Subject: Re: 3c509 again
To: Herb Peyerl <Herb.Peyerl@sidney.novatel.ca>
From: Drew Hess <dhess@CS.Stanford.EDU>
Date: 06/30/1994 13:04:52
Herb Peyerl writes:
> : I agree with the other poster that VLB systems with more than 2 slots at
> : 33MHz should be avoided. I would tend to blame your problems on the fact
> : that your VLB board is outside VLB 1.0 specs before condemning VLB as a whole.
> : I have one of the earliest VLB designs, a Micronics JX-30 (basically a
> : Gateway machine), with an ATI GUP VLB and a Buslogic BT445S, and only 2 VLB
> : slots; it has worked flawlessly in NetBSD, Linux, NT, Windows/DOS, and
> : OS/2 2.1.
> I'm sorry. I'm not so sure if I buy this. ie: If the thing had N vlb slots
> but I was only using 2 of them; does it really matter? ie: only 2 slots
> were actually populated at the time.
Yes, of course it does. Each physical connector is a capacitive load on the
bus; and since VLB 1.0 is basically being driven by the i486 address and
data bus drivers, at 33MHz 2 physical connectors is a safe limit. Any
number of connectors beyond this point is asking for trouble (the i486 simply
can't drive enough current). In fact, you'll notice that the VLB 1.0 spec
does not include support for 486DX50s because the 50MHz part simply couldn't
drive *any* loads other than the standard external cache, memory controller,
VLB 2.0 corrects these limitations by properly isolating the processor's
memory bus from the "local bus."
> I dunno; I'd prefer to stick with stuff that seems to have been engineered
> with some forethought. Just MHO.
I agree with you that VLB 1.0 is a hack; and some boards (notably the
Ultrastor 34F) are notorious for having problems with certain VLB
implementations. However, it's a hack that works well if the designers are
careful to stay within the spec's guidelines.
If you can afford it you should go with PCI or EISA; if you're looking for
cheap video performance, a properly implemented VLB machine is a fine
solution. (I won't get into the VLB/SCSI argument because nobody seems
to agree on whether or not it's a "good thing" for system performance; I'm
not sure how I feel about it because I haven't run enough comparisons with
various OSes/mboards/controllers/benchmarks. I *will* say that VLB SCSI
boards seem to perform better for disk benchmarks than ISA SCSI boards;
whether or not they make a difference in system performance (or even hurt
it) is what I'm not sure about.)