From ns.itep.ru!CERN.ch!EU.net!gatech!news.mathworks.com!newshost.marcam.com!usc!math.ohio-state.edu!news.cyberstore.ca!nntp.cs.ubc.ca!unixg.ubc.ca!duffy Sun Aug 20 21:04:11 1995
Path: ns.itep.ru!CERN.ch!EU.net!gatech!news.mathworks.com!newshost.marcam.com!usc!math.ohio-state.edu!news.cyberstore.ca!nntp.cs.ubc.ca!unixg.ubc.ca!duffy
From: duffy@theory.chem.ubc.ca (Pat Duffy)
Newsgroups: comp.os.os2.misc,comp.os.os2.setup.misc,comp.os.os2.setup.storage,comp.sys.ibm.pc.hardware.misc,comp.sys.ibm.pc.hardware.storage
Subject: SUMMARY:  PCI SCSI controllers for OS/2
Date: 16 Aug 1995 05:36:10 GMT
Organization: The University of British Columbia, Vancouver, B.C., Canada
Lines: 407
Message-ID: <40s04a$f2j@nntp.ucs.ubc.ca>
Xref: ns.itep.ru comp.os.os2.misc:130250 comp.os.os2.setup.misc:1696 comp.os.os2.setup.storage:1166 comp.sys.ibm.pc.hardware.misc:34565 comp.sys.ibm.pc.hardware.storage:40478

          Document:  PCI SCSI controllers for OS/2 list
        Maintainer:  Patrick Duffy, duffy@theory.chem.ubc.ca
Last Revision Date:  August 15, 1995
       Archived at:  ftp.netcom.com, in directory /pub/ab/abe/
         Web pages:  http://warp.eecs.berkeley.edu/os2/workbench/work.htm

This is the PCI SCSI controllers for OS/2 list. I've placed a '*' beside
the cards which I would recommend for anyone assembling an OS/2 system.
If you see '**' beside a particular card, this is an indication that I
would choose it myself when picking a PCI SCSI card for my own system.
The PCI vendor name and ID have been added in brackets below each card's
description.  The first number is HEX and the second is decimal.

As usual, please feel free to correct any continuing misconceptions I
might have, or to insist that I should recommend a component I don't, or
to add a new card/bit of information to what's here already.  Your
contributions and/or corrections are always welcome and certainly
appreciated.

General comments:

Make sure that for any SCSI device (if you have an Intel board) you have
at least the seventh or eighth revision of the Intel PCI BIOS on your
motherboard (rev. 13 is out for the 90/100 MHz boards, and fixes some
problems with rev. 8 and SCSI support, and so is preferred).  Earlier
AMI BIOS revisions would not work properly with SCSI and (apparently)
protected mode.  In addition, it turns out that some SCSI cards will
only work with some motherboards. If you're not sure what card to pick,
cross-check this list with the PCI motherboard list.  Note also that the
Intel PCI Saturn chipset (rev. 1) had problems with cache integrity, and
rev. 2 had problems with PCI to CPU burst mode, so that you're probably
really better running these motherboards without SCSI controllers, or at
the very least with all caches off.  See the PCI chipsets list for
further details on known problems with (Intel) PCI chipsets.

One other motherboard-related issue apparently is that not all PCI slots
on all motherboards allow busmastering.  As all the cards in this list
_are_ busmastering, ensure that the slot in which you put your SCSI card
will accommodate a busmaster.  Certain controllers (the Adaptecs most
prominently) behave very erratically when put in a non-busmastering
slot.

Many motherboard manufacturers simplify the choice of SCSI controller
further by including an NCR SCSI chip (for which there are OS/2 drivers)
in their boards, so there is the potential that if you buy one of the
cards listed below you could be buying a redundant SCSI interface. Check
first.  Make sure too that you get the latest drivers for the NCR chip
(available from ftp.symbios.com); these now allow DMA transfers above 16
MB where earlier drivers did not.  Note too that PCI SCSI controllers do
not include floppy controllers on-board, as most PCI motherboards have
these built in already, and fixed-address devices like floppy drives are
not a good thing to put on a PCI bus.

