the Albatron KX18D Pro II board which wasn't tested before. It's not the fastest board among those based on this chipset but it doesn't matter (later you will see why). 

Performance

Testbed:

• CPU: AMD Athlon XP 3200+ (11x200 = 2200 MHz), Socket A(Socket 462)
• Mainboards:
  • ABIT KV7 (BIOS 11) on VIA KT600
  • Albatron KX18D Pro II (BIOS R1.03) on NVIDIA nForce2 Ultra 400
  • ASRock K7S8XE (BIOS 1.20) on SiS748

• Memory: 2x256 MB PC4000(DDR500) DDR SDRAM DIMM OCZ, CL 2 in DDR400 mode
• Video card: Manli ATI RADEON 9800Pro 256 MB
• Hard drive: Western Digital WD360 (SATA), 10000 rpm (IDE-SATA SABR200HV adapter)

Software:

OS and drivers:

• Windows XP Professional SP1
• DirectX 9.0b
• VIA Hyperion 4.49
• NVIDIA UDP 2.45
• SiS AGP Driver 1.16a
• SiS IDE Driver 2.03
• Silicon Image Driver 1.0.0.32
• ATI Catalyst 3.6

Test applications:

• Cachemem 2.4MMX
• 7-Zip 2.30
• WinRAR 3.20
• VirtualDub 1.5.4 + DivX codec 5.05a Pro
• Return To Castle Wolfenstein v1.1, Gray Matter Studios & Nerve Software
• Serious Sam: The Second Encounter 1.07, Croteam/GodGames
• Unreal Tournament 2003, Digital Extremes/Epic Games/Atari

Test results

Actually, there's no need in comparing scores in one benchmark to those in another. Just look through all the diagrams.

Low-level memory test

Archiving

Media content encoding

3D games

The picture doesn't change at all: judging by the SiS748 based sample we got, this chipset is the slowest among the modern solutions. But the problem is not in the memory access speed because the gap should grow in applications critical to this parameter (up to 20-25%). Well, we didn't expect any other outcome: SiS chipsets could never boast of their speeds if their FSB/SDRAM clock speeds were equal to their competitors'. Their advantage was price and sometimes functionality (SiS745 coupled with FireWire), but not the speed. The SiS748 is no exception. But it doesn't mean that it's a bad chipset: it just depends on where one uses it. It doesn't suit powerful Athlon XP 3200+, but the lower the clock speed, the narrower the gap, right?

Conclusion

The SiS748 is a low-end chipset for low-end processors, adequate to them both in price and performance. Surely, the FSB 400MHz makes no sense, but it doesn't matter. The lower clock speeds are also supported. The release of the SiS748 will probably mean that the SiS746/746FX will be taken out of production. They are really unneeded anymore. The platform itself is not that important for SiS judging by its activity on this market and the number of mainboards based on SiS chipsets for Socket A.

Write a comment below. No registration needed!