From owner-freebsd-hackers Tue Apr 4 16:37:44 1995 Return-Path: hackers-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id QAA26456 for hackers-outgoing; Tue, 4 Apr 1995 16:37:44 -0700 Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241]) by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id QAA26438 for ; Tue, 4 Apr 1995 16:37:34 -0700 Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.8/8.6.6) id PAA08721; Tue, 4 Apr 1995 15:35:21 -0700 From: "Rodney W. Grimes" Message-Id: <199504042235.PAA08721@gndrsh.aac.dev.com> Subject: Re: PCI/EISA/ISA performance To: terry@cs.weber.edu (Terry Lambert) Date: Tue, 4 Apr 1995 15:35:21 -0700 (PDT) Cc: matt@lkg.dec.com, vernick@cs.sunysb.edu, freebsd-hackers@FreeBSD.org In-Reply-To: <9504042255.AA20914@cs.weber.edu> from "Terry Lambert" at Apr 4, 95 04:55:48 pm X-Mailer: ELM [version 2.4 PL23] Content-Type: text Content-Length: 2575 Sender: hackers-owner@FreeBSD.org Precedence: bulk > > > > Anyone have any performance results comparing the three types of > > > buses? I'm interested in disk I/O performance and/or network > > > I/O performance using the three buses and DMA. > > > > As a rule of thumb, PCI > EISA > ISA. > > Whose thumb are you looking at? 8-). Yours has a few facts wrong :-(. Replace speed limit with clock frequency in all of the following. > VLB has a speed limit of 40MHz (typical cards die > 33MHz, though) > PCI has a speed limit of 33MHz > ISA has a speed limit of 12MHz ISA does not have a specified clock frequency, I have seen it running as fast as 16Mhz. Most boards die above 10Mhz, but some of the more specialized industrial applications boards are spec'd upto 12 or 16Mhz. > EISA has a speed limit of bus clock (mine runs at 50MHz) The EISA spec says that the BCLK signal shall be 8Mhz +/- 5%. Your CPU may very well be running at 50Mhz, but I'll beat you $100 your EISA BCLK is running at 50Mhz/6 or 8.3333Mhz (just inside the upper limit of 8.4Mhz). > PCI has a width limit of 64bits PCI has a standard bus width of 32 bits, with 64 bits being an extension. The only 64Bit PCI connectors I have ever seen where on prototype systems inside of Intel. There are no production 64 bit PCI motherboards or add in cards that I have been able to find at this time. So in effect PCI is still a 32 bit bus, forget all the sales hype! > EISA has a width limit of 32bits > VLB has a width limit of 32bits > ISA has a width limit of 16bits (AT) or 8bits (XT) > > EISA and VLB fall back to 16/8 (ISA) depending on the card The card is either in ISA emulation mode, or using ISA I/O port addresses. > So which is faster depends on your relative bus clock rate (it has > to be pretty high to beat PCI, however). Not really true. You'll never crank an ISA bus upto anywhere near the speed of any of the others. Hard top speed limits are more like: ISA: 5MB/sec EISA: 33MB/sec VLB: 132MB/sec PCI: 132MB/sec > For instance, a DX4/75 PCI has the same bus transfer rate as a DX/50 EISA. This is false, the PCI is most likely running at 25 Mhz, and the EISA at 8.33 Mhz. Due to the number of clocks cylces required to transfer data the max bux speeds would be 100MB/sec for PCI and 33MB/sec for EISA. > Actually, MCA looks pretty good, compartively. 8-). And it would have probably done very well had IBM not required all the stupid licenseing to use it. -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Custom computers for FreeBSD