Also note:  OS/2 2.1/2.11 does not recognize PCI as anything different
from a (very fast) ISA bus.  It therefore is unaware of PCI's ability to
share interrupts, and so if you put two of any one controller in any
slot you're likely to have problems.  This has been fixed in Warp, which
is now apparently PCI 2.0-compliant.

One more thing:  If you have a motherboard with the NCR BIOS built in
and you get an NCR card, make sure you get a card without the NCR bios
on it, to save yourself some potential trouble that some card BIOSes
seem to have coexisting with motherboard BIOSes (which cannot be
disabled).  I'm told that the newest rev. of the Intel BIOS
(1.00.13.AX1) will also fix this problem.

Useful Numbers:
---------------
Adaptec:         (800) 959-7274 (tech. support)
                 ftp.adaptec.com (ftp site)
                 http://www.adaptec.com (web site)
                 ftp.adaptec.com (ftp site)

Advansys:        (408) 383-9400 (sales)
                 (800) 525-7440 (tech. support)
                 (408) 383-9753 (FAXback)
                 (408) 383-9612 (FAX)
                 (408) 383-9540 (BBS - N81)
                 ftp.advansys.com (FTP site)
                 http://www.advansys.com (WWW site)

BusLogic:        (408) 970-1414 (tech. support)

DPT:             (800) 322-4DPT (sales)
                 (407) 830-5522 (tech. support)
                 (407) 831-6432 (BBS)

Intel:           (800) 628-8686 (tech. support) -- for the NCR controller

NCR:             stu.nuffer@ftcollinsco.attgis.com (OS/2 tech. support)
                 ftp.symbios.com (NCR SCSI drivers)

QLogic:          (800) 867-7274

Washburn (AMI):  (800) 836-9026 / (716) 248-3627 (General inquiries)
                 (800) 836-8028 (faxback and information about specials)
                 (716) 383-6086 (tech. support)
                 (716) 381-7549 (FAX)

Manufacturer           Model         Comments
------------------------------------------------------
 Adaptec               2940*         This is Adaptec's non-RAID SCSI
                                     controller for the PCI bus.
The basic 2940 uses fast SCSI (SCSI-2, 10 MB/S), and the 2940W uses
fast/wide SCSI (SCSI-2, 20 MB/S).  There latest driver for the card
(available from the Adaptec FTP site) now supports the extra driver
switch /PCIHW, which tells the driver to bypass the PCI BIOS and query
the PCI hardware directly for configuration information.  This switch
should now enable the adapter to work on most systems.  The controller
is definitely known to work with the Intel Premiere and Premiere II
boards as well as most (if not all) of the Asus boards. The cards will
not coexist with the Quantum Empire 1080S drive when communication speed
between the drive and controller is set to 10 MB/sec (they won't
recognize it during the BIOS scan, which makes it difficult to boot...).
If you're not sure whether the 2940 will work in your setup, you can
call the Adaptec FaxBack service (408-957-7150) and request document
21105, which gives a list of PCI systems tested with the card.  (It may
be necessary to disable the green functions in the motherboard's BIOS to
get the card to work properly.)  There have been reported problems with
getting the 2940W to consistently recognize either a Fujitsu or a
Seagate 1 GB HD.  As the 2940 and 2940W are essentially similar
controllers, this problem may be endemic to both.

(Adaptec:  9004/36868)

 Under 2.1/2.11:  This card will work with the driver found on
ftp-os2.cdrom.com.

 Under Warp:  Driver support is built in, though not as robust as it
could be (from reports I've seen on the .net).  The latest driver
(mentioned above) should fix this.

Note:  Many of the problems people have reported with the card may be
due to things beyond Adaptec's control:

Some problems may be caused by BIOS bugs.  The reason for this is that
the drivers included with OS/2 Warp make use of BIOS calls to configure
the PCI BUS. These calls fail in older BIOSes which do not support this
feature properly.  The latest drivers have as an option writing directly
to the PCI hardware instead.  This option should be used (by adding the
/PCIHW switch to the basedev line for the adapter) if the driver fails
(either that, or update the BIOS, if possible).  Of course, if you can't
update the BIOS and your motherboard does not support a driver which
writes to the PCI hardware directly (some don't), you're stuck.

