OS/2 Shareware BBS: 14 Text/14-Text.zip/video2.zip/VIDEO2.TXT

Text File | 1994-01-30 | 16KB | 339 lines

SIXTH REVISION (01/27/94):

From all that I've learned, the choices for a good OS/2 2.1 video card
as I see them (in increasing order of performance and to some degree
cost) are:

* XGA-2 (CatsEye/X and Radius):  ISA cards using the IBM chipset that
have excellent driver support under OS/2 2.1.  Speed is reportedly
good, but not as fast as the latest S3, Weitek and IIT accelerators.
Color depth is limited to 16 bits (static 64K color pallette).
However, further development of the XGA architecture has reportedly
been abandoned (and the CatsEye/X discontinued) as IBM has adopted S3
for future products.

* ATI Graphics Ultra Pro [VRAM] and Plus [DRAM]:  Able to use generic
8514/A drivers in many environments, which makes them a good choice
for OS/2.  ATI native OS/2 drivers have been released, but they aren't
much faster than the 8514/A (i.e., significantly slower than ATI
Windows drivers) -- essentially they are 8514/A drivers hacked to
support more resolutions, and have proved to be a bit buggy.  Thanks
to lukewarm support for OS/2, ATI drivers won't be getting hardware
support for Ultimotion video.  These cards used to be quite pricey,
but have now come down to a somewhat more reasonable level.  ISA, VLB
and EISA versions are available.  (The reincarnation of the old 8-bit
Graphics Ultra is unimpressive.  The newer XLR is claimed to be up to
twice as fast as the Ultra Pro, but real-world results seem to be more
in the range of 10-20%.  Caveat emptor in both cases)

* Tseng ET4000W32-series [DRAM] (Hercules Dynamite, Diamond Stealth
32, and a number of smalled suppliers):  Fastest DOS and dumb frame
buffer performance.  The best non-accelerated driver support of any
card.  A dumb frame buffer ET4000 performs much faster on VLB than on
the ISA bus, giving perhaps 1/2 the speed boost of an ISA bus S3
accelerator (though much less than a VLB accelerator), so a VLB Tseng
ET4000W32 should give at least respectable performance even without
accelerated drivers.  Very good accelerated Windows performance.
Accelerated drivers for OS/2 are just becoming available.  A good
value, with prices about the same as the S3 801/805.  However, VESA
70+ Hz refresh rates are only supported at up to 8-bit (256) colors;
16-bit (64K) and 24-bit (16M) color are limited to 60 Hz.  A faster
version, the ET4000W32i ("i" as in interleaved memory access), has
performance that is closer to S3.  An even faster version (thanks to a
drawing engine at last), the ET4000W32p, which is just becoming
available, finally has performance comparable to the S3 805-series;
one of the first is the Diamond Stealth 32 (which, despite the
confusing "Stealth" name is *not* an S3-based card -- OS/2 drivers due
to be released "soon").

* S3 801/805 [DRAM] (Diamond Stealth 24, Orchid F1280+, Actix, STB):
S3 is well-supported by drivers in most environments.  IBM has
commited full 32-bit seamless driver support for certain S3 cards, and
excellent 16M-color drivers have been released (on CompuServe GO
OS2SUPPORT and look for S3_16M.DSK); a driver update is due sometime
soon.  The 801/805 cards are a good deal faster than the old fast S3
911/924 cards (even unaccelerated and especially in DOS text mode),
and are a real bargain thanks to the use of lower cost DRAM (about
$160 on the street with 1 Mb).  24-bit color in 640x480, 16-bit (64K)
color at 800x600, and 8-bit (256) color at 1024x768.  Greater color
depth and up to 1280x1024 resolution (depending on the video card)
with 2 Mb of DRAM.  VESA-standard refresh rates possible at all
resolutions (again depending on the video card).  Both ISA and VLB
versions.  A faster version of the 805, the 805i ("i" as in
interleaved memory access), is just becoming available.  (Note that
the Diamond Stealth 32 is *not* an S3-based card.)

