What's wrong with Hint EISA implementation
Michael Condict
been trashed in this forum as "not real EISA" and not worth paying
for. (It costs about $100 more than an equivalent ISA board.) Well
I just looked at the official HINT specifications, and I think these
opinions are way off base. My motivation was that I have a NICE
Mini-EISA motherboard and am interested in possibly getting an EISA
SCSI controller. I wanted to know if my SCSI I/O would be crippled
in some important way by the limitations of the HINT EISA, which is
a subset of the full EISA spec. (They call it P-EISA, for Pragmatic
EISA.)
Here's what I found out:
The chipset supports all features of the official EISA spec
except for one: in its least expensive configuration, i.e.
with no help from outside chips, it cannot do DMA to physical
memory addresses that have bits 28,29 or 30 turned on. That's
it -- the only restriction.
In particular, it can do full 32-bit, burst-mode transfers, allowing
up to 33 MB / sec. I/O speeds.
So the restriction means that you can't do DMA to physical memory
above 256 MB. Wow, this is really crippling, isn't it?
Does anyone have any reason why I should be disatisfied with my
EISA board now?
Michael
Ralph Valentino
>The HINT chipset used in some motherboards (such as NICE Mini-EISA) has
>been trashed in this forum as "not real EISA" and not worth paying
>for. (It costs about $100 more than an equivalent ISA board.)
I agree with the "not real EISA" statement, but it is cheaper than
EISA so you're the final judge in determining if it was worth it.
>Well I just looked at the official HINT specifications,
What exactly is this?
>and I think these opinions are way off base. My motivation was that I
>have a NICE Mini-EISA motherboard and am interested in possibly
>getting an EISA SCSI controller. I wanted to know if my SCSI I/O
>would be crippled in some important way by the limitations of the HINT
>EISA, which is a subset of the full EISA spec. (They call it P-EISA,
>for Pragmatic EISA.)
Yes, it will be crippled, never mind cause you some headaches.
>Here's what I found out:
>The chipset supports all features of the official EISA spec
>except for one: in its least expensive configuration, i.e.
>with no help from outside chips, it cannot do DMA to physical
>memory addresses that have bits 28,29 or 30 turned on. That's
>it -- the only restriction.
Where did you get this info? From a Hint motherboard manual (G486HVL):
HiNT Caesar Chipset does not support the following functions:
Non-ISA compatible DMA
Non-ISA compatible Interrupts
Non-ISA compatible Timers
[list of EISA cards that won't work with it deleted]
and my favorite quote from the section entitled "Further Information":
In no way can EISA deviate for[sic] the ISA standard and
therefore HINT decided to abandon Non-ISA compatible redundant
features defined by EISA.
Translation: ISA cards will work in our P-EISA slots just as well as
ISA cards work in real EISA slots. <sigh> :-)
It is then nice enough to explain to you how you can turn off all of
those "redundant" features on the Adaptec 1742A SCSI controller to get
it to work, though.
>Does anyone have any reason why I should be disatisfied with my
>EISA board now?
As I said, you're the final judge. No one can tell you if you should
be satisfied or not. But if you're asking if I would be satisfied,
then I'd have to say definitely not.
-Ralph
===============
Ralph Valentino (ra...@chpc.org) (ra...@wpi.wpi.edu)
Hardware Engineer, Worcester Polytechnic Institute
Center for High Performance Computing, Marlborough MA
gary korenek
> been trashed in this forum as "not real EISA" and not worth paying
> for. (It costs about $100 more than an equivalent ISA board.)
>
Back in April ('93) when I went into a local shop, asked for a price
on a EISA machine, and got 'converted' to a system including a HINT
based EISA/ISA/VL motherboard.
After I had the machine for several weeks, I read in another newsgroup
that the HINT chipset has 24-bit (ISA) DMA, rather than 32-bit (EISA)
DMA. So I called the MCCI/NICE company (800-697-0660), and the HAWK
company (408-436-8999). They both make VL/EISA/ISA m-boards based on
the HINT chipset. They both verified 24-bit DMA (with a 16 mb ram
addressability limit of ISA).