 Adaptec               3940          This is Adaptec's two-channel SCSI
                                     controller for the PCI bus.  Like
the 2940, it comes in a fast-only model (3940, 7 devices on each channel
for a total of 14 devices, 7 of which may be external) and a fast/wide
model (3940W, 15 devices on each channel fora total of 30 devices, 15 of
which may be external).  To put more than one channel on a card, Adaptec
employs two AIC7870 processors (one for each channel) linked by a
PCI-PCI bridge.  In order for these cards to work, therefore, your
system _MUST_ support PCI-PCI bridges, and not all do. Check with
Adaptec and/or your system motehrboard manufacturer before proceeding.
That said, I'll say that I've had no reports of success with the card
(likely because it's so new), but given that the 2940 now seems to
co-exist peacefully with most motherboards it should work well.

(Adaptec:  9004/36868)

 Adaptec               3985          This is Adaptec's RAID controller,
                                     and performs some RAID functions
automatically with more than one drive attached. In a manner similar to
the 3940, the 3985 uses multiple (three in this case) AIC7870 chips
linked by PCI-PCI bridges to support up to 21 devices on the adapter. It
also has a dedicated RAID coprocessor to handle all RAID 5 parity
calculations.  It comes with software (Netware-based, I think) to
monitor the array status and performance from the server console or any
windows-based client on the network.  The adapter supports hot swapping
of drives and hot spare drives.  All connections to the card are
standard 50-pin, and no wide model exists.  I've had no reports of
success with the card.

(Adaptec:  9004/36868)

 Advansys              ABP9XX        Advansys is a relative newcomer to
                                     the SCSI controller market, but
they make what appear to be a very nice line of SCSI controllers.  All
adapters run on the same driver (drivers are available for OS/2, NT,
DOS, SCO, Interactive, Win/95, and Netware).  All adapters are capable
of starting a local I/O request 10 microseconds after the SCSI bus is
detected as being free.  They should, therefore, be ideal for RAID
applications (especially the higher-end controllers).  The controllers
also auto-terminate, so that you need not worry about termination
problems if you have a removable external device attached.  I've had no
reports of success with these adapters and OS/2, though I suspect that
since the driver exists, they should work.

(Advanced System Products:  10CD/4301)

The controllers themselves are:

      1. ABP950 -- 2 channel SCSI adapter with 500 I/O requests stored locally
             on the adapter.  50-pin high-density external connector.
      2. ABP940 -- 1 channel SCSI adapter with 250 I/O requests stored locally
             on the adapter.  50-pin high-density external connector.
      3. ABP930 -- 1 channel SCSI adapter with 20 I/O requests stored locally
             and a 50-pin high-density external connector.
      4. ABP920 -- 1 channel SCSI adapter with 20 I/O requests stored locally
             and a 25-pin low-density external connector.

 AMD                   <various>    AMD makes a SCSI controller
                                    chip which is apparently used by
Zeos and Compaq in their computers. OS/2 drivers are apparently shipped
with the boards which use it, and it is reported to work well, though
I've had no direct reports about it yet.

(Advanced Micro Devices:  1022/4130)

 AMI                   MegaRAID      Here's a controller for the truly
                                     power-hungry.  This is a fast/wide
controller with three separate SCSI channels, each of which supports up
to 15 peripherals (so you can attach up to 45 SCSI devices to _one_
card).  It uses an Intel i960 RISC processor to control all this, and
will take up to 128 (!) MB of cache (in 72-pin SIMMs) on the controller
itself. It is PCI 2.0-compliant and supports all the regular SCSI
features, as well as RAID.  Three NCR53C720 processors (one for each
channel) are used.  The card has 50 and 68-pin internal connectors and a
120-pin external connector.  It is fault-tolerant.  I've no reports of
success with this card as yet, but driver support is there, apparently.