* S3 928 [VRAM] (Diamond Stealth Pro, Actix, Metheus, Number Nine,
STB):  Faster than the S3 801/805, but significantly more expensive
(about $220 on the street with 1 Mb) due to the use of VRAM; the speed
difference only really shows up at the higher color depths,
resolutions (up to 1280x1024 with 1 Mb) and/or refresh rates
(VESA-standard at all resolutions).  Upgrade to 2 Mb ($80-100) allows
(unaccelerated) 24-bit color in 800x600, 16-bit (64K) color at
1024x768, and 8-bit (256) color at 1280x1024.  Driver support is the
same as the S3 801/805; i.e., excellent (see above).  Some cards
(e.g., Metheus and STB) support up to 4 Mb of DRAM, which gives 24-bit
color at 1024x768 and resolutions as high as 1600x1200, although OS/2
drivers for these resolutions do not yet exist.  (Note that the
Diamond Stealth 32 is *not* an S3-based card.)

* P9000 [VRAM/DRAM] (Diamond Viper, Orchid P9000):  The P9000 is very
fast, faster than the S3 928 as measured by WinBench 3.11, although
the real world performance difference is small.  However, performance
in DOS text mode is normally poor due to the need for a separate VGA
chip that typically is slow.  (An improved version due in 1994, the
P9100, will have fast integrated VGA.)  Thanks to top accelerated
performance, separate VGA and pricey VRAM (resolutions up to
1280x1024x256, all at VESA-standard refresh rates), at the moment it's
about $300-400 on the street.  OS/2 drivers are thus far slower and
more limited than Windows drivers.  Early cards were a bit buggy, so
be sure to get the latest rev (chipset and BIOS).

* IIT X-15 (Hercules Graphite and Orchid Celcius):  The X-15 (an
improved version of the X-14) is claimed by IIT to be the fastest
current accelerator; it probably is for drawing (CAD), but is
otherwise only comparable to the S3 928 and Weitek P9000.  (Orchid
thinks the X-15 is a bit faster than the P9000, which Orchid also
sells.)  OS/2 driver support is still problematic.  The X-14 and X-15
are *not* fully XGA-compatible, so XGA drivers won't work.  Watch out
for cards that are FCC Class A rather than FCC Class B.

* Matrox MGA and MGA-II [VRAM]:  Probably has the greatest hardware
potential (including support for 3D) of any current single chip
solution, but I have not yet had a chance to test it.  High pricing to
match the high performance, but the new MGA-II is a bit less expensive
than the original MGA (due to the ommission of 3D features that aren't
used for OS/2).  Windows drivers are reported to be faster than other
accelerators.  8-bit drivers for OS/2 are reported to be buggy and
slower than the Windows drivers.  Drivers are also claimed to be
available for Windows NT and X-Window.


At 1993 Fall COMDEX I saw MediaVision's new color video card, which is
claimed to be very fast in 24-bit (16M true color) mode and looked
very impressive.  Full 32-bit seamless OS/2 drivers are claimed to be
in development and are supposed to be released sometime soon.  This is
a card to watch.

Other interesting developments at 1993 Fall COMDEX included new, ultra
fast 64-bit chips from S3, together with a motion video accelerator
chip.  However, boards with these chips won't be available before
min-1994, and OS/2 drivers are anybody's guess.


(BTW, my research has convinced me that WinBench is NOT an accurate
predictor of real-world performance.  The tests are poorly designed
and not typical of real-world usage.  This goes not only for version
3.11, but also for the new version 4.0 introduced at 1993 Fall COMDEX.
Certain chips and/or drivers have been optimized to win the WinBench
contest without providing as much real-world performance as the
numbers would suggest; some are even slower in real-world performance
than chips and/or drivers with lower WinBench ratings.  Caveat
Emptor.)