Over time, when e-news posts have come up asking about this board, I
have recommended passing this m-board by, in favor of any EISA m-boards
with 32-bit DMA that also supports VL. I have also said that despite
24-bit DMA, the HINT based m-board works. I have run MS-DOS 5.0/6.0,
MS-Windows 3.1, and ISC UNIX SVR3.2 quite happily with my HINT based
m-board.
The last new item of information I received (from a fellow Use-netter):
When a bus-mastering EISA card is used with the HINT based m-board,
since DMA on the busmaster card is 32-bit, then you regain standard
EISA RAM addressability for that card. So for example, if you use an
Adaptec 1742a EISA SCSI host adapter (in its enhanced mode), there will
not be a 16 mb ram addressability limitation because the 1742a uses its
onboard 32-bit DMA (rather than the 24-bit DMA on the HINT m-board).
I am happy that the m-board works. I am not happy that what was adver-
tised as 'EISA' is EISA with 2 exceptions (in my mind EISA is EISA with
no exceptions). Exception #1 is the 'watchdog timer' EISA function is
not implemeted. For this I do not care at all. It is used (as I recall)
to make sure a busmaster EISA card does not take the bus away from the
CPU forever. Exception #2 is the 24-bit DMA. The DMA ram addressability
limitation is darn hard to determine by the casual (or even advanced) user
looking in the -board manual.
The HINT chipset based VL/EISA/ISA m-boards are unusual compared to
other m-boards. I wish that HINT would make it more obvious where
it deviates from EISA. I wish the buyer would be informed of these
deviations *before* the purchasing dollars are spent. If I had known
then what I think I know now, I would not have bought the HINT based
m-board.
Regards,
--
Gary Korenek (kor...@nmti.com)
Network Management Technology Incorporated
Sugar Land, Texas (713) 274-5357
SHASHANK KUMAR
> Where did you get this info? From a Hint motherboard manual (G486HVL):
> HiNT Caesar Chipset does not support the following functions:
> Non-ISA compatible DMA
> Non-ISA compatible Interrupts
> Non-ISA compatible Timers
> [list of EISA cards that won't work with it deleted]
> and my favorite quote from the section entitled "Further Information":
> In no way can EISA deviate for[sic] the ISA standard and
> therefore HINT decided to abandon Non-ISA compatible redundant
> features defined by EISA.
May I ask you as to where you found the above statements in the G486HVL
manual. The very 1st line of Chapter 1 INTRODUCTION (p. 1) says,
With the Ceasar chipset, the EISA G486HVL supports all main EISA
features such as down-shift masters and 32-bit Master DMA capable
of bursting at 33MB/sec transfer rate, EISA specification
configuration, refresh, extended NMI status and controls, EISA
bus master and command timeout, software NMI ...
On next page under 1 - 2 SPECIFICATION it says,
* 10 expansion slots 3 * 32-bit EISA Bus slot for EISA Bus master/
slave add-on cards, .....
Though I must say that I was informed by the mb. manufacturer JOIN DATA that
Hint Chipset has 24-bit DMA, (as usual the Comtrade Tech Support gave the
wrong information of it being 32-bit DMA).
> <id.6ZU...@nmti.com> ........
> The last new item of information I received (from a fellow Use-netter):
> When a bus-mastering EISA card is used with the HINT based m-board,
> since DMA on the busmaster card is 32-bit, then you regain standard
> EISA RAM addressability for that card. So for example, if you use an
> Adaptec 1742a EISA SCSI host adapter (in its enhanced mode), there will
> not be a 16 mb ram addressability limitation because the 1742a uses its
> onboard 32-bit DMA (rather than the 24-bit DMA on the HINT m-board).
Does this mean that mine MORSE IDE Blazer (VLB Disk Controller) which is
supposed to be a true 32-bit IDE disk accelerator, has 32-bit DMA? I don't
seem to facing any problem with 20MB RAM, though I can't tell if full
20MB is being used or only 16MB is being used ???
