wales@valeria.cs.ucla.edu (07/27/88)
A while back, I asked for comments on 286-based accelerator boards for
PC's or XT's (or clones thereof). I have a Taiwanese turbo XT clone
(8-MHz NEC V-20, 8/4.77 MHz clock, 0 wait states). The RAM is a mix of
120-ns and 150-ns chips; yes, I know that 150-ns chips should theoret-
ically not work at 8 MHz with 0 wait states, but apparently the toler-
ances on these chips are such that they do work quite reliably (at
least, I've never had any problems, and I run *everything* at 8 MHz).
The "turbo board" market appears to be a lot less clear-cut and satis-
fying than I had initially supposed. Several people (including people
on the net, as well as several computer dealers) advised me against
this route -- saying that they had seen many, many times when these
boards simply didn't work right -- and even when they did, the speedups
in actual practice were not nearly as dramatic as one might hope for,
since you're still stuck with the 8-bit bus of the original system.
I had initially been attracted to the Orchid "Tiny Turbo 286" acceler-
ator board. But a call to Orchid's technical support dampened this idea
-- they told me in no uncertain terms that the Tiny Turbo 286 DEFINITELY
WOULD NOT WORK in a turbo clone built around an 8-MHz processor, EVEN if
the clock were always run at 4.77 MHz. I couldn't get any comprehensi-
ble explanation of why; they just said it wouldn't work and wasn't worth
even trying, period. Orchid does have a faster turbo board, as does AST
-- but in either case I would very probably have to replace all my RAM
with 100-ns or faster (a sizable expense nowadays).
I also looked into the idea of upgrading to a 286-based motherboard.
AST, for instance, has one called "Xformer/286" ("X" pronounced "trans")
that runs at 10 MHz with 0 wait states. It has 512K of RAM on the board
and can take either 128K or 512K of additional chips. Of course, these
chips had better be 100-ns (at least, AST was unwilling to recommend the
use of slower chips -- not surprising, considering the clock speed and
the 0 wait states). The AST Xformer/286 is said to be exactly the same
size as the standard XT motherboard (except for a little extension on
the front), and is supposed to fit in the standard XT cabinet. (My
clone's motherboard and cabinet are the standard XT size. Note that the
AST Xformer/286 is known *not* to be suitable for some clones; the AST
people mentioned the Tandy 1200, Leading Edge Model D, Compaq Deskpro,
AT&T 6300, and Epson Equity II in their "don't waste your time" list.)
After considering the idea for some time, though, I have decided not to
try to upgrade my present system. I've already indicated my less-than-
positive conclusions regarding accelerator boards. As for swapping the
motherboard, four factors tend to make me decide against it:
(1) I would probably have to upgrade my current power supply (which is
rated at about 140 W) to a 200-W supply. One dealer suggested that
I *might* be able to get away with my current supply after all, but
I'm not sure I'm prepared to take that gamble.
(2) I would probably have to buy a load of 100-ns RAM chips -- though I
could save some money by expanding the 512K on the AST motherboard
with just 128K extra (i.e., 64K chips instead of 256K).
(3) I would probably have to get a new video board. My current board
(a Taiwanese Hercules clone) has 150-ns RAM chips; while it works OK
at 8 MHz (see above comment on lax timing tolerances in RAM chips),
I would consider it unlikely to work at 10 MHz. (Then again, are
video boards set up to use extra wait states to the video RAM, inde-
pendently of the wait states the CPU may or may not be using for the
regular RAM?) I do plan to upgrade my video stuff eventually, but I
don't really want to do so *now*.
(4) I'm not sure how much of a speedup (if indeed any at all) I could
expect to get from my hard disk card, since it would still use only
the 8-bit XT bus connections. And I'm not really of a mind to buy a
new hard disk card specifically designed for an AT bus.
In conclusion, I suspect that upgrading my system with a new motherboard
would likely end up costing almost as much as a new system -- and that
I'd be better off just waiting until I can afford to trade in my whole
system for a "real" AT clone.
Comments, anyone?
-- Rich Wales // UCLA Computer Science Department // +1 (213) 825-5683
3531 Boelter Hall // Los Angeles, California 90024-1596 // USA
wales@CS.UCLA.EDU ...!(uunet,ucbvax,rutgers)!cs.ucla.edu!wales
"Spiff's hyper-freem drive malfunctions! The aliens close in!"phil@amdcad.AMD.COM (Phil Ngai) (07/27/88)
In article <14742@shemp.CS.UCLA.EDU> wales@CS.UCLA.EDU (Rich Wales) writes:
about the choices for upgrading his turbo XT clone.
I'd like to throw in a few commments. These are just opinions.
First, I would avoid a 286-based accelerator boards for XTs.
You'd be stuck with the 8-bit memory bus and I think you'd notice
the performance difference very easily.
Buying an AT clone motherboard seems more attractive. If you do so,
watch out for card slot spacing and the location of the keyboard
connector. Either make sure it fits your case or buy a new case to go
with the new motherboard.
Expect prices of around $90 for an AT class power supply and $90 for
an AT clone case. You can pay less if you shop hard.
You probably won't have any problem with your existing video board.
In any case the MIT Herc clones we have use only two RAM chips and
replacing them shouldn't be too expensive, if you need to, which
I doubt.
I have heard that most of the disk speed up between XT and AT is due
to faster seek times on the AT disk itself. There are those who claim
the 16-bit bus does not make disk IO much faster, although it does
make a difference for memory accesses. If you figure the disk can
deliver either 5 or 10 million bits per second and the memory can
deliver 16 or 32 million bits per second (either a byte every 500 ns
or two bytes every 500 ns) then you wouldn't expect an 8-bit disk
controller to be too much of a bottleneck.
