Benchmarks Windows NT
Of course the motherboards were tested under the same conditions:
CPUs: Intel Pentium MMX 233 MHz (66 x 3.5), AMD K6 at 233 MHz (66 x 3.5) and if possible at 250 MHz (83 x 3) as well as Cyrix/IBM 6x86MX PR233 (187 MHz, 75 x 2.5).
Memory: If supported 64 MB SDRAM (2x 32 MB Toshiba 10ns or Samsung 12ns), else 64 MB EDO DRAM (2x 32 MB TI, 60ns)
Harddisk: Quantum Fireball ST 3.2 EIDE (if supported using Ultra-DMA/2 interface, else DMA/2)
VGA: Matrox MGA Millennium 4 MB, 1024x768 High Colour
OS: MS Windows 95 OSR 2.1 (USB supplement), MS Windows NT 4.0 Server (+ Service Pack 3)
Benchmarks: Business Winstone 97
I did not try mixing EDO and SDRAMs. It should work on many boards, but I don't recommend you to do it. You won't get more performance if you add some new SDRAM to your EDOs. If you want to use 75 or 83 MHz bus frequency, change over to the faster SDRAM, but keep your EDO memory in case you're going to stay at 66 MHz. The performance advantage at this bus speed is not worth talking about it. Another issue is memory expansion, which has become quite elegant since DIMM modules can be installed seperately; you don't need to obtain an identical pair and you won't get trouble thanks to different contacts or even chip manufacturer. In some boards I tried the mixture of Samsung, Toshiba, Hitachi and LGS SDRAM memory, without any problems.
The Quantum Fireball ST obviously is faster than the Fireball TM used in the former tests. But the performance difference between EIDE hard drives is only visible under Windows NT. Windows 95 does not profit that much from a faster hard drive. This is also a reason why I did without the well known performance champ Seagate Cheetah.
You might ask why I did not use the faster Millennium II. This video card is certainly faster, but not fast enough to clearly push up the Winstone results. The 2D improvements are fine, but 3D quality is, I'm sorry to say, a huge disappointment. For the same money you can get one of the new generation 3D accelerators like Diamond Viper 330 or nVIDIA RIVA128, you can get some information about the new cards at the graphics card review .
Some boards do not always appear in the charts; the reason usually is instability or the missing support. Only if the particular configuration was able to complete the benchmark did I quote the result. The performance relation between the boards does not necessarily keep the hierarchy from the 66 MHz charts (K6 and Intel MMX); some boards have to be run at slower memory timings to work stable, so a board being average at 66 MHz could be very fast at 75 or 83 MHz thanks to the optimizations of its manufacturer.