The bottom line is that *any* of these cards are reasonable choices,
depending on what you want and what you can afford.  If I wanted to
run *anything*, I'd go with the Tseng ET4000W32p.  If I wanted the
best OS/2 support and bang for the buck I'd go with the S3 801/805i.
If I wanted top performance (and could afford it), I'd go with the S3
928, IIT X-15 or Weitek P9000, but for all but S3 only after actually
testing the drivers I needed (especially with the Matrox MGA).

-----

Another possible option is the Lava, a Canadian 24-bit board based on
three 8514/A chips that is claimed to have OS/2 support.  Because of
the very high price (US$1300) and uncertain support I have not pursued
it.

-----

I did not find that current Western Digital (Paradise) or Cirrus Logic
chipsets made sense for OS/2 users due to poor support and
performance.  (Although accelerated Cirrus Logic drivers are now
available, current performance does not measure up to other GUI
accelerators.)

-----

My personal choice was the Diamond Stealth Pro VLB, which was faster
and higher quality than most of the other S3 928 cards I tested.  (I
also liked the ability to save monitor settings in EEPROM.  The one
weakness is that 1024x768x64K is limited to 60 Hz refresh, but that's
a mode I do not use.)  The decisive factor for me on choosing the S3
928 over other chips was the excellent IBM 32-bit seamless S3 driver
support and the desire for top performance at an affordable price.

-John Navas
 CompuServe: 70244,2046
 Internet: [e-mail]
           [e-mail]

-----

GO GRAPHVEN  on CompuServe for information on STB and ATI.

GO GRAPHBVEN on CompuServe for information on Tseng, Diamond, Hercules
             and Genoa.

-----

Internet mail to video card companies can be addressed through
CompuServe as follows:

 STB Systems                Eric Lofland   [e-mail]
 ATI Technologies           Corilee Fox    [e-mail]
 Tseng Laboratories         Joe Curley     [e-mail]
 Diamond Computer Systems   Lance Tom      [e-mail]
 Hercules Computer Tech.    Dylan Rhodes   [e-mail]
 Genoa Systems              Neil Roehm     [e-mail]

-----

Additional contact information for high-end S3 928 suppliers:

ARTIST Graphics Co.
2675 Patton Rd.
St. Paul, MN  55113
800-627-8478; 612-631-7800
FAX: [phone]
Tech support: Use toll-free no.

ELSA America, Inc.
400 Oyster Point Blvd., Ste. 109
South San Francisco, CA  94080
800-272-3572; 415-615-7799
FAX: [phone]
Tech support: Use toll-free no.

Metheus Corp.
1600 N.W. Compton Dr.
Beaverton, OR  97006-6905
800-638-4387; 503-690-1550
FAX: [phone]
Tech support: Use main no.

National Design, Inc.
1515 Capital of Texas Hwy., S, 5th Fl.
Austin, TX  78746
800-253-8831; 512-329-5055
FAX: [phone]
Tech support: Use main no.

Number Nine Computer Corp.
18 Hartwell Ave.
Lexington, MA  02173
800-GET-NINE; 617-674-0009
FAX: [phone]
Tech support: Use main no.

STB Systems, Inc.
1651 N. Glenville, Ste. 210
Richardson, TX  75081
800-234-4334; 214-234-8750
FAX: [phone]
Tech support: Use main no.

-----

FOOTNOTES:

-----

 > In one of your posts in OS2DF1 you mentioned that some video cards
 > need a memory aperature on ISA machines and that this was a problem
 > if the machine had 16 Mb or more of RAM.  Well, I'm in the market
 > for a video card too and have an ISA machine with 20Mb of RAM and
 > was wondering what a memory aperature was and what problems the
 > high memory caused.

The problem seems to be most commonly associated with the ATI Ultra
Pro (VRAM) and Ultra Plus (DRAM) on the ISA bus.  For best performance
they need a linear memory aperature of at least 2Mb that is 2 Mb above
all physical RAM (which must be contiguous).  On an ISA bus that
aperature must be below the 16Mb line because of addressing limits on
the ISA bus.  That means the memory aperature can't be used with more
than 12 Mb of RAM.  A VLB or (expensive) EISA card solves the problem.
(Putting an ISA card in an EISA slot doesn't help.)

