Welcome to the 6th episode of Worst Gaming Cards and our journey to deep dive in the absolute worst the 90’s had on offer. Today we’ll be taking a look at a company
that’s almost forgotten these days, Cirrus Logic and more specifically their 1st 3D
chip and also the last graphics chip they ever developed. Cards from Cirrus Logic were quite successful
in the early 90s, but the unstoppable wave of 3D acceleration during the second half
of the decade forced a lot of companies to adapt or die. The company managed to release only one 3D
capable chip, the Laguna3D, which came out too late and after its failure they were force
to totally abandon the graphics chip market. We’ll analyze what’s so wrong with this
chip in a moment. But first, as always, let’s have a quick
look into the company’s history. Originally known as Patil Systems until eventually
getting its final name in 1984, Cirrus Logic was started in 1981 in Salt Lake City. Founded by former MIT professor Dr. Suhas
Patil, the aim of the company was to pursue the development of software that would in
the design of VLSI chips using new techniques invented by its founder. In 1984, Michael Hackworth, impressed by this
new technology, left his prestigious position of senior vice president at Signetics and
joined the company in January 1985 as president and CEO. Patil Systems was also renamed to Cirrus Logic
and moved its headquarters to Silicon Valley. Hackworth then suggested that they should
start designing audio, graphics and controller chips with their unique tools. The plan was to sell these chips on existing
large or emerging markets. But initially, they started with something
quite different. Their first product was an early “set-top”
cable TV converter chip code-named “Starfish”. After Starfish’s success, they looked more
closely into the controller market. Developing controller chips for upcoming IDE
drives, they proved once again to be successful, starting a long line of very profitable products. The next step was to infiltrate the graphics
chip market! Cirrus Logic engineers were looking into designing a display graphics controller in 1987 just when IBM announced its new Video Graphics Array
(VGA) technology standard. Just a few months later they had their own
VGA chipset. The resulting chips that formed this chipset,
the GD410 and GD420, ended up on a card by Video 7 and offered support for both EGA and
VGA screens, with the card itself barely exceeding the length of a standard 8bit ISA connector. Unfortunately, the chipset wasn’t perfect
and reviewers noted, that card didn’t seem to offer 100% compatibility with the VGA standards based on various tests and benchmarks of the era. This was addressed a year later with the release
of a replacement based on newer chips named GD510 and GD520. Monitor compatibility was preserved and the
chipset also got support for 16bit BIOSes, which increased performance when used with
386 CPUs. Of course this was achieved only in combination
with 16bit ISA slots. By 1989, the time had come to focus on another
important segment of the computing market, that of notebooks. And Cirrus Logic did just that with an upgraded
variant of GD510 / GD520 chipset, once again with boring numbers used for the naming scheme. The GD610/GD620 chipset, as it came to be
known, was destined for notebook markets with new features such as mapping of colors to
shades of grey. This was another successful entry into a new
market segment for the company and started what was to become a long line of popular
mobile chips. We’ll be skipping the rest of these for
today, as there were far too many and are beyond the interest of this specific video. The following year brought Cirrus Logic’s
1st single chip VGA controller, which was called GD5320, a very low-cost solution with
no new exciting technologies behind it. In 1992, Cirrus Logic acquired one of its
competitors, a smaller company by the name of Acumos, which had only 2 graphics chips
on the market by that point. These were the AVGA1, released in 1991, and
the AVGA2 released just a year later with twice the maximum memory size and SVGA support. Both of them were faster than Cirrus’ very
own GD5320 and what’s more, they featured an integrated RAMDAC. This allowed them to compete very well with
other low-cost chips like Trident’s TVGA 9000. In the same year as the acquisition of Acumos,
Cirrus revealed their first True Color VGA chips to be added to their line of products. As such, the entry level, low-cost GD5420 which was limited to just a 16bit memory bus and 256 colors was complemented by the more mainstream