Also in my AMI BIOS Chipset up I have added 1 clock cycle delay to run my
VLB cards (MORSE disk controller and Hercules VLB Graphics Accelerator) to
make them run at 50MHz, will make them faster or slower ???
Thanks in advance.
Cheers!!!
-Kumar
Ralph Valentino
>[Re: HiNT Ceasar chipset and G486HVL limitations]
>May I ask you as to where you found the above statements in the G486HVL
>manual. The very 1st line of Chapter 1 INTRODUCTION (p. 1) says,
>
> With the Ceasar chipset, the EISA G486HVL supports all main EISA
> features such as down-shift masters and 32-bit Master DMA capable
> of bursting at 33MB/sec transfer rate, EISA specification
> configuration, refresh, extended NMI status and controls, EISA
> bus master and command timeout, software NMI ...
>
>On next page under 1 - 2 SPECIFICATION it says,
>
> * 10 expansion slots 3 * 32-bit EISA Bus slot for EISA Bus master/
> slave add-on cards, .....
Same with the manual I have - this is the marketing mumbo-jumbo. I
bring to your attention that there are not really 10 slots, there are
only 8 (look at the picture). They are counting the 2 VLB slots as
both ISA and VLB. I _suppose_ this is 10 slots if you find a VLB card
that doesn't use (or block) the ISA connectors. Pretty slimly if you
ask me.
Also note the words "all main EISA features" in the intro. This quite
nicely implies that it doesn't support some non-main EISA features.
Anyhow, the EISA compatibility issues are in Chapter 4 "Technical
Information" sections 4-3 "Controllers Not Supported" and 4-4 "Further
Information". Do you have these chapters? (the documentation id on
the last page of the table of contents reads: G486HVL-V2 1992-12)
I don't suppose you have a more recent copy in which they decided this
information wasn't necessary for you to know...
>Though I must say that I was informed by the mb. manufacturer JOIN DATA that
>Hint Chipset has 24-bit DMA, (as usual the Comtrade Tech Support gave the
>wrong information of it being 32-bit DMA).
From the documentation, I can not determine which is the actual case.
My only guess is that the 32-bit DMA that it can do is 32-bit data,
not 32-bit address. (They quite cleverly leave out the words
"address" and "data" when ever referring to 32-bit DMA).
>> <id.6ZU...@nmti.com> ........
>> The last new item of information I received (from a fellow Use-netter):
>> When a bus-mastering EISA card is used with the HINT based m-board,
>> since DMA on the busmaster card is 32-bit, then you regain standard
>> EISA RAM addressability for that card.
I don't know how this can be true. What type of DMA is only limited
to 24-bits then? The motherboard's DMA controller??? Who uses that
thing any more? (besides the occasional floppy drive). If that's the
case, then heck, doesn't sound like a big loss to me.
Brian Hook
supposed to be a true 32-bit IDE disk accelerator, has 32-bit DMA? I don't
seem to facing any problem with 20MB RAM, though I can't tell if full
20MB is being used or only 16MB is being used ???
IDE doesn't use DMA, so it doesn't have a problem with 24-bit vs. 32-bit
DMA, unless you have some oddbal IDE adapter.
Brian
--
//----------------------------------------------------------------------------
// Brian W.K. Hook "Stop! Stop in the name of all that
// ( b...@beach.cis.ufl.edu ) which does not suck!" - Butthead
// ( 72144,3662 CI$ )
//
// "Oooooooo-la-tec! Ooooooo-la-tec!"
//
// finger b...@beach.cis.ufl.edu for more info on my 3d graphic project
//----------------------------------------------------------------------------
Michael Condict
>In article <24tk89$o...@paperboy.osf.org> con...@next19.osf.org (Michael Condict) writes:
>>The HINT chipset used in some motherboards (such as NICE Mini-EISA) has
>>been trashed in this forum as "not real EISA" and not worth paying
>>for. (It costs about $100 more than an equivalent ISA board.)
>
>I agree with the "not real EISA" statement, but it is cheaper than
>EISA so you're the final judge in determining if it was worth it.
