[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Some questions
- To: Mailing List Recipients <pci-sig-request@znyx.com>
- Subject: Re: Some questions
- From: "John R Pierce" <pierce@hogranch.com>
- Date: Fri, 6 Dec 1996 18:37:25 -0800
- Resent-Date: Fri, 6 Dec 1996 18:37:25 -0800
- Resent-From: pci-sig-request@znyx.com
- Resent-Message-Id: <"MAC3b.0.FV2.8cDgo"@dart>
- Resent-Sender: pci-sig-request@znyx.com
Mark Gonzales <markg@scic.intel.com> writes...
> I don't know of any CPU-PCI bridge that will prefetch for a CPU read
> from the PCI bus. Most CPU-PCI bridges can be configured to prefetch for
> PCI bus master reads from cacheable system memory that lives on the CPU
> bus.
>
> This is why it is essential to implement bus mastering for good PCI bus
> performance. Using CPU reads and writes to do I/O to PCI will give poor
> performance for your device, and it will usually slow down any other bus
> mastering PCI devices in the system.
actually, host WRITE performance to a PCI target is quite acceptable, I've seen
on the order of 60-80MByte/sec continuously for large transfers (100's of
kbytes to several megabytes) to various targets (mostly graphics cards, thats
what I deal with). Its host READ performance thats quite limited. (I'm
talking about reasonably fast pentiums (133, 166) with triton chipsets here).
-jrp
w L ;