GD5422 offering a 32bit bus and true color support. The GD5426 was essentially the same chip as
the GD5422, but it featured support for VL-Bus. 1993 was the year of the VL-Bus and Cirrus
released two new chips with an increased maximum memory size of up to 2MB. The GD5428 got also featured a BitBTL engine
for Windows 3.0 and 3.1 acceleration. The last members of the True Color family
were the GD5429 released in 1994 and the elusive GD5425 announced in 1995. The GD5429 had further enhanced Windows Acceleration and the GD5425 may have been a low-cost chip with TV-Out support. But we haven’t found any photos of it, so
the only evidence to its existence are press releases and datasheets. There’s a chance that interest in cheap
TV-out enabled cards on a bus other than PCI was low and chips were therefore canceled
or produced in very limited numbers. 1994 also saw the release of a new family
of chips named Alpine with support for the PCI bus and offering 64bit GUI acceleration,
with few exceptions such as the entry level GD5430 limited to a 32bit bus and a maximum
of 2MB of memory. The mainstream GD5434 of the same line of
products offered a full 64bit engine and a maximum memory size of up to 4MB, although
it looks like no vendors bothered equipping cards with that size. 1995 saw another 4 new chips. The low-end GD54M30 limited once again to
2MB and mid-range GD5436 with a max supported 4MB and an enhanced BitBLT engine as well
as better performance in 32bit color modes. The two remaining chips were both low end
with a 32bit engine, but this time with video acceleration and zoom functions. Features are almost the same between them
with only minor differences in dropped support for VL-Bus on the GD54M40 chip. It instead gained support for an external
MPEG decoder. 1996 was the year that most companies debuted their 1st chips to take on 3D accelerator market. But this was not the case for Cirrus Logic,
who were starting to get left behind. Both new chips, the GD5446 and GD5480 had
enhanced video acceleration and supported more external decoders. The cheaper and much more common GD5446 was
limited to EDO memory support. The GD5480 by contrast supported far superior,
but more expensive SGRAM modules. Another 2D chip, the GD5462, was announced
in 1995 using then hot & new RDRAM memory modules from Rambus. Unfortunately, cards using this chip only
started appearing in 1996 and were meant to be 2D high-end video multimedia accelerators. Problem was, that no-one was at that time
was interested in pure 2D cards and the GD5462, codenamed Laguna was somewhat of a flop. Cirrus needed its own 3D card to compete with the shifting tides and it had to be really fast. The company chose an unusual route to achieve
this, announcing in March 1996 that they’d be licensing the 3D core of the then upcoming
3DO M2 console. 3DO also exclusively licensed the architecture
of the console to Panasonic, stepping away from the business model established with their first console. Unfortunately, despite the impressive hardware,
by mid 1997, Panasonic decided to pull the plug on the project due to the cost of the
hardware however, due to high costs and the fierce competition they’d face. So, Cirrus Logic now had a console class 3D
core capable of drawing 100Mpixels and 500.000 polygons per second. These were pretty high-end numbers even by
1997. But there was only one small problem. The technology demo board from 3DO was too
big and expensive to manufacture as standalone consumer card. Cirrus needed a more integrated solution and
that’d be cheaper to produce. And thus in September 1996, the resulting product
was finally announced, Cirrus Logic’s first 3D chip was to be the Laguna3D. The 2D core was lifted of the GD5462 and once again made use of fast RDRAM modules. But, the texel rate was dropped to 50Mtexels
per second, so some sacrifices had to be made. Still, this was a chip equaling the performance
of Voodoo 1 cards, so compared to the competition it seemed to have the upper hand, at least
on paper. The first cards were shipped in March 1997
using the PCI bus. Pricing for 4MB models was set at a suspiciously
low 99$. Something was bound to be really wrong here. Why else would anyone sell a high-end chip
for low-end prices? In addition to that, the cancellation of 6MB
and 8MB variants must have raised some eyebrows as well, with only 4MB and 2MB versions hitting the market. Αn AGP version was also announced in August 1997, but by this point interest had dwindled and the chip ended up as a low-cost multimedia solution sold mostly by various small noname companies. We’ll look into what happened in just a
