I can't tell about other buses, but the FDC and serial ports run at fixed speeds. Because VLB can run at different speeds, it makes more sense to have an independent clock to ensure proper speeds.

33MHz is the bus speed, but remember -- IDE / ATA is actually based on the AT ISA bus. This is something I'm just now getting into thoroughly (ISA Systems Architecture book arrived last week), but to my understanding, there's a buffer in the controller chipset that reconciles the difference between the particular ATA revision's speed and that of the bus. So the throughput can be up TO 133MB/s, but not until... what was it, ATA-7? Which came about a little while after the whole VLB thing came and went.

Edit: On my particular VLB controller, the 24MHz crystal is connected to the chip next to the floppy connector. I don't see any other crystals, so it's either redistributed (possibly at multiple different divisors for legacy I/O and disk I/O) or the IDE controller chip gets its clock from the bus connector. There seems to be a pretty distinct division on the board between the ISA side (floppy, serial, parallel, game) and the VLB side (IDE), but there are still a couple traces between them.

SirNickity wrote:

...So the throughput can be up TO 133MB/s, but not until... what was it, ATA-7? Which came about a little while after the whole VLB thing came and went.

--If I were using an IDE to CF adapter (Rated up to 133MB/s) and a 1066x CF card (Rated up to 160MB/s), what speeds could I reasonably expect to see here? Would this essentially be an Ultra ATA/133 set-up (ATA-7) @133MB/s, or is there another bottleneck I'm missing?

--Or...Are we saying that because the controller card itself IS NOT ATA-7, the controller card would be the bottleneck?

--Additionally, I recently ran across a VLB controller card that had the two typical IDE connectors, 1 & 2. One of them was labeled "Vesa IDE" and the other "ISA IDE". I wonder if the ISA IDE is subjected to the 24.00mhz crystal speed or if it still uses the 33mhz external bus....But 8-16 bits wide instead of 32. Unsure, just thinking out-loud.

1) 24MHz is for creating baud rates for the floppy controller and UARTs.
2) Most VLB transactions take 2 clocks to do except when bursting is done, similar to PCI. Maximum is 134MB/sec but reality is half and less due to various overhead.
3) Crystal has no effect here. HDD controllers are all just normal PIO stuff with relatively strict timing settings, taking many cycles per transaction. That's why you cannot get more than few MB/sec with them even on VLB. Floppy and other parts are sitting on ISA bus and not the VLB end and will be slow no matter what.
5) Bus mastering stuff such as SCSI controllers have the best performance.

The fastest VLB IDE cards I have seen top out at 16MB/s. They actually have jumpers for manually setting the cycle time (automatic setting is also possible) for communicating with the drive. 120ns is the fastest setting available (8.3MHz), and the interface is 16 bits wide, so 16MB/s in theory.

I haven't tried connecting CF cards or newer HDDs to see how high of a transfer rate I could get. Currently I have a 2.5GB Quantum Fireball connected which benches at 6MB/s sequential, it's pretty snappy for a 486.

I can't wait 'til I get to the chapter on IDE. *nerd joy*

bakemono wrote:

The fastest VLB IDE cards I have seen top out at 16MB/s. They actually have jumpers for manually setting the cycle time (automatic setting is also possible) for communicating with the drive. 120ns is the fastest setting available (8.3MHz), and the interface is 16 bits wide, so 16MB/s in theory.

I haven't tried connecting CF cards or newer HDDs to see how high of a transfer rate I could get. Currently I have a 2.5GB Quantum Fireball connected which benches at 6MB/s sequential, it's pretty snappy for a 486.

which one? practically I never measured more than 3.3mbyte/s (very late model, with pio4, and timing settings, 33mhz). the more common models topped out at 2mbyte/s.
i benchmarked with core.