Introduction
Some people tried to interpret the meaning of the name “Slot-1” and came to the conclusion that it stands for Intel who wants to control this market all on one’s own. After losing ground on the low cost sector, Intel is now starting a new attack to regain market shares with their Celeron processor and Slot-1, now with 128 KB L2 cache on-die (Celeron 300A and 333) which practically reaches the same performance as a Pentium II at the same clock speed. It’s not exagerated to say that Intel has created some kind of own competitor for the Pentium II CPUs up to 333 MHz, for overclockers even up to the 450 MHz top model.
Intel tries to spread their CPUs almost emphatically and since the Intel chipsets are not always cheap enough for low end multimedia or gaming computers, somebody else has to provide an adequate product. Cheap chipsets of course lead to cheap motherboards, and the more cheap Slot-1 motherboards become available, the bigger ist the amount of CPUs Intel would be able to sell. SiS and VIA now have got all necessary licenses for a Slot-1 chipset. VIA is now shipping the Apollo Pro chipset and SiS offers the 5600/5595, both ready for 66 and 100 MHz FSB.
The end user benefits in two ways from Intel’s efforts: First the price war could become harder again, since some people who were out for a K6-2 could change their mind and get a Slot-1 system, and second most low end CPUs like the Celeron are already fast enough for most applications. After this review you will know if the new chipsets could force you to come to some kind of compromise or if they can be recommended as real alternative to Intel’s BX chipset.
SiS 5600/5595
The new Slot-1 chipset from SiS consists of the north bridge 5600 in a BGA package and the south bridge 5595. A single chip solution would have been even cheaper, but it seems as if this isn’t possible yet. Let’s take a look at the technical data:
5600 North Bridge
- 487-ball BGA package, 3.3V CMOS
- up to 100 MHz FSB support
- support for single Pentium II or Celeron CPUs
- integrated DRAM controller supports SDRAM, EDO and even FPM memory as well as ECC (max. 768 MB), 3/6 memory banks (double/single sided)
- AGP 1x or 2x interface
- Supports SideBand Addressing (SBA) mode
- Max. 4 PCI busmaster slots
- Integrated PCI EIDE controllers (PIO mode 4 or UltraDMA mode 2)
- PC98 compliant
5595 South Bridge
- 208-pin PQFP package, 0.5V CMOS
- PCI-to-ISA bridge
- ACPI 1.0 and APM 1.2 compliant
- Suspend-to-RAM and suspend-to-disk supported
- USB interface, 2 ports
- SB link and IІC serial bus supported
- ISA bus interface, IRQ and DMA controller, timer, real time clock and keyboard and PS/2 mouse controller
- Voltage (5x), temperature (1x) and fan (2x) monitoring capability
- Support for real time clock and LAN wake up
- Jumperless CPU core frequency setting supported
The data sounds promising. Integrating the keyboard controller saves costs and room on the motherboard. The memory controller also supports EDO and even FPM memory which does only work in case of 66 MHz FSB, but leaves open all paths of equipping memory. I would really like to see a Pentium II motherboard offering the option of clocking the main memory asynchronously to run the Pentium II over 333 MHz at 66 MHz memory clock. As you’ve seen in our review regarding the AMD K6-2’s performance at different bus and memory speeds, the most performance relevant factor is the L2 cache speed, not the memory clock speed. The same applies to Slot-1 CPUs. Only the “old” Celeron without L2 cache would lose performance considerably, but for Pentium II users such an option would be great.
VIA Apollo Pro
Today the Apollo MVP3 is one of the most popular socket 7 chipsets on the market. Of course VIA now wants to introduce an attractive chipset for Pentium II or Celeron CPUs as well.
VT82C691 North Bridge
- 492-pin BGA package
- Supports single Pentium II or Celeron
- 66 / 100 MHz FSB
- AGP v1.0 and PCI v2.1 compliant
- Supports SideBand Addressing (SBA) mode
- Memory controller supports SDRAM, EDO and FPM memory, max. 1 GB RAM, ECC
- 4/8 memory banks (double/single sided)
- max. 5 PCI busmaster slots
VT82C596 South Bridge
- 324-pin BGA package
- PCI-to-ISA bridge
- PC98 compliant
- ACPI 1.0 and APM 1.2 compliant
- USB controller, two ports
- Integrated PCI EIDE controllers (PIO mode 4 or UltraDMA/66)
- ISA bus interface, IRQ and DMA controller, timer, real time clock and keyboard and PS/2 mouse controller
The chipset does also support EDO or FPM memory which theoretically allows to use your EDO memory with 66 MHz CPUs. But I doubt that many manufacturers will equip their boards with SIMM sockets. Also EDO DIMMs were not very common and thus rare. The final release of this chipset includes an asynchronous memory mode, which makes it possible to use a 100 MHz FSB CPU with 66 MHz memory. The only board which does support this is the Tekram P6Pro-A5. More later.