>
>>Well I just looked at the official HINT specifications,
>
>What exactly is this?
The reference documentation from HINT Corp. titled "Caesar Chipset
Specification", Rev. 1.01, Jan. 5, 1992. It describes the pinouts and
functionality of the chips numbered CS8001 and CS8002. It was obviously
written for motherboard designers and is not promotional literature.
First let me point out that the last quote above does not imply that
the HINT EISA is crippled in any important way (note the use of the
word "redundant"). For example, it does not imply that the chips cannot
do 32-bit DMA, since that is not a redundant feature of EISA, compared
to ISA.
But we still have a major inconsistency here, possibly caused by my
misreading of the documentation on my first pass. Let me quote from the
Specification document mentioned above:
"With the Caeser Chipset, an EISA motherboard can be manufactured
at $150 cost. There are less than 20 most common TTL discrete logic.
Such a system supports all main EISA features such as down-shift masters
and 32bit Master DMA capable of bursting at 33MB/s transfer rate. EISA
specification Configuration, Refresh, Extended NMI status and controls,
EISA bus master and command timeout, software NMI generation, Local Bus
Devices can also be supported."
So why do you have to turn off features of the Adaptec 1742A? Did they
say which features don't work? Certainly, 32-bit bus mastered DMA should
work, according to the quote above. Is it possible that the HINT
motherboard does not allow the full functionality of the HINT Caesar
Chipset?
The only possible explanation I see for this cognitive dissonance is
in another section of the Specification, which says:
"With the Caesar Chipset, different system configuration can
be achieved:
1. P-EISA: a low cost version of EISA providing performance of
EISA at close to ISA price.
2. Super-ISA: an ISA cost board providing EISA features and
performance.
3. Intel EISA: The CS8001 can be used with the Intel 357 + 358"
It looks like the HINT motherboard you are referring to is operating
in the Super-ISA mode, which probably cannot do more than 16-bit
data transfers. I deduce this limitation from part of the description
of the CS8002 chip, which says:
SD0-15 [data bits 0 through 15] is supported through the chip
while SD16-31 is supported through external buffers, with all
the controls provided by the CS8002.
I take this to mean that in the Super-ISA mode (which operates with
standard AT BIOS, by the way), the motherboard manufacturer fails to
provide the external support for SD16-31, saving some money, but
limiting the system to 16-bit transfers. Apparently, HINT did this
with the motherboard you are referring to.
My claim was about the P-EISA mode, in which the board provides the
external buffers to support 32-bit data transfers. I think the only
limitation of this mode is that it can't do DMA to addresses with bits
28,29,30 on. This is an unimportant restriction on today's systems.
My mistake was that I thought at the time that this was the cheapest
configuration of a HINT-based motherboard. I now see that it can be
operated in a crippled mode called Super-ISA. I agree that when used in
this manner it is of questionable value. Does anyone know whether the
NICE Mini-EISA board operates in Super-ISA mode or P-EISA mode?
Michael Condict
OSF Research Inst
Cambridge, MA 02173
Michael Condict
>> HiNT Caesar Chipset does not support the following functions:
>> Non-ISA compatible DMA
>> Non-ISA compatible Interrupts
>> Non-ISA compatible Timers
>> [list of EISA cards that won't work with it deleted]
>> and my favorite quote from the section entitled "Further Information":
>> In no way can EISA deviate for[sic] the ISA standard and
>> therefore HINT decided to abandon Non-ISA compatible redundant
>> features defined by EISA.
>
>
>May I ask you as to where you found the above statements in the G486HVL
>manual. The very 1st line of Chapter 1 INTRODUCTION (p. 1) says,
>
> With the Ceasar chipset, the EISA G486HVL supports all main EISA
> features such as down-shift masters and 32-bit Master DMA capable
> of bursting at 33MB/sec transfer rate, EISA specification
> configuration, refresh, extended NMI status and controls, EISA
> bus master and command timeout, software NMI ...
