Discussion
We,ve got the results - what we gonna do with'em ?
First of all, don't forget that all speeds except the L1 one depend significantly on the bus clock. A system with a bus clock of 60 MHz always has a slower memory speed than a system with 66 MHz bus clock. Therefore a P180 has a slower memory speed than even a P100. It'll be very interesting to see, how systems with 75 MHz bus clock will perform, if they ever should turn up.
- L1 Cache Speed
As already said - this speed only tells us how fast the CPU speaks to its own chip cache, so it only depends on the clock speed of the CPU. Nice number, but it doesn't tell us a lot at all. - L2 Cache Speed
This one is important. It tells us how fast the CPU is able to transfer data between itself and the L2 Cache. Highly dependent on Cache Type used - PB SRAM is favorite here. Also shows quality of M/B design, the faster, the better the M/B design - but don't get fooled by small differences here ! More or less all boards are able to reach the physical limitation of 66.6 MB/s at 66 MHz clock speed - due to the 3-1-1-1 cache read cycle of all the chipsets using PB Cache. If you've got more it's either due to measuring tolerance of ctcm or you might have a slightly higher clock speed than 66 MHz. Only 75 MHz bus speed boards can top that value. - Memory Speed
The most interesting one ! This value depends on three things: - the memory type used:
the faster the memory, the higher the value. EDO performs quite a bit better than FPM , and I'm waiting impatiently on results for BEDO and SDRAM , which should be even faster than EDO.the Chipset :
One of the main differences between the 430FX and 430HX PCIset is the ability of the HX to access EDO RAM faster than the FX, the EDO timing is much better (max 5-2-2-2 for EDO in HX, max 7-2-2-2 for EDO in FX). There HAS TO BE a significant difference between the speed for FX and HX. It also would be very interesting to know how the new other chipsets (VIA Apollo Master, VLSI Lynx,....) perform here.the RAM timing:
The less cycles it takes to write to DRAM, the faster the memory performance. You are normally able to adjust this in your BIOS setup - as long as you don't run auto config. The values are RAM read, RAM write, Leadoff cycle, RAS to CAS delay. Keep them small to gain maximum performance but be aware of system crashes if your RAM isn't up to it - this is the reason why I'm using 45 ns EDO.
Having said that, it's up to the M/B manufacturer to make the most of it - so not every HX board is that much faster here than a FX board ! You should expect up to almost 40 MB/s for a FX at 66 MHz (my own I got tuned up to 39.3 !) and up to almost 48 MB/s for a good HX at 66 MHz. Have a look at the motherboard comparison table to see which boards seem to reach that. The reason why the HX chipset has to be faster is the different memory timing. The fastest read timing (with EDO) of the Triton FX chipset is 7-2-2-2, the fastest of the Triton II HX chipset is 5-2-2-2. Much faster even should be BEDO and SDRAM. Their memory timing in the Triton VX or the VIA Apollo VP is 5-1-1-1, so we should expect a speed of up to 57 MB/s ! - DOS Simulation Speed
This one tests the speed in the good old 640 kB block of main memory. Because this is very small, it's mostly dependent on your L2 cache size and speed. Systems with 512 kB L2 cache are significantly faster than systems with only 256 kB, cause 640 kB fit almost completely in the 512 kB cache. - Win Simulation Speed
It's tested using 4 MB memory to run the benchmark in, so it's very dependent on your main memory speed. The cache plays only a small role here, for 4 MB are significantly bigger than 256 or 512 kB.