moment, but it’s fair to say the Laguna3D did not fare well in the market and Cirrus
Logic was forced to sell its entire graphics division to ISDCorp in October 1998 and instead
focused on audio chips along with device controllers. This decision ended up saving the company
in the long run, as unlike many of its many competitors at the time, Cirrus Logic survives
to this day. ISDCorp provided support and updated drivers for the graphics chips for the following 2 years until 2000. Marketing had, as it always does, very high
hopes for the Laguna3D. According to the company’s old webpage,
Microsoft was excited about its advanced acceleration features and Intel liked its early adoption
of the then brand new AGP bus. Also, Creative Labs were satisfied with its
outstanding performance. To achieve such performance, another technology
dubbed “TextureJet” was used. This was a small texture cache, in which the
chip stored the addresses for all textures to be used. But it only operated on AGP variants. Unlike previous episodes, we’ve finally
managed to have framerate numbers for Glide and 3Dfx cards thanks to thp’s glidefps
tool which you can find on VOGONS. Sadly, FPS counter only appears in the top left corner,
which means we had to re-arrange our screens a little bit to accommodate this change. So, as always, the reference card is once
again 3Dfx’s Voodoo 1 from Diamond with 4MB of memory. And because our test PC is sometimes too fast
for the Voodoo 1, we also used an NVIDIA Riva 128 card for Direct3D games with color and
texture issues that usually occur on an original Voodoo when used with CPUs that exceed 400MHz. Both cards will be competing against the AGP
version of Laguna3D from Chaintech for which we could not discern the core clock. According to the datasheets it’s supposed
to have a clock of up to 85MHz in 2D modes, but 3D clockspeeds remain a mystery. Also, memory could reach all the way up to
300MHz, but our card only goes up to 215MHz. We’ll be showing results results from a
PCI card from DSystems as well, with the memory clock on that one hanging at 229MHz. Drivers are localized, so you won’t be able
to read much unless you can read Czech :-). On the left side you can choose between speed
or quality of rendering. On the right side is an option that reads
“display list processing”. Another tab allows changing resolutions and
refresh rates for the monitor and the final one is for bus configurations. Powerstrip exposes largely the same options
plus some totally random memory clocks :-). These could only be detected properly by Hwinfo
thanks to the swift response of its author to our request. In conclusion, the drivers offer some basic
options, but nothing advanced or very useful for gaming. Final Reality shows that the Laguna3D has
almost all of the required features to run properly, as its only missing some advanced
Alpha Blending modes. These are important in more modern games as you will see later. FPS almost never go above 15 however, and figures under 10fps are common in more complex scenes. The missing Alpha Blending modes only create visible issues when text is displayed on screen. 3DMark99 had some driver issues and wasn’t
able to select a 640×480 resolution. In end, we managed to get it to run through
Batch Run, but the results were far from being impressive. Image quality is quite good with just one
small detail. Perspective Correction often isn’t working
properly as you will see on wall and ground textures soon enough. Tomb Raider 2 is fine and definitely playable. The only issues arise in more complex scenes,
where framerate can fall to just 14FPS. Forsaken looks perfect and produces a good
framerate, so long as not many explosions occur on-screen, at which point performance tanks. One of our new additions to the pool of games
to be tested is Star Wars: Shadows of the Empire, which warns of missing features and
Alpha blending modes. Despite that, the game still looks great and
is very playable. Warhammer Dark Omen works flawlessly even
on the highest details. It’s unfortunate that the framerate doesn’t
exceed the mid-teens often, but since this is strategy game and not fast action shooter
it’s less distracting here. Another addition to our test suite is Moto
Racer which appears to be playable, but we’d like a somewhat higher framerate. Missing blending modes cause black smoke to
appear and the broken perspective correction deforms road textures. We had to use the Riva 128 in this game, as
the Voodoo combined with our fast CPU changed the colors for every biker to magenta, giving