>
>On next page under 1 - 2 SPECIFICATION it says,
>
> * 10 expansion slots 3 * 32-bit EISA Bus slot for EISA Bus master/
> slave add-on cards, .....
>
>Though I must say that I was informed by the mb. manufacturer JOIN DATA that
>Hint Chipset has 24-bit DMA, (as usual the Comtrade Tech Support gave the
>wrong information of it being 32-bit DMA).
This is very confusing you are saying that the manual claims 32-bit DMA
but the mb manufacturer told you 24-bit DMA. Which is true? I've double
checked the HINT Specification document and I cannot figure out any
way that it can be operated such that only 16-bits of address are used.
As I said in my first mailing, it does not allow address bits 28,29,and 30
to be on, but that limits you to 256 MB, not 16 MB. Is this a piece of
misinformation being promulagated by mb manufacturers?
And what about the limitation of 16-bit transfers? It certainly seems that
the chips can be configured in that mode (called Super-ISA). Why doesn't
any mb manufacturer mention that?
Curious,
Michael Condict
OSF Research Inst.
Cambridge, MA
SHASHANK KUMAR
> both ISA and VLB. I _suppose_ this is 10 slots if you find a VLB card
> that doesn't use (or block) the ISA connectors. Pretty slimly if you
> ask me.
Sure. I saw the motherboard of my friends PC, its VLB seems to be pretty
different from others I have seen. It actually had two ISA slots in a
single line to make a 32-bit slot. Also he had a Tseng ET2000... Graphics
Card which had 32 _notch_.
Is this 32-bit slot and card different from VLB cards that I have which
have _notch_ randomly present, I don't think they total to even 32. Is
it a better card in performance based on its architecture???
> Also note the words "all main EISA features" in the intro. This quite
> nicely implies that it doesn't support some non-main EISA features.
What am I missing by not having 'non-mail EISA features'? I suppose
24-bit addressing is enough for upto 256MB memory (too much for PC).
32-bit data transfer should be good enough.
> I don't suppose you have a more recent copy in which they decided this
You are right, its a recent one as G486HVL-3-V1-1993-06 w/o any 4.3 or 4.4.
Though right now I am not using any EISA based card but in future I plan
to get SCSI-2 card and SCSI drive, and both VLB slots are filled with
MORSE IDE VLB card and VLB Video card.
I remember someone pointing a VLB/EISA/ISA mb. in price range of $300
w/o CPU is expensive one, do we any other with full specification in this
price range.
Cheers!!!
-Kumar
Brian Hook
> What am I missing by not having 'non-mail EISA features'? I suppose
> 24-bit addressing is enough for upto 256MB memory (too much for PC).
> 32-bit data transfer should be good enough.
You're missing full EISA compatibility. Does it support level based
interrupts? And 24-bit addressing vies you only up to 16MB of addressing,
which is rather limiting on a "Real" system.
Later,
Hook
Michael Condict
on the Hint Caesar Chipset), and further perusal of the Hint Specification
document, we now have the full story, which is:
The Hint Caesar ChipSet can be used to build one of three
different motherboards. From best to worst they are:
Intel EISA: By adding the two Intel chips that implement
full EISA, you end up with an unrestricted EISA board.
This configuration uses only one of the two Hint chips
(the CS8001 and not the CS8002). This is *not* the
configuration of the NICE MINI-EISA, nor probably the
HINT motherboards.
P-EISA: By adding a small amount of supporting TTL chips,
an 80C206 timer, external 16-bit data buffer, etc, to the
two HINT chips, you get a restricted EISA board. But
the only restriction is in the DMA controller that is
built into the HINT CS8002 chip. It is basically an
ISA DMA controller, limited to 16-bit data transfers
and 24 bits of address (16Mbyte addressing limitation).
But this is unimportant to your bus-mastering EISA
SCSI controller, such as Adaptek 1742, because it
has its own on-board DMA controller. The HINT
chipset does not restrict the full 32-bit DMA
capabilities of bus-mastering DMA devices attached
to the EISA bus.
