PCI IDE controller on Mediator?

[Karlos]

mboehmer_e3b wrote:


ad 2): depends on the driver. It won't reduce CPU load significantly, as one still has to copy all data from/to the GFX card mem to Amiga side by CPU.

That sounds a performance killer. I have yet to see any Voodoo card on any mediator score more than 5MB/s for VRAM -> RAM copy. Without making the area cacheable reading from vram is painful on pretty much all cards, mediator or otherwise.

I'm sure someone we all know and love will dive in soon and point out that on the A4K, a PCI IDE controller would just be able to DMA straight into the motherboard memory.... ;-)

[Karlos]

x56h34 wrote:
He will most likely blame it on the poorly written benchmarking software. ;-) :-P

Exactly. As we all know, it is fundamentally flawed of me to use the CPU to read longwords or move16's in a hand written, icache friendly asm loop to benchmark this, when used inside a locked environment, accumulatively timed and pre calibrated using something as terribly innacurate as the EClock over a several second duration...

The card it self will do it all 10x faster over the exact same bus using Direct Magic Access