Banshee: Good or Bad? article by Michael Flemming.
I copied and pasted it here from his site
voodooextreme.com
Everyone and their kid brother should read this, I agree with it wholeheartedly.
-Tae
Banshee: Good, Bad or Ugly?
In a nutshell: Envelope, please... the answer is: "Good!" 3Dfx's Banshee chip looks
great and, in any sane world, it should do very well in its target market, system OEM's
(e.g. Dell, Compaq, etc.). 3Dfx had to dance a fine line between killing the cash cow
(Voodoo 2) and undershooting the competition performance-wise in the fight for the
OEM market: to their credit, they seem to have executed quite solidly, with only a little
wobble here and there. All in all, the Banshee is, in my opinion, a success for both the
company and the consumers.
What is it? The Banshee is 3Dfx's long-awaited single-chip 2D/3D solution. It is hoped
that the Banshee will enable 3Dfx to move into new markets (system OEM's), and will
provide people who want Voodoo 2-powered 3D graphics with a cheaper and more
complete (one board for all graphics needs) option than is currently available.
Speculation, concerns and hopes have run rampant for months as 3Dfx has been more
tight-lipped than the CIA: everyone wanted to know - just what would the Banshee be
like?
As of this morning at 12:01 a.m. we finally know. The reactions have been many and
varied, but, in my opinion, most of the negative press has been the result of ignorance,
bias or a misconception of the Banshee's design goals. I feel that, as a whole, 3Dfx
succeeded admirably in making the Banshee what they wanted it to be; an inexpensive,
one-stop solution for 2D and Voodoo-class 3D which should attract a whole lot of
system OEM interest. By doing this, 3Dfx stands not only to make a lot of money on
Banshee sales in new computers, but also to strengthen the presence and viability of the
company's proprietary (but very fast) 3D API, Glide.
These and many other rewards await 3Dfx if the Banshee is good enough. So, is it? Let's
see.
The Positive: There are a lot of good things about the Banshee, but most are rather
standard for 2D/3D boards, such as 3D acceleration in a window, support for various
video options, etc. Suffice it to say that the Banshee does everything it should to be an
upstanding and respectable 2D/3D card - in other words, the standard features are all
there. For the low down on the chip's basic functionality you're going to have to look
elsewhere: for the purposes of this article, I'm going to assume that you'd rather look at
what makes the Banshee stand out from the crowd, and/or from previous expectations.
First, and perhaps most importantly given the Banshee's target market (system OEM's),
the 2D portion of the chip is fantastic. In fact, indications are that the Banshee is the
fastest consumer 2D graphics board ever produced: it is within about 1% of being as fast
as is theoretically possible in current machines. This is a coup for 3Dfx, which had a
mediocre 2D performer in the Voodoo Rush and which most people didn't trust to go
outside of their realm of expertise (3D), much less supersede more-experienced 2D
companies like Matrox. In fact, this is more than a bit of an embarrassment for Matrox,
Number 9, etc. While many people have been dismissing the importance of the
Banshee's 2D performance on the grounds that games don't need it any more, I think
they are missing the meat of the matter: 2D is crucial to system OEM acceptance, and
that is the Banshee's true mission. The Banshee isn't meant to be a gamer's add-on
peripheral, but rather an everyday main graphics system which has the added capability
to play games at an outstanding clip. Having the world's fastest 2D is a huge step in this
direction.
Second, the 3D speed is also excellent. The chip manages 100 million pixels per second,
trilinear-filtered: on real world game benchmarks like the 3D GameGauge, the alpha
release of the Banshee and its drivers place second only to Voodoo 2 SLI - the nearest
current 2D/3D competitors (i740, Riva128) are about 1/2 as fast. In relation to the
announced competition (more on this below), the Banshee also fairs well for the most
part: only one chip (the Riva TNT) looks to completely out-class it, and that chip should
be in a different price category, and thus a different market. Moreover, 3Dfx has a
reputation for providing on it's speed claims (their worst gaffe being a 10% shortfall on
the Voodoo Rush), while the competitors' products routinely miss their advertised
performance by 50% or more. Finally, since the Banshee is a Voodoo-based chip, it is
able to run Glide versions of games, which add yet more speed in comparison to other
companies' 3D chips.