Super-ISA: The details of this one are unclear, but it
appears to mean that the board maker does not supply
the external 16-bit data buffer necessary to support
the top half of a 32-bit data transfer, resulting in
a brain-damaged EISA that can only do 16-bit transfers.
In this configuration, I don't even think EISA cards
would work, unless (like the Adaptek 1742) they have
an ISA-compatibility mode.
NICE has assured us that the NICE Mini-EISA uses the P-EISA con-
figuration. So basically, if the only EISA performance you
care about is for I/O through your EISA SCSI controller, any
bus mastering DMA SCSI controller, such as Adaptek, Bustek,
DTC, Ultrastor will work fine and will be as fast as on any
EISA bus. However, if you've got a slave IDE controller on
your EISA bus (would anyone actually do this?), DMA to it will
be crippled, because it will use the HINT CS8002 DMA controller,
which has only the functionality of ISA DMA. I don't know
about floppy controllers on the EISA bus. If your floppy
controller is built into your EISA SCSI card (e.g. Bustek 742),
it probably can use the same full-featured 32-bit DMA that is
used for I/O to/from the SCSI devices. But I'm not sure about
that.
One last note. NICE says that you need to use different ".CFG"
files when you configure your Adaptek or Bustek controller.
This is because of some minor incompatibilities of the HINT
EISA, perhaps the previously quoted thing about not supporting
watchdog timers? In any case, the incompatibility has nothing
to do with not supporting 32-bit data / 32-bit address DMA.
The bottom line is that for the price, I am completely satisfied
with my NICE Mini-EISA board, and will probably buy an Adaptek
compatible EISA SCSI controller for it so I can run various
versions of UNIX.
Does this close the issue?
Ralph Valentino
>Okay, after a long talk with NICE (who makes the MINI-EISA board based
>on the Hint Caesar Chipset), and further perusal of the Hint Specification
>document, we now have the full story, which is:
>
>NICE has assured us that the NICE Mini-EISA uses the P-EISA con-
>figuration. So basically, if the only EISA performance you
>care about is for I/O through your EISA SCSI controller, any
>bus mastering DMA SCSI controller, such as Adaptek, Bustek,
>DTC, Ultrastor will work fine and will be as fast as on any
>EISA bus.
Doesn't this mean that the EISA bus mastering devices have to use ISA
compatibility mode for arbitration? (I was under the impression that
EISA arbitration was a bit more sophisticated when dealing with
multiple bus masters). How does this effect the performance with
multiple bus masters?
You didn't mention interrupts at all (ISA edge triggered vs EISA
level). Level triggered interrupts are important if you plan on
sharing them (lord knows there aren't nearly enough).
Do some devices use the watchdog (sanity) timer for anything?
>The bottom line is that for the price, I am completely satisfied
>with my NICE Mini-EISA board, and will probably buy an Adaptek
>compatible EISA SCSI controller for it so I can run various
>versions of UNIX.
Ah, at least we agree here. The Adaptec 1742A is an excellent controller.
>Does this close the issue?
If the bottom line is "it's good enough for some," then I guess it
does, well, at least till you need to purchace an EISA ethernet or
video card... :-)
SHASHANK KUMAR
In Article 69978 b...@beach.cis.ufl.edu (Brian Hook) writes:
> You're missing full EISA compatibility. Does it support level based
> interrupts? And 24-bit addressing vies you only up to 16MB of addressing,
> which is rather limiting on a "Real" system.
Does that mean I can't use above 16 MB RAM, though I am right now using
20MB w/o any prob. Both NT & DOS seem to be using full 20MB RAM. My system
has 2 Ethernet Cards, 1 VLB Graphics card , 1 VLB IDE card, a MODEM, a 8bit
SCSI card, and I plan to get few more, till now I haven't had any problem
with interrupts, if thats what you mean by 'support level based interrupts'.
I didn't have to disable any options ( maybe I am not using any EISA card).
Cheers!!!
-Kumar
SHASHANK KUMAR
In Article 69869 b...@beach.cis.ufl.edu (Brian Hook) writes:
> IDE doesn't use DMA, so it doesn't have a problem with 24-bit vs. 32-bit
> DMA, unless you have some oddbal IDE adapter.
