From: Andreas Beck <becka@rz.uni-duesseldorf.de>
  To  : ggi-develop@eskimo.com
  Date: Thu, 5 Aug 1999 21:12:16 +0200

Re: Matrox GGI accellerator

> True.  People write benchmarks which do nothing but throw accels at
> the driver as fast as possible, see that writing directly to the registers
> generates the highest benchmark scores, and fail to realize that vitually any
> real-world app (games included) is going to introduce enough latency to wipe
> out those gains.

Yes. That also leads to silly drivers that use polling instead of IRQ, as
this saves the context switch overhead. ARGL.

I have shown that to people in the past using a hybrid system that had both
IDE and SCSI HDDs. At the first glance, the IDE was faster (it was newer),
but when you ran a thorough disk-test on the drivers and simultaneously
started a calculation, it ran significantly faster with the SCSI, as the IDE
ate up 30% CPU (as it was polled), while the SCSI disk only used <5% (due to
busmaster DMA).
So overall system performance was higher with the SCSI disk (except when
running braindead OSes of course).

The very same applies to graphics.

CU, ANdy

-- 
= Andreas Beck                    |  Email :  <andreas.beck@ggi-project.org> =