this more of an aesthetic vaporwave look! Croc runs fine, producing an almost locked
20FPS due to V-Sync, but once again hitting a stable 30 FPS would have been nice. Note that the blinking sky textures on the
side of the voodoo side are caused by the glidefps utility. Subculture looks nice and the framerate is
pretty decent, but the voodoo can be almost 3 times faster in some scenes. Unfortunately, MDK doesn’t play nice with
our framerate counters, so we can’t give you exact numbers, but the game renders correctly
and plays very smoothly which is what matters in the end. On first glance, Turok seems to render correctly,
however bright squares are visible around fire and blood effects indicating transparency
issues. FPS drops are common, but overall this is
a good showing for the card. The final game we’ve added to our tests
is Wing Commander Prophecy. FPS is limited to 30 by the game’s engine
and Laguna3D manages to hit it most of the time. But overall image quality is much lower, because
the game won’t allow us to use some effects such as Textured Space and the radar screen’s
transparency is broken. Over on the Voodoo side you can see how it’s
supposed to render, however we had to use the Direct3D renderer, because Glide caused
constant freezing on our setup. Incoming is not only too slow, but also displayed very wrong due to missing transparency modes. Carmageddon 2 suffers from a very low framerate. 10FPS just isn’t enough for a racing game. There are also visual issues like deformed
road textures thanks to the bugged perspective correction and lack of transparent smoke on the track. Shogo once again lacks visual features such
as fog and on medium details it’s very often below 10FPS even on a simple scene such as
the first level here, thus making the game unplayable. Wipeout looks the part, with the darker gamma
actually being closer to the intended look than the Voodoo here, however the framerate
is very low for a racer requiring twitch reflexes. Expendable looks perfect which is an unbelievable
sight after so many cards that broke this game. The framerate though is very low once again. We couldn’t get the framerate counter to
work on this unpatched version of Half-Life, but it’s not a big deal anyway, as the game
lacks almost all world textures. After patching the game to version 1.1.0.4,
textures are restored and we get our framerate counter back, which gives us the full story
and it ain’t pretty. The Laguna3D has another issue as well. You can’t use the game’s menus at all or
the engine flips out and switches into a crazy yellow blinking mode. Alien vs Predator suffers most from missing
transparency modes. These are used not only for smoke and lights,
but also for nightvision mode, which is now useless. This game runs poorly anyway. As we approach the end, it’s time for the
ultimate test. The lingering question is, Can it run Unreal? And yeah, it can, but with broken lightning
and transparencies. Besides the very dark and broken areas, you
can also expect low framerates and missing text, which means we can’t give you a detailed
performance report. I guess it’s time we reached for a conclusion. The Laguna3D has a bright side and that is
its performance. Unlike, most 1st gen cards it’s actually
able to run many early games at a resolution of 640×480 and produce very playable results. But that only applies for the AGP version. The PCI variant we tested was much slower
and ended very close to the S3 ViRGE/DX. The list of negatives is much longer. OpenGL is not supported in any shape or form,
so Quake engine based games are no go. Also, many games that came out in the following
years suffer a lot from the missing transparency modes which really hurts image quality. In addition, many racing titles exhibit the
broken perspective correction more clearly and the fluctuating framerate makes it hard
to play them. It is also safe to say that in more complex
games, the framerate leaves a lot to be desired. As such, the Laguna3D gets 5 stars out of
10 for its ability to play early 3D games sufficiently. Sadly, it’s not quite fast enough for more
modern titles and issues with transparency can often occur. There are much better and more common alternatives
from that generation such as ATi’s Rage PRO and of course the heavyweights, Nvidia’s
RIva 128 or 3DFX’s Voodoo 1 and even the Rush. That’s all for this episode! Once again, thank you for watching and now
as always you can enjoy 3DMark99’s demo.

Worst Game Graphics Cards – Cirrus Logic Laguna3D
Tagged on:                 

Leave a Reply

Your email address will not be published. Required fields are marked *