So where is the problem with 24-bit addressing for EISA, you don't really
need it for Modem, Ethernet, Sound, .... except for ofcourse price of around
$50 extra for EISA.
BTW, what does JOIN DATA mb. have, P-EISA, Super-ISA or Intel EISA ???
Cheers!!!
-Kumar
Prattle
SHASHANK KUMAR <ku...@sloop.cis.ufl.edu> wrote:
>In Article 69978 b...@beach.cis.ufl.edu (Brian Hook) writes:
>> You're missing full EISA compatibility. Does it support level based
>> interrupts? And 24-bit addressing vies you only up to 16MB of addressing,
>> which is rather limiting on a "Real" system.
>
>Does that mean I can't use above 16 MB RAM, though I am right now using
>20MB w/o any prob. Both NT & DOS seem to be using full 20MB RAM.
NT uses a double buffered scheme to handle >16mb on systems
with 24bit DMA limits (i.e., ISA, bad versions of EISA). One
interesting experiment might be to try OS/2, which doesn't--and
will treat memory >16MB on ISA systems as a fast swap area. IF
it does the same on your EISA system, it might be a quick way
to reveal its limits.
> My system
>has 2 Ethernet Cards, 1 VLB Graphics card , 1 VLB IDE card, a MODEM, a 8bit
>SCSI card, and I plan to get few more, till now I haven't had any problem
>with interrupts, if thats what you mean by 'support level based interrupts'.
>I didn't have to disable any options ( maybe I am not using any EISA card).
I don't think that was what was meant.
--
all new thread elimination .sig: (make your own!)
The 21st century begins at 12:00:00 on 1/1/2001. The proper date for new
millenium parties is 31/12/2000, not 31/12/1999. Years are cardinal
numbers. 2000+1=2001. New, creative counting schemes need not apply.
Brian Hook
In Article 69869 b...@beach.cis.ufl.edu (Brian Hook) writes:
> IDE doesn't use DMA, so it doesn't have a problem with 24-bit vs. 32-bit
> DMA, unless you have some oddbal IDE adapter.
> So where is the problem with 24-bit addressing for EISA, you don't really
> need it for Modem, Ethernet, Sound, .... except for ofcourse price of around
> $50 extra for EISA.
Okay, let's get something clarified. First, the EISA bus full spec says
that you get 32-bit DMA and 32-bit bus addressing. This translates to the
need of full EISA if you have a 32-bit DMA SCSI controller. IDE doesn't
use DMA period, so EISA vs. ISA is irrelevant in theory. As for bus
addressing, if you have a card that needs to access > 16MB, say a video
card that wants to map itself above 16MB, then EISA makes more sense than
ISA.
Jerry Shekhel
:
: The bottom line is that for the price, I am completely satisfied
: with my NICE Mini-EISA board, and will probably buy an Adaptek
: compatible EISA SCSI controller for it so I can run various
: versions of UNIX.
I don't know about the NICE board, but some time ago I tried some generic
EISA/VL motherboard using the HiNT Caesar chipset. It worked fine with
DOS, Windows, and Linux, but it had one problem. If I ran any application
which played digitized sound (I have a PAS-16), it would run fine, and go on
running fine, but the next time I tried to reboot the machine it would lock
up and start beeping in low-pitched tones at me. At that point I'd have to
use the big red switch. Seems like a DMA problem to me.
BTW, I have now used 3 motherboards using the SIS chipset -- a 386SX, 386DX,
and 486VL. All have performed flawlessly.
: Michael Condict
--
+-------------------+----------------------------+---------------------------+
| JERRY J. SHEKHEL | Molecular Simulations Inc. | Time just fades the pages |
| Drummers do it... | Burlington, MA USA | in my book of memories. |
| ... In rhythm! | je...@msi.com | -- Guns N' Roses |
+-------------------+----------------------------+---------------------------+
Bill Campbell
part of a month trying to get an EISA system up using Hint main
boards! We tried both Adaptec 1742s and BusLogic EISA
controllers, and both worked (more or less). The real killer was
the Racal Interlan ES3210 NIC. One failed both DMA and Interrupt
tests so we got another which did the same thing.
