
Original Link: https://www.anandtech.com/show/187
Remaining fairly quiet throughout the 1998 hardware rollout season, Guillemot International hasn't had any major successes in any PC peripheral related field since the introduction of their top of the line Home Studio Pro 64 sound card which put the original Sound Blaster AWE64 to shame. | |
With history prone to repeat itself
eventually, and Guillemot well overdue for a comeback, it seemed almost fitting that
Guillemot should be one of the first to jump on the next-generation 2D/3D accelerator
bandwagon with one of the first video accelerators based on the 3Dfx Banshee
Chipset. Almost predestined to follow in the footsteps of the Voodoo2 chipset, the Banshee will bring even more truth to the generally vague statement that all 3Dfx accelerators are created equal, meaning a Voodoo2 is a Voodoo2, and a Banshee is a Banshee, regardless of manufacturer. With that in mind, let's take a look at the mold for 3Dfx's newest chipset, will it salvage 3Dfx's name in a market that is quickly falling into the hands of the competitors? |
Banshee vs Voodoo2
Before getting into the specifics of the card itself, we have to separate the top two contenders from the 3D proprietors themselves, 3Dfx Interactive. The primary difference between the two chipsets is definitely the obvious, the Voodoo2 is a 3D-only solution while the Banshee is an integrated 2D/3D combination product.
Presented in a single chip package, the Banshee contains essentially the same 3D texture processing capabilities as the Voodoo2 with one major disadvantage in order to decrease cost. The Banshee only has a single texture processor (known in the Voodoo2 community as a Texelfx processor) meaning it requires two passes to render multi-textured environments.
If you look at 3D rendering as painting a wall, a single coat of paint can easily be accomplished by virtually any brush, while that same brush will require two strokes to place two separate coats of paint on the wall (1 texture processor). Now imagine a brush capable of placing two coats of paint on a wall in a single pass (2 texture processors). By using the latter type of brush you are essentially doubling your productivity. In 3D gaming and rendering situations the application of such a technique is a bit more complex, yet it follows the same basic principle. If a wall in a game, such as Unreal, happens to have a texture placed on it, such as a brick texture, followed by another layer, say a reflection from a nearby fire, you basically have two textures on that one surface.
A Voodoo2, or any other chipset which has two separate texture processors can render that surface in a single pass (as in a single pass of a brush from the example above) while a Banshee, or any other chipset which only has a single texture processor, must make two passes (as in two strokes of a brush) to render the entire surface.
The Banshee boasts the same pixel processing unit which is found on all Voodoo2 boards, and therefore does retain some of the power of its bigger brother. In spite of its semi-crippled nature, the Banshee can surpass even the Voodoo2 in terms of performance when placed in the proper situation; at the same time, it can also be put to shame in a slightly different one. How can this be?
Well, in order to compensate for the lack of a second texture unit, 3Dfx clocked the Banshee at a full 100MHz clock, producing a fill rate of 100 million pixels per second compared to the Voodoo2's default fill rate of 90MP/s. In situations where only a single layered texture is used, the Banshee will outperform a Voodoo2, however in multi-textured situations, the Banshee will be shadowed by the raw power of the Voodoo2's second texture processor.
Also, unlike the Voodoo2, the Banshee was catered to be an AGP solution rather than a PCI slot-hog as its counterpart turned out to be. Unfortunately, this decision came as a marketing ploy more than anything else for 3Dfx. Since AGP is a newer technology, the market immediately assumes that it is inherently better and worth the extra attention, however in the case of the Banshee, this is definitely not true. The Banshee's implementation of the AGP specification only allows it to take advantage of what is known as the 1X AGP Bus Transfer Mode, operating at 66MHz. Not only does it operate at the AGP 1X specification, it also doesn't allow for the transfer of textures over the AGP bus due to its lack of support for AGP side-banding (the transfer of data on both the rising and falling edges of the AGP bus, i.e. making use of the entire width of a road rather than the central portion exclusively). This makes the AGP capabilities of the Banshee chipset virtually useless to the gamer from a performance perspective, virtually bridging the gap between AGP and PCI Banshee based cards.
The Banshee, as mentioned earlier, is a completely integrated 2D/3D solution in a single package. This is an obvious deferral from the standard 3Dfx methodology of graphics accelerator production which normally calls for a separate 3D accelerator. For those of you that have been soured on 3Dfx's 2D/3D combo cards from the horrid days of the VoodooRush, the Banshee isn't going to give you the same problems you experienced with the Rush chipset. It fully supports Direct3D as well as Glide, meaning it will be able to run all of your current games that will work on 3Dfx based products virtually flawlessly. The inclusion of 16MB of SDRAM or SGRAM on board allows for 3D resolutions up to 1024 x 768, with 16-bit Z-Buffering (no 32-bit support, meaning you get the same image quality as the Voodoo2). The 2D performance is pretty much on par with that of the nVidia TNT Chipset, still keeping the generous gap between itself and the 2D monster, the Matrox G200. This performance is courtesy of the 128-bit 2D core that is a part of the Banshee chipset, unlike the original VoodooRush which contained a separate, external, 2D core which wasn't of the absolute best quality. While the 2D image quality could definitely be improved, for most users with 17" or smaller monitors, the Banshee shouldn't be a huge disappointment. If you are used to a Matrox or maybe even a high end Number 9 2D accelerator on a 17" monitor you may want to try the Banshee out on your system before making any final judgements as to the course of your purchasing decision.
With that said and done, and with the Banshee formally introduced, let's get to the very first review of a 3Dfx Banshee based card on AnandTech, the PCI Guillemot Maxi Gamer Phoenix Banshee.
Graphics
Engine
Memory Technology
Board Features
3D Features
|
System
Requirements
OS Support
API Support
|
Removing the Phoenix from its extremely eye catching package revealed a fairly tightly packed cardboard box containing the board itself, along with the installation guide, an installation CD-ROM, as well as the game bundle among the normal warranty card mess you find in all products of this nature.
Exploring the board itself, the Phoenix turns out to be much like any other next-generation combo board either out on the market now or soon to be released. The slot-length board fit snugly into the PCI slot of AnandTech's ABIT BX6 Pentium II test system and provided absolutely no problems at start-up in terms of the physical hardware installation. | ![]() |
The test system was purposely prepared with a previous set of drivers from various other 3rd party video adapter manufacturers in order to test compatibility in carelessly installed situations. The Phoenix didn't prove to fly high in this case and caused numerous crashes and conflicts resulting in a horrendous 640 x 480 x 16 bit color resolution until the previous drivers were removed in safe mode and Windows reverted to the standard VGA graphics driver. After this, the Guillemot driver installation went without a hitch.
Guillemot included a CDR outfitted with their latest drivers at the time of the review, before the card begins shipping you can expect a bit of a maturing process for the drivers as the set AnandTech tested with had no native OpenGL support rendering the Quake2 and SiN benchmarks useless as the test system simply wouldn't run the tests. Luckily the drivers seemed to work fine with Glide implementations under Unreal as well as Direct3D games such as Forsaken. Currently released Banshee cards such as those from companies like Quantum3D do have operational OpenGL support so it seems as if the problem with the Guillemot card AnandTech tested can easily be fixed with