(American Megatrends:  101E/4126)

 BusLogic              BT946C        This is BusLogic's PCI SCSI
                                     controller.  The same SCSI
controller chip is apparently used in their VL card, and so the drivers
for that card will work with the PCI card as well.  There are apparently
five revisions of the PCI card out, the first two of which (released
before Nov. '94) were not PCI 2.0-compliant.  The newer revsions (C, D,
and E) _are_ PCI 2.0-compliant, but people have been having problems
getting revision C running properly on motherboards with Opti chipsets
(apparently the PCI controller is not being programmed properly).  To
get the older cards (A and B) running, you have to either set the card
to match the IRQ pin in your PCI slot (usually A, B, or C), or move the
card to slot A (where the card is set initially). Failure to do so will
result in hangs under Warp (but apparently not 2.1/2.11). To tell what
version of the BsuLogic Card you have, either look at the revision
number in the model (it's indicated there), or have a look at the
firmware and BIOS levels on startup.  The firmware version should be
4.22 for rev. B, and 4.23 for rev. C.  Bios version 4.86 is for rev. B,
and 4.90 is for rev. C.

(Buslogic:  104B/4171)

Deal of the week:  $180 from ASA Computers
                             2354 Calle Del Mundo
                             Santa Clara, CA 95054
                             (408) 496-6853, X201 (phone)
                             (408) 988-0359 (fax)
                             Billy@ASAComputers.com (Billy Bath, sales)

 DPT                   PM2024*       This card sounds like a less
                                     extravagent version of the PM3224
below.  The card apparently uses a Motorola 68000 chip for SCSI
operations and, like the PM3224 below, will do caching and RAID, but
apparently you must purchase the cache module and RAM separately for the
card in order to do it.  The card is reported now to work well with Warp
(drivers in the Warp box), with the caveat that the card is full-length,
and that there is no plastic guide for the 50-pin connector on it.

(Distributed Processing Technology:  1044/4164)

 DPT                   PM2124*       This is a similar card to the 2024
                                     above.  The primary difference
between this card and the 2024 is that the 2124 uses a 20 MHz Motorola
68020 for SCSI operations, in contrast to the 16 MHz 68000 used by the
2024 above.  The card has been reported to work very well with Warp and
Warp Connect (with 8 MB of cache installed).

(Distributed Processing Technology:  1044/4164)

 DPT                   SmartRAID     Now _this_ sounds like a card for
                       PM3224**      high-performance fanatics.  Driver
                                     support is included with the card.
The card is caching (RAM must be purchased separately) and, once RAM is
added, will do RAID automatically out of the box (once additional drives
are added). Quite obviously, these cards are _not_ cheap...  They are
reported, however, to work very nicely under Warp.

(Distributed Processing Technology:  1044/4164)

 Future Domain         TMC-3260SVP   This card _will_ work with the 60
                                     MHz Intel Pentium motherboard, with
one reported exception, and one note. The exception:  It hangs during
BIOS initialization during the device scan if a SCSI tape device (the
reported one was a Wangtec 5525ES) is connected.  Replacing this tape
drive with a Conner 2GB DAT drive solved the problem (switching the
older tape drive to SCSI-2 mode did not).  The note:  FD sets things up
differently than most other SCSI adapters, in that the hard disk with
the HIGHEST ID is set to drive C:, and is the boot drive.

(Future Domain:  1036/4150)

Deal of the week:  $105 from ComputAbility (800) 554-9950/(414) 357-7814

 NCR                   <Various>**   NCR makes boards and chips for OEMs
                                     (but does not sell the boards
themselves).  Boards using this chip are available from Acculogic, Asus,
and others (note that Intel is no longer selling this board).  One of
these boards is used in OS/2-certified systems of which I know. As all
of the NCR chips support busmastering, are PCI 2.0 compliant, and are
available in your choice of fast/wide SCSI combinations (the Acculogic
is made special note of here for its wide array of possible connectors),
they are both fantastic bargains and highly recommended at typically
about 25% of the price of the Adaptecs. One thing to check for when
picking your SCSI card is whether or not your motherboard has the NCR
BIOS built into the system BIOS.  The Intel Premiere and Premiere II
motherboards mentioned in the PCI motherboard list certainly do, as do
many others. If you don't have the supporting BIOS built into your
motherboard, cards are available with BIOSes on them (the card by
NexStor for example) which have BIOS support built in.