Unfortunately, none of the support people we talked to at Adaptec
or Racal Interlan was aware of the incompatibilities (they will
be after I e-mail them the articles I found today :-).
Bill
--
INTERNET: bi...@Celestial.COM Bill Campbell; Celestial Software
UUCP: ...!thebes!camco!bill 8545 SE 68th Street
uunet!camco!bill Mercer Island, WA 98040; (206) 947-5591
SPEED COSTS MONEY -- HOW FAST DO YOU WANT TO GO?
Chris Smith
It's the Super-ISA species.
The "EISA" bus has 24 address lines and 32 data lines. It can be used with
Adaptec 174x and DTC 329x 32-bit EISA controllers. I'm pretty sure it does
32-bit data transfers (from the specs, and the speed). But nothing, DMA,
bus mastering, or black magic, could address above 16meg since there are no
address lines for that memory.
It works fine with dos, os/2, and linux.
The i/o speed (with the DTC controller) is extremely impressive. The video
speed (to an S3805 card across the VL bus) is extremely impressive. The
cache refill time is extremely awful. The cache hit time is fine.
I don't remember the exact numbers right now, but i/o transfers from the
controller cache to memory run at about the same speed as memory-cache
transfers, something like 10 mb/sec. (Actually I think it's mem-mem transfers
too big to fit in the cache that run at ~10 mb/sec, but any way you slice
it the i/o is very well implemented and the cache controller is very badly
implemented.)
With my configuration, os/2 and linux use memory above 16mb about the way
dos uses memory above 1mb, more or less treating the extra memory as a
ram drive. My "os/2 floppy" partition does not work with >16meg, I never
figured out why.
This is NOT the motherboard to get if you want to use tons of memory, and
os/2 can always use a little more memory. If <=16meg will do, it's a very
nice and very cost-effective system.
To find out what mode things will work in on the Nice mini-eisa board,
I suggest that you simply look up the bus specification in the manual
and count the address lines. (You may have already; I missed the beginning
of the thread.)
>One last note. NICE says that you need to use different ".CFG"
>files when you configure your Adaptek or Bustek controller.
>This is because of some minor incompatibilities of the HINT
>EISA, perhaps the previously quoted thing about not supporting
>watchdog timers? In any case, the incompatibility has nothing
My motherboard manual (for a different board) says you must change
level-triggered interrupts to edge-triggered (no IRQ sharing).
Andrew Williams
> NT uses a double buffered scheme to handle >16mb on systems
> with 24bit DMA limits (i.e., ISA, bad versions of EISA). One
> interesting experiment might be to try OS/2, which doesn't--and
> will treat memory >16MB on ISA systems as a fast swap area. IF
> it does the same on your EISA system, it might be a quick way
> to reveal its limits.
But then again, it might not... OS/2 can still use the >16Mb area for
code that can't be swapped, and apparently it is smart enough to do so.
Putting all the kernal and device drivers up there will save a LOT of
space in the rest of the RAM, and you will definitely notice an
improvement over 16Mb.
And whats more- can you tell me the exact difference between 'a
double buffered scheme' and 'fast swap space'? Both involve moving the
code to low memory before writing it to the swapfile, and it sounds like
they might be different names for the same thing...
Andrew Williams,
Physics, Uni of Western Australia.
Dzung L. Pham
An EISA/VESA motherboard with a Hint chipset IS worth getting since it can
be had for about the same price as a VESA only motherboard. As long as
the motherboard operates in P-EISA mode, you'll get no deficiencies as
long as you're using bus-mastering cards, allowing 32 bit DMA and 32 bit
addressing.
Any corrections ? additional caveats ?
-- Dzung Luu Pham
The Applied Physics Laboratory or The George Washington University
(dzung...@jhuapl.edu> (pha...@seas.gwu.edu)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~