Third, speaking of graphics API's, the Banshee, along with the other 3Dfx products,
enjoys the singular position of supporting every major API - OpenGL, Direct 3D, and
3Dfx's own Glide. The value of this level of compatibility is difficult to overstate - no one
(especially not casual gaming purchasers of pre-built OEM systems) wants to find out
that the game they want won't run, or won't run well on their machine. From the OEM's
point of view, compatibility (and thus smoothness and ease of use) is a very big deal.
Fourth, contrary to at least one analyst's predictions, the Banshee has a 128-bit memory
bus. This fast, wide bus provides enough bandwidth to allow the chip to sustain its peak
performance with pretty much all rendering features turned on. The memory pool that the
bus serves can contain up to 16 MB of industry standard (and thus very cheap) SDRAM
or SGRAM.
Fifth, as a result of the large potential memory pool, the Banshee can do 3D at very high
resolutions (more important for 3D-in-a-window than for games, currently), and can hold
a very large amount of textures locally. The Banshee's reliance on local memory rather
than AGP texturing has been roundly criticized by some. In my opinion, this is downright
silly - it's a big plus that the Banshee doesn't use AGP to do direct texturing from main
memory: there is no possible way that AGP (1x, 2x, or even 4x) could keep the
Banshee's hunger for textures satisfied (see AGP: Myth & Truth).
Sixth, contrary to the concerns of many, the Banshee has a very high speed RAMDAC
(230-250 MHz), which allows it to produce rich, saturated colors, and to display 24-bit
true-color screens at very high resolutions (up to 1920 x 1440) at 70+ Hz.
Seventh, along with its Voodoo siblings the Banshee shares a consumer brand name that
is the envy of it's competitors. Similarly, the mindshare of 3Dfx's 3D hardware is
extraordinary among 3D game developers, giving the Banshee another leg up.
Finally, the Banshee will apparently be upgraded in the Fall, gaining 25% in speed and
dropping in cost.
The Negative: There's not a lot negative about the board from my point of view. The
only things that disappointed me were finding out that they were going to wait until Fall to
move to a 125 MHz clock speed (the 25% gain mentioned a moment ago), and that the
price differentials for the different memory configurations seem usurious to me. In my
opinion, it would have been better to come out with a 125 MHz version first and maybe
moved to 150 MHz in the Fall. However, it's hardly the end of the world. As for the
price vs memory configuration issue, that's just strange and may be sorted out by board
manufacturer competition: basically, while 100 MHz SDRAM retails for less than
$1/MB, the cost to go from a 4 MB Banshee ($125) to a 16 MB Banshee ($225) is
more like $8/MB. While this is outrageous, there may be a good explanation (I'm waiting
on a newsgroup reply from 3Dfx), and, as I just said, there is always the hope that
competition will bring the price differential into line with reality.
All in all, there's just not much wrong with the chip, given its target market. People are
complaining that it doesn't have two texturing engines like the Voodoo 2 and the Riva
TNT, that it won't be the fastest chip for the remainder of the year, etc. I say, so what?
The Banshee isn't and wasn't meant to be the absolute 3D high-performance leader - the
Voodoo 2 is 3Dfx's entry in that race. The Banshee, as I've said above, is supposed to
be a low-cost mainstream system OEM part for the general populace and casual gamers.
In that light, these `flaws' are no big deal.