The chipset does have one technical advantage compared with the Intel or SiS chipsets: It already supports UltraDMA/66. It’s questionable whether this is a feature which should be present today, but of course it’s no disadvantage.
Test Setup
Playing around with pre-release or prototype boards is not very funny since it takes quite a while to find out which BIOS setting can be enabled safely and to discover which others cause hang ups. This applies particularly to the SiS board, which are reference boards from SiS. The Apollo Pro boards we got are the KA-6100 and VB-601-V from FIC and the Tekram P6Pro-A5. Since VIA and SiS are now entering a market which is very demanding, I wanted them to challenge one of the very best. So the BX competitor was the well known Asus P2B. All boards ran with a 400 MHz Pentium II.
I benchmarked with the fast DVGS high end harddisks, since the results shouldn’t be influenced by slow or little optimized IDE drivers. By the way, the VIA chipset revision which was used on those boards does not support the busmaster mode, so that’s another reason for using SCSI in these benchmarks. The final revisson which will be shipped does support it, of course. The operating systems I used were Windows 98 and Windows NT 4.0 Service Pack 3.
I would have used a Riva128 card, but the VIA chipset board didn’t permit this. Neither the Asus V3000 nor the Canopus Total3D 128V ran with the nVIDIA drivers 1.50H under Windows 95 or NT. So I used the Matrox Millennium II for the WinStone benchmarks. To determine the game performance I used a single VoodooІ 3D accelerator equipped with 12 MB memory and the games Quake II and Incoming at 640×480 and 800×600 resolutions. Of course I used the actual 3Dfx reference drivers (Glide 2.53, Glide3 3.0, Direct3D drivers 4.10.01.0180-2.17) and Microsoft DirectX 6. All benchmarks ran with 128 MB PC-100 SDRAM (2x 64 MB, LGS, 7 ns) and activated L2 ECC.
The Tekram board also offers an asynchronous memory mode. I also added benchmarks with 66 MHz memory to show you how small the difference really is (comparison with the LX board P2L97 from Asus).
SiS 5600 Motherboards
The first board is equipped with the maximum amount of three DIMM sockets and four PCI slots. Two ISA slots have been placed on this board besides the controllers and other interfaces. The manufacturer decided to use the Award BIOS, which (still?) lets you adjust many settings regarding the memory and bus timings. On the one hand this makes it possible to coax the highest possible performance out of the hardware, on the other hand it’s very easy to ruin the complete software installation if a certain setting is enabled which does not run properly. 🙁
The second board features the AGP 6326 graphics engine as well. It uses the AGP port, so there’s no way to use another AGP card. That’s why I used the on board graphics engine for the Winstone runs, so don’t be surprised if the results are a bit slower.
Both SiS boards we got do not reach the same high memory and PCI bandwith as the other two candidates; but also the VIA boards are only slightly better. This becomes clear especially in the game benchmarks: The Asus P2B motherboard with the 440BX chipset is between 10 (Incoming 640×480) and 30% (Quake II, 640×480) faster than the SiS boards. With higher resolutions the importance shifts from the chipset and its bandwith to pure CPU power. A much faster CPU (500+ MHz) would be necessary to fathom the motherboard’s limits.
It’s only fair to mention that the BAT SiS motherboard does not yet support the required asynchronous AGP clock. Thus the AGP bus runs at 100 MHz which prevented the usage of the Millennium G200. Of course this is a performance advantage for this board. Nevertheless this isn’t too important, since the performance difference between a PCI video card at 33 MHz bus and an AGP card at 66 MHz is also very small.
The memory compatibility test was not what I would call convincing: At 100 MHz the 256 MB Samsung modules did not work at, the 64 MB Toshiba modules (8ns, PC-100) only ran very instable even at the slowest memory timings. But our good old PC-66 memory from Samsung (10ns) was very reliable. LG Semicon’s PC-100 DIMMs ran fine, the 32-chip PC-66 module as well. I made exact the same experiences with both boards.
FIC KA-6100 with VIA Apollo Pro
FIC also offers an ATX board with the Apollo Pro chipset, but the sample we got is the baby AT version. It is a board crowded with connectors, chips, jumpers or other parts to make the board’s size as small as possible. In this case I think the components are so close together that it’s almost impossible to plug in the PS/2 mouse port cable if the CPU is already installed. The clear jumper instructions which FIC usually prints on the board do suffer from this.
Even a soundchip (Yamaha) has been placed on the board. The idea FIC had was certainly good, but using a board which is one or two inches longer would improve this lack of space. Due to this, the board comes with three DIMM sockets, but also only three PCI slots. Here I have to criticize two things: First it is very difficult to plug in a long PCI card, because the IDE connectors are both right in front of the PCI slots. All people who use both ports will have to fiddle around with the cables to arrange them properly. Second the board’s IRQ assignment system is not very advanced; every slot has its fixed IRQ. The PCI slot one does use the same IRQ as the AGP slot, and in slot three my Adaptec controller collided with the IDE controller. In slot two if worked fine, but than there’s no chance to run two VoodooІ cards in SLI mode on this board; the only way would be getting a very large SLI cable (I don’t know if that’s a good idea) or using a VoodoІ card with two chipsets on one board (e.g. Quantum 3D X-24).
