The Quest for Speed: Octek DCA-2
by Alan Zisman
(c) 1995. First
published in Our Computer Player, February 1995
Octek Hippo DCA/2 486DX4/100 motherboard
$350 (US list)
Standard Computronics
Unit 333 Camridge Industrial Centre
13988 Cambie Road
Richmond, BC
V6V2K4
(604)273-7884/273-7889 fax
It seems to be an unwritten law of computing that
whenever some component
gets significantly faster, some other part of the system turns into a
bottleneck,
keeping you from really getting all the benefits you'd hoped for.
Take CPUs, for instance... the so-called brain of the
computer. The
original IBM PC, way back in 1981 used an Intel 8088 processor... a
hybrid
8/16 bit chip running at a ferocious speed of 4.77 mhz. (By comparison,
the Apple 2's processor, a true 8-bit chip ambled along around 1 mhz).
Now you can buy true 32-bit 486s running up to 100 mhz
(well, really
99 mhz, but who's counting), and Pentiums running at the same speeds,
but
at 64 bits internally. And that's not counting the more exotic
workstation
RISC chips with still higher speeds.
Even my now run-of-the mill 486DX2-66 gets rated by
the old-standby
Norton SI utility as having 97 times the power of that original PC.
Well, I'm really not computing 97 times quicker.
Take hard drives. That Seagate ST-225 20 megger in
that old XT had an
average access time of 65 msecs or so. Your new EIDE or SCSI drive has
an access time of maybe 9-10 msecs. Faster, yes. 97 times faster, no.
That old PC or XT used 150 nsec ram chips. Today's 486
or Pentium probably
is filled up with 60 or 70 nsec ram. That's a little more than twice as
fast, while trying to keep up with a CPU that's running at a clock
speed
that's 20 times faster.
How to keep up with the huge increase in CPU speed and
power has been
a problem ever since the spread of fast 286s in the late 1980s...
machines
that ran at 12 mhz or more.
First, we saw 'wait-states'... essentially these
solved the problem
by forcing the machine to wait-- pause to let the ram catch up. In
other
words, slow the CPU down.
This was workable, in an ironic sort of way, until the
era of the 386-33...
around 1991 or so. Then, a new trick became common-- cache ram.
You see, it had always been possible to make ram that
ran faster than
the common DRAM (Dynamic RAM). Much faster SRAM (Static RAM), at speeds
of 20 nsecs or so, had been manufactured for years. The problem was
that
it was much more expensive than DRAM. So some bright soul started
producing
motherboards with 64 or 128kb (or more) of fast, expensive SRAM in a
memory
cache.
Don't confuse this with your disk cache, most often
provided by a software
program such as SMARTDRIVE. A disk cache stores the most recent sectors
of information read from your hard drive in ram, on the assumption that
you're most likely to want to read it again sometime in the near
future.
This can dramatically speed up hard drive operations (helping your
relatively
slow hard drive keep up with the much faster CPU).
Your memory cache takes this one step further... here,
you're storing
in fast ram (SRAM) the most recent operations performed in memory...
again
on the assumption that you're likely to want to do them again. If
that's
the case, they can be performed in fast SRAM, rather than in the
relatively
slower DRAM that makes up your main memory.
486 chips got yet another, even faster level of cache.
A tiny (8kb in
most models) cache was built right onto the CPU itself... so if the
information
required is stored there, it can be performed even faster than if it
has
to be read from the SRAM 2nd-level (L2) cache.
All tricks to try to keep up with the increasingly
powerful CPU.
However, as software moves to 32-bit operating
systems, such as OS/2,
Windows NT, or the often-promised Windows-95, more and more strain
falls
on the L2 cache. And this gets magnified with the common clock-doubling
and tripling chips, such as my DX-2/66 or the DX-3/99, or with all
Pentiums.
For with these chips, while the CPU is speeding along
at 66 to 99 mhz,
the memory subsystem is running at half or a third that speed. Working
at 16 bits in DOS or Windows 3.1 isn't much of a problem. But pumping
data
through 32-bits at a time can turn it into a bottleneck once again.
Just
like with my old 286-12, it's back to hurry up and wait (though at a
much
faster pace!)
Enter the DCA/2... a motherboard from Ocean
Information Systems, with
the perhaps unfortunate full name of 'Octek Hippo DCA/2'. For this
motherboard
has absolutely no resemblance to a large African mammal that wants to
spend
its life wallowing in the swamp.
It gets around the L2 cache bottleneck by supplying
ALL the ram on the
motherboard-- a full 16 meg on the system I tested, in the form of fast
15/35 nsec EDRAM. In effect, all 16 megs can run as your cache...
compared
to the 256-512 kb on your average 486 or Pentium motherboard.
The results are dramatic. By some tests, this 486/100
outperforms the
supposedly more powerful (and much more expensive) 100 mhz Pentium,
making
it the hottest motherboard currently available for Intel-standard PCs.
And it'll give you the correct answers while dividing prime numbers,
too!
But what about cost, I hear you ask. Hasn't the
problem with this solution
been the high price of the ram? Didn't I just mention this a few
paragraphs
ago?
While there is still a price differential, it seems to
have shrunk dramatically.
The distributor's quoted me a price of $201 for a standard 4 med memory
SIMM, while for the much faster ram for this motherboard, I got a price
quote of $216... $60 more for 16 megs. That's not much to pay for such
a performance benefit, especially since the price difference between
100
meg 486s and Pentiums is quite a bit greater.
(Careful readers will have noted that I've referred to
this ram as EDRAM,
while the chips in your current memory cache are SRAM... SRAM is still
pricy; EDRAM (Enhanced Dynamic RAM ?) is almost as fast, but unlike
SRAM
needs to be refreshed-- continually rewritten).
All in all, this seems to be a clever and affordable
way to bypass one
of those performance bottlenecks. The DCA/2 is currently available in
versions
for the 486/66 and 486/100, which it is suggested, are comparable to a
Pentium 60 and 90 in performance. But this strategy can be used with
even
more powerful CPUs, resulting in still better performance. And they've
announced upcoming motherboards using Pentiums, multiple-CPUs
(symmetric
multiprocessing) and even Power-PCs.
I wouldn't be surprised if ALL motherboards work this
way in a year
or so, but if you want increased performance now, this could be the way
to go.