BTW, I don't think the memory aperature affects the performance using
IBM's 8514/A drivers, only ATI (Windows and presumably Windows NT
beta) drivers.  However, the Ultras are slower and more limited in
8514/A mode than they are with native ATI drivers, and ATI has yet to
release drivers for OS/2 (they're in beta).  Note also that ATI is
known for buggy drivers.

ATI claims only a 5% slowdown when the memory aperature is disabled,
but there have been several reports of big performance losses and an
inability to use 24-bit color; ATI's response is that it is still
working on optimizing its Mach32 drivers.

I received a report that the Radius XGA-2 card has the same video
aperture problem as the ATI card; i.e., it won't enable the 1MB
aperture in ISA machines with 16 Mb or more of RAM.  I therefore
assume that the CatsEye card has the same problem.

Note also that, while current S3 drivers generally do not use a memory
aperture, the newer S3 chips (801/805/928) do support it, and it's
likely that future driver performance increases will require it.

In fact, as more and more video cards require a memory aperature for
best performance, it becomes even more important to have a motherboard
that supports EISA, VLB or PCI, any of which solves the problem.

-----

 > Do you know how EISA figures in to the video board choice problem?
 > I have heard that EISA is much faster for DOS applications
 > (non-Windows, non-OS/2).  Since much of what I do is DOS apps, I
 > would like good performance there as well as any Windows or OS/2
 > graphics apps I later add.

The best chips for DOS applications (that do not have accelerated
drivers) are the Tseng ET4000 (unaccelerated) and ET4000W32
(accelerated), because they have the highest memory to screen
bandwidth and the fewest wait states (despite using cheaper DRAM) of
any VGA-type cards.  (Note, however, that the S3 805/928 on local bus
as fast or faster than the fast ET4000 on ISA bus.)

Graphic speed is, however, limited by the slow (usually) 8 MHz 16-bit
ISA bus.  One way to boost graphic (and other I/O) speed is to run the
ISA bus at 10 or even 12 MHz, assuming that your motherboard can do it
and that *all* your peripheral cards can handle it.

The second step up is 32-bit EISA, which has two or more times the
throughput of the 16-bit ISA bus.  However, you'll only really get
that speed boost with drivers or applications that are smart enough to
take advantage of the wider EISA bus; i.e., if the application only
writes graphics data 8 or 16 bits at a time (which most do), the EISA
bus probably won't be that much faster than the ISA bus (the only
difference being bus wait states).  EISA also has the disadvantage of
being quite a bit more expensive than ISA.

The third step up is VESA Local Bus (VLB).  Since VLB runs 32-bit
transfers at processor speed (up to a practical max. of 40 MHz), it
has up to 10 times the speed of the ISA bus, and is much faster even
for 8- or 16-bit transfers; i.e., no special driver or application
support is needed to get the benefit.  The good news is that VESA
costs only slightly more than ISA, which is why it's taking the market
by storm.

The fourth step up is Peripheral Component Interconnect (PCI), a new
standard in the making.  PCI is not widely available yet and it's
probably going to be more expensive than VLB.  In the real world there
probably won't be a noticeable speed difference between PCI and VLB
graphics cards on the 486.  It appears that VLB will continue to be
dominant on the 486, with PCI taking over on the Pentium.

The bottom line?  For about the same money that you'd pay for an EISA
graphics card you can get a VLB graphics card and a new naked VLB
motherboard, and transfer over your processor, RAM and peripheral
cards.  The result will be significantly faster in most applications.
This is why I *don't* recommend EISA for graphics cards.

BTW, real performance freaks can get an EISA/VLB motherboard and have
the best of both worlds.  <g>

<<END>>