The memory tests were passed mixed: The LGS 7 ns PC-100 DIMMs ran fine, but the Toshiba PC-100 memory ran fine only if just one module was installed. Two modules lead to exception errors and blue screens. The 256 MB module did not run at all, but then again the 32 chip module with LGS chips (PC-66, 10 ns) ran properly as well as the PC-66 memory from Samsung.
FIC VB-601-V with VIA Apollo Pro
Here we have the second board with the Apollo Pro chipset. This one is an ATX board with 5 PCI slots, 2 ISA slots, the AGP slot and four DIMM sockets. The other boards use the Award BIOS – this one comes with AMI. A lot of performance influencing settings can be changed, but most of them lead to instabilities or hang ups during the Business Winstone runs.
After trying and playing and configuring one whole day I finally got the benchmark results; FIC added a note which said that this board used a chipset revision which is not going to be shipped. Luckily, because of the trouble you can really easy get into it’s not very good to start optimizing…
The low price will make such a motherboard a good competitior for LX and EX motherboards. Power users should only go for a BX board.
Due to the difficulties of the pre-release chipset I don’t want to evaluate the board itself now. The memory tests were not too successful: Only the Samsung 10 ns memory and the PC-100 type ran reliable at 100 MHz – all others caused problems. The 256 MB module did not want to run at all once again. It seems as if such XXL memory modules are a problem for many motherboards.
Tekram P6Pro-A5 with VIA Apollo Pro
As mentioned above, the Tekram motherboard allows the main memory to run at the AGP clock. This of course makes it possible to use PC-66 SDRAM memory, so you have two ways of saving money: First a VIA Apollo Pro board is much cheaper than Intel BX or LX boards, second you can avoid buying new PC-100 memory.
The Winstone performance is as fast as with the KA-6100 (and some slower than good boards with Intel BX chipset), but there was an interesting phenomenon with the game benchmarks: At 66 MHz FSB the benchmarks are as fast as we would expect it, but at 100 MHz all results were clearly slower (using a VoodooІ card). Another run with a Riva128 card shows that the board can’t be condemned for this. There seems to be some problem with the VoodooІ at 66 MHz FSB or with some drivers. I added a chart with the according benchmarks to show you that the difference between 66 and 100 MHz memory clock is really small in deed.
The board has 5 PCI slots, but only 3 DIMM sockets. But it is also the first Apollo Pro board which is suited for overclockers: Besides the standard clock speeds of 66 and 100 MHz you can also chose 75, 95, 112, 124 and even 133 MHz FSB! I did not do detailed testings, but the only clock speed which seems not to be absolutely reliable is 133 MHz.
The Benchmarks
If you took a closer look at the Quake II benchmarks you will have noticed slower benchmark results at 100 MHz than at 66 MHz with the Tekram board. There seems to be some problem with the VoodooІ cards making them at 66 MHz faster than at 100 MHz. The following chart makes clear that this doesn’t happed with another AGP video card.
Comparison with the Intel LX chipset
Summary / Conclusion
Starting a complete new chipset for a basically different system is not quite easy. The manufacturers of chipsets (or motherboards) don’t get the chance to play around with the cache timings in Slot-1 systems any more, so they have to try optimizing the bus and memory performance. In the last months and even years Intel has proved this to be one of their best fields. That’s why I’m not very surprised that the first generation of non-Intel Slot-1 chipsets wouldn’t be able to beat them.
In Quake II, which is very bandwith-sensitive, both the 5600 and the Apollo Pro chipset are clearly slower than the 440BX, which shows a little bandwidth lack of the newcomers. With Incoming the difference is not that big. The Winstone runs do not differ that much, since large parts of the program’s code is written to optimally take advantage of the fast L2 cache. The Highend Winstone requires a lot of CPU power and makes the differences clear again.
There’s still a lot of work to do for both the chipset and motherboard manufacturers. As you have seen now, the alternative chipsets can not reach for the crown. But they are a quite cheap alternative for all who want to have a Slot-1 system (with a 66 MHz CPU, e.g. Celeron A) which is capable of running all available CPUs.
I personally wouldn’t want to chose between one of the newcomers and a BX board (as power user I would take the BX board), but getting a Apollo Pro board instead of a EX or LX motherboard is now a real alternative. An EX board has only few expansion slots and doesn’t run at more than 83 MHz FSB. An Apollo Pro based motherboard may have twice as much slots, one or two more DIMM sockets, 100 MHz capability, ECC support and the option of running the memory at the AGP clock at safe 66 MHz. And if you would get both boards for the same price…
On the one hand neither VIA nor SiS are able to offer a chipset which would beat Intel’s BX, but on the other hand lots of new board with both new chipsets will be available soon and poach market shares in the low cost sector which Intel was originally trying to regain.