Finally, a lot of noise has been made about the lack of AGP 2x and AGP texturing. As I
have mentioned elsewhere, this is a total non-issue: AGP 2x is too slow by a country mile
(or 5) to serve as a primary texturing system and AGP 1x is plenty fast for the other
duties the system-to-board bus must perform. In any case, long before the Banshee
family will actually need AGP 2x, that feature will be added (in the Fall upgrade
mentioned above). The only semi-valid concern I've seen raised about the lack of AGP
2x support in the Banshee is that OEM's might be unhappy that they won't be able to put
that feature on their bullet lists. However, I suspect that this is also no big deal, as Joe or
Jane Consumer is almost certain to be more impressed by "fastest 2D in the world,"
"3Dfx inside," "works with more games than anyone else," etc. than "has AGP 2x." I
mean, please: Banshee boards can rightly claim to have "AGP" - how many people are
going to know or care that there's no "2x" tacked on?
How does it stack up? In comparison to existing products, the Banshee Rocks, Kicks
butt, Takes out the Trash, etc. Pick your favorite descriptive phrase :-). As mentioned
above, the Banshee performs twice as well as it's current competitors in actual games
(3D GameGauge). The situation with the announced competition, however, is more
complex.
Even with many of the upcoming products, the Banshee does extremely well. The
much-ballyhooed Matrox G200, for instance, will be slower than the Banshee in nearly
every particular, including, it would seem, Matrox's forte - 2D performance. The S3
Savage 3D should be slightly faster if it lives up to its paper specs, but it is beginning to
look like it won't: my own experience at the Computer Game Developer's Conference
(CGDC) and Tom of Tom's Hardware's more recent experiences indicate that the
Savage 3D may well fall short of it's specs. NEC/Videologic's PowerVR Second
Generation is a big question mark: it could be considerably faster than the Banshee in
most cases if it lives up to its PR and hype. However, again, it looks like they're having
problems with the chip - it's not up to speed and may well be late to market whether or
not they manage to get it up to its promised specs.
The biggest apparent threat is the Riva TNT. On paper, this chip should blow the
Banshee out of the water in 3D performance. In fact, the TNT's 3D should blow
everything but Voodoo 2 SLI out of the water. However, there are two important
ameliorating factors in the Banshee's favor (above and beyond the branding, and
compatibility mentioned above): cost, and the familiar specter of real-world performance
shortfalls. Cost may be the most important difference - the cheapest announced version
of a TNT board to date is set to retail for $299, as compared to $125/$150 for the
Banshee. This is a big difference for the system OEM's, and, as a result, the TNT is less
a true competitor for the Banshee than it is for the Voodoo 2: the Banshee and the TNT
are currently aimed at fundamentally different markets. As for the TNT failing to meet its
paper specs, that also seems likely. First, because nVidia has failed to meet it's predicted
performance before (by around 50% with the Riva 128). Second, because the TNT
needs 200 MHz SGRAM to even hope to maintain peak performance, but that type of
memory isn't even formally announced yet, to my knowledge. I suspect that the TNT was
designed to be performance capped with current memory and to speed up as faster
memory became available, but that marketing decided to claim the future numbers as the
initial ones for PR purposes. Even if 200 MHz SGRAM is somehow available by Fall, no
one will be using it but nVidia, meaning that the TNT will cost even more in its high-speed
configuration. Finally, nVidia is currently hurting fiscally, as its main profit center, the Riva
128, is being squeezed hard by the Intel i740, and the company's planned IPO is late and
looking more and more distant. If nVidia can't secure the capital they need in some other
way, they may have serious problems finishing the TNT, ramping its production up, etc.
Similarly, they may be handicapped in a price war if they have little in the way of cash
reserves and need to maintain high margins in order to survive.
The Bottom Line: Basically, the Banshee looks like it will perform about as well as or
better than everything in its price range and target market. The Banshee's 2D and 3D
performance and features, together with the 3Dfx name, unequaled software
compatibility, huge developer mindshare, and the performance boost of access to Glide
versions of games (which often run as much as 30% faster than the Direct 3D versions),
should make it a real success. Banshee is right on target: Good going, 3Dfx!
|