(NCR:  101A/4122)

In any case, the chips themselves are as follows:

                NCR53c810 - FAST SCSI-2, no BIOS
                NCR53c815 - FAST SCSI-2, BIOS
                NCR53c820 - FAST WIDE SCSI-2, no BIOS
                NCR53c825 - FAST WIDE SCSI-2, BIOS

Setup tips:          It may be necessary on the Intel motherboards
                     (during install only -- I haven't had this problem
                     when running OS/2 the rest of the time) is to set
                     IRQ9 to 'used' in the flash BIOS, so that the card
                     will be assigned IRQ 10 (which isn't this weird
                     cascaded thing like 9 is anyway).  Other cards
                     (like the one by Asus) have the IRQ for the card
                     set to 10 ('A') by default.  If you're installing
                     2.10 (not 2.11), you may have to rem out the
                     dpt20xx.add in your disk 1 config.sys in order to
                     get the install to proceed if you have this card.
                     It may also be necessary to set the "2nd SCSI"
                     jumper on some adapters to get them to work (even
                     when there are no other SCSI adapters in the
                     system, oddly enough).

Potential problems:  There appears to be a problem with one of either
                     the NCR OS/2 driver, the 53c810, or the Quantum
                     Prodrive 540S, as in order to get this combination
                     to work together synchronous negotiation must be
                     disabled between drive and card.  The problem does
                     not seem to appear under DOS, however.  The Quantum
                     Empire 1080S works well, however. Some NCR cards
                     come configured for edge-triggered interrupts as
                     default, which causes problems for OS2CAM.ADD.
                     Level-triggered interrupts must be used (despite
                     what the manual may claims) to make the card work
                     properly (or at all) under OS/2. The latest
                     OS2CAM.ADD was supposed to solve these problems but
                     apparently did not, and has been reported to be
                     suspected (in one person's system) to cause
                     corruption of the system files (the system in
                     question was the Asus SP3G board using the built-in
                     NCR chip).  The 4/28/94 OS2CAM.ADD is reported to
                     work well in that system, however.

Deal of the week:  $199 for a (brand-unspecified) 53c825-based fast/wide
                   SCSI controller (with 50 and 68-pin internal
                   connectors and a 50-pin external connector), from
                   JDR Micro
                   NCR 8250, $110, from
                   Insight, (206) 820-8100

 QLogic                IQ PCI**      Qlogic makes what are apparently
                                     very good fast and fast/wide SCSI
cards which have drivers for OS/2 (1.12 are the latest and are reported
to work well under Warp).  They feature a 12 MIPS processor (as opposed
to the 1 MIPS one which forms part of the NCR fast/wide processors, for
whatever that gets you) and are fully SCSI-2 compliant.  The IQ PCI
supports two SCSI channels and 1000 queued I/O operations, and the IQ
PCI/10 (fast only) supports one channel and 400 queued I/O operations.
These cards will work with the Intel 60 MHz motherboard, and one (the
fast/wide one) is in fact being used with a #9 GXE64 in a 60 MHz system
with no problems (under 2.11 or Warp).  The IQ PCI is now at revision 2.
One caveat:  The IQ PCI series of cards uses extended BIOS translation
for 1GB+ drives.  There's no way to disable this, so if you're switching
from a controller where you had extended translation disabled, you'll
have to reformat your drive.

(QLogic:  1077/4215)

Setup tips:  If you are using the controller with an Asus P54/P55
             motherboard, get the newest BIOS chip from QLogic to fix a
             problem caused by the motherboard, which does a PCI bus
             reset after the controller scans the SCSI bus, causing
             problems for the controller.  (It will be obvious if you
             have the problem -- your system will not boot.)  The
             ISP1020 firmware level should be 1.27 or greater.

Deal of the week:  $209 from Hard Drive Super Source, (800) 252-9777
                                                      (510) 494-8501

 QLogic                PCI Basic*    This card differs from the previous
                                     two QLogic cards in that it's
cheaper ($135 MSRP), and that it uses a different chip (the "400"
series).  I've had a few reports of success with this card under Warp.

(QLogic:  1077/4215)

There's what I know.  Please E-Mail suggestions/corrections and I'll
post again.
-- 
Patrick Duffy, duffy@theory.chem.ubc.ca

      -- I am Homer of Borg.  Resistance is fu-- mmmmm... donuts...

