The ASUS P55T2P4 is one of the few Intel Triton II (HX) based motherboards that will support reliable operation of the CPU bus at speeds above 66 MHz. The P55T2P4 can easily be overclocked to 75 MHz. Operation at 83 MHz seems problematic, at least with Cyrix 6x86L CPUs, but is possible with 6x86MX CPUs and 60ns EDO DRAM (HX-based motherboards do not support SDRAM). This motherboard has switching voltage regulator circuitry, and supports all present 6x86, 6x86MX, K5, K6, Pentium and Pentium MMX processors.
I have found that the ECS P5VX-Be motherboard will support host bus speeds of 75 and 83 MHz, just like the ASUS (but not as reliably). For 75 MHz operation, jumper the 3-4 pins. For 83 MHz operation, jumper the 1-2 pins. This motherboard supports SDRAM DIMMs.
Why do I insist so much on HX-based motherboards? Simply because HX-based motherboards combine two essential features for GNU/Linux users, besides their excellent performance:
1) Linux bus mastering DMA mode 2 driver compatibility
EIDE DMA mode 2 operation with the excellent Linux triton.c driver written by Mark Lord is only supported by Linux 2.0.x kernels on the Intel VX, HX and TX chipsets. Recently, a patch against kernel 2.1.55 has enabled UDMA mode 2 (with theoretical tranfer rates of up to 33 Mb/s) support on the TX chipset (see the Kernel patches page). The main advantage of bus mastering IDE DMA transfers is the lower protocol overhead and CPU utilization. It doesn't make EIDE as efficient as SCSI-3, but it certainly provides an affordable solution for above-average disk I/O performance.
2) L2 caching with > 64 Mb of RAM
The VX and TX chipsets have been voluntarily crippled by Intel engineers to accommodate some devious marketing plans: they won't cache memory accesses above 64 Mb! If you happen to install > 64 Mb of RAM on such motherboards, expect a huge performance hit on uncached accesses. EDO DRAM is cheap nowadays, and large amounts of RAM will provide much better performance for GNU/Linux systems under some circumstances (e.g. heavily loaded Web, News and Mail servers).
John Howland sent me detailed information on the MTech R534 motherboard, which is based on the SiS 5571B chipset. The SiS 5571B supports the Linear Burst Mode memory read address sequence of the 6x86 CPUs, which is said to provide a 3-4% performance improvement by lowering bus utilization. Since I don't have an SiS 5571B based motherboard, I am unable to verify this assertion.
The following information applies only to current R534 revisions F & G motherboards: switching voltage regulator circuitry (with all current CPU voltages supported), 75 and 83 MHz base CPU clocks, and the option to run the PCI bus asynchronously at the standard 33 MHz, or synchronously at 1/2 the base CPU clock. General quality and motherboard stability are considered good, documentation just adequate. I have no information on the kind of decoupling capacitor used.
A recent patch is supposed to make the Linux DMA mode 2 driver compatible with motherboards based on the SiS 5571B and other chipsets, but again I haven't been able to verify this information. The SiS 5571B and most recent non-Intel motherboard chipsets do not suffer from the 64Mb caching limitation of the VX and TX chipsets.
Cyrix sent me a beautiful Chaintech 5SDA SiS 5598 based motherboard for testing (together with a shiny new 6x86MX-PR233 chip rated at 2.5x75MHz). This high quality motherboard has everything one could dream of in terms of 6x86 processor support: switching voltage regulator circuitry, 512Kb pipelined burst cache, Linear Burst mode support, 75 and 83 MHz base CPU clocks, full DRAM cacheability to 256Mb, synchronous and asynchronous PCI bus operation, standard AT and ATX power connectors, UDMA mode 2 support (33Mb/s), EDO and SDRAM support, a Flash EEPROM BIOS and even an integrated SVGA graphics adapter (which can be disabled with a small dip switch). A larger JPEG picture (144Kb) of the Chaintech 5SDA motherboard is available by clicking on the small JPEG at the left.
The SiS 5598 chip has a small heatsink with a spring mounting system that maintains constant pressure on the chip, improving cooling. The board seems adequately decoupled with various large electrolytic capacitors rated at 105°C. General layout for this Baby-AT form factor motherboard is good, with the Socket 7 out of the way of the floppy/HD card cage and clear silk-screen markings for all jumper and dip switch settings. Documentation is excellent, and a CD-ROM with drivers and utility programs is included.
I would say this board would be an excellent choice for a fast and stable 6x86MX-based Linux server.
The VIA VP-2 chipset is used in the PA-2007 (Baby AT format) and PA-2011 (ATX format) motherboards from FIC. The PA-2007 seems particularly interesting since it is available with up to 1Mb of pipelined burst L2 cache, and has switching voltage regulator circuitry. Top host bus speed is 75 MHz. Ian Main has reported that the board works very well with his Cyrix 6x86L 150 MHz CPU.
The VP-2 features match those of the HX chipset, but the VP-2 also supports the Cyrix 6x86 Linear Burst Mode memory access, SDRAM DIMMs and IDE DMA mode 3, and will cache up to 512Mb of RAM. We can say it combines the best features from the HX and TX chipsets, and even adds a few of its own.
VIA and FIC have gone one step further with the new PA-2012 motherboard (ATX format) based on the VP-3 chipset. The PA-2012 also sports 1Mb of L2 cache, but the new VP-3 features should provide enhanced performance: 83 MHz host bus speed, AGP and USB ports, and more.
The first report I received on this board came from Mike Brownlow, who is using it with a 6x86MX-PR200 processor. Mike has setup a very nice page showing how much he spent to build a high-quality system and what criteria he based himself on to select various components.
The most interesting feature on the PA-2007 and PA-2012 mainboards is the 1Mb L2 cache; it will not make any difference on most benchmarks, but it would be noticeable on a system running full-tilt, for example a heavily loaded Web or news server.
The Cyrix Web site has specific pages on motherboard compatibility with the 6x86 and 6x86MX processors.
Since this is not a GNU/Linux specific subject, I recommend you take a look at the Docs & Links page for sites that have a far more thorough description and testing of Socket 7 motherboards. I am also including some comments on motherboard quality in the FAQ page.
Last updated on January 8, 1998.
Copyright 1997, 1998 Andrew D. Balsa