If you do decide to choose an AT clone motherboard, I would look for
something that allowed you to split 1 meg on the motherboard into 640K
for DOS and 384K for LIM expanded memory. Boards built around the new
CHIPS NEAT chip set are in this category.
Some MBs only allow you to put 640K on (16 256K and 16 64K chips).
Others allow 1 meg split as 640K DOS and 384K extended.
Access to extended memory (above the 1 meg that DOS understands) is
slower because you have to go into protected mode and leaving it
is usually slow. (But NEAT speeds this up too.)
Access to LIM expanded memory goes at normal memory speeds since you
never have to leave real mode. You will find this useful for putting
RAM disks or disk caches in.
You should also look for the ability to clock the bus at a different
speed from the CPU. Various MBs can clock the bus at a constant 8 MHz,
eliminating your worries about slow video, etc, while running the CPU
at 10, 12.5, or beyond.
WD has an interesting motherboard which only has three slots. They
can do this because almost everything you can think of is integrated.
The only thing you need to add is a networking card. Memory, video,
disk, serial, mouse, printer, they have it all. Unfortunately, I
don't know the price yet.
You should be able to get a new motherboard for around $400. Add
another $180 for case and power supply and $300 for a disk. Now you're
up to nearly $900. You may need a new keyboard too. Is this still
cheaper than a complete new clone. At least you're reusing your
monitor, video card, serial, floppy, etc.
We haven't even begun to discuss the 386...
--
I speak for myself, not the company.
Phil Ngai, {ucbvax,decwrl,allegra}!amdcad!phil or phil@amd.comRalf.Brown@B.GP.CS.CMU.EDU (07/27/88)
In article <14742@shemp.CS.UCLA.EDU>, wales@valeria.cs.ucla.edu writes: }120-ns and 150-ns chips; yes, I know that 150-ns chips should theoret- }ically not work at 8 MHz with 0 wait states, but apparently the toler- }ances on these chips are such that they do work quite reliably (at }least, I've never had any problems, and I run *everything* at 8 MHz). In an XT, each memory access takes four clock cycles. At 8 MHz 0 waits (I've NEVER heard of an XT with waits!), the memory cycle time is 500ns, which means 250ns memory chips will work just fine (memory cycle time = 2 * access). Even the fastest 8088 machine I've seen advertised (12.8 MHz) would run just fine on 150ns chips. It's only on 286 machines that memory access time becomes important, because a memory access on a 286 or 386 takes two clocks plus any wait states. } I would consider it unlikely to work at 10 MHz. (Then again, are } video boards set up to use extra wait states to the video RAM, inde- } pendently of the wait states the CPU may or may not be using for the } regular RAM?) I do plan to upgrade my video stuff eventually, but I Most video boards insert loads of wait states because they can only let the CPU access the RAM when the video circuitry isn't. This mean that for most boards the CPU must wait an average of 1.1 microseconds to get access. }In conclusion, I suspect that upgrading my system with a new motherboard }would likely end up costing almost as much as a new system -- and that }I'd be better off just waiting until I can afford to trade in my whole }system for a "real" AT clone. I've come to much the same conclusion about upgrading my AT to a 386. -- UUCP: {ucbvax,harvard}!cs.cmu.edu!ralf -=-=-=- Voice: (412) 268-3053 (school) ARPA: ralf@cs.cmu.edu BIT: ralf%cs.cmu.edu@CMUCCVMA FIDO: Ralf Brown 1:129/31 Disclaimer? I |Ducharm's Axiom: If you view your problem closely enough claimed something?| you will recognize yourself as part of the problem.
seeger@beach.cis.ufl.edu (Charles Seeger) (07/27/88)
In article <22465@amdcad.AMD.COM> phil@amdcad.UUCP (Phil Ngai) writes: >In article <14742@shemp.CS.UCLA.EDU> wales@CS.UCLA.EDU (Rich Wales) writes: >about the choices for upgrading his turbo XT clone. > >I have heard that most of the disk speed up between XT and AT is due >to faster seek times on the AT disk itself. There are those who claim >the 16-bit bus does not make disk IO much faster, although it does >make a difference for memory accesses. If you figure the disk can >deliver either 5 or 10 million bits per second and the memory can >deliver 16 or 32 million bits per second (either a byte every 500 ns >or two bytes every 500 ns) then you wouldn't expect an 8-bit disk >controller to be too much of a bottleneck. But, the speed of the controller itself is *really* important, i.e. can it handle the full disk speed (1:1 interleave)? For most applications, disk transfer rate is more important than access time. My 1:1 RLL (7.5 Mbps) controller gives cheap speed, in my 20 MHz 386/387 2MB box (~$3500.00). Such controllers are too fast for 8 MHz 8 bit buses. Afterall, the HDC can't have all the bus bandwidth. >If you do decide to choose an AT clone motherboard, I would look for >something that allowed you to split 1 meg on the motherboard into 640K >for DOS and 384K for LIM expanded memory. Boards built around the new >CHIPS NEAT chip set are in this category. I prefer to use that 384K to shadow the system BIOS into faster RAM. It really speeds memory checking, BIOS video, etc. The NEAT chips do this. My AMI BIOS doesn't copy my HDC BIOS into RAM (it does the main BIOS), but my disk performance should go up a significant bit when I write the code to do it. >We haven't even begun to discuss the 386... ..., which can, with the right software, emulate LIM in extended memory. They are sooo nice, but if you don't need high 32 bit performance, it'll be interesting to see what happens with the 386SX. Personally, I wouldn't get a accelerator board. You can put together a new turbo AT clone for ~$1500.00, and sell your old turbo XT for ~$400.00. Really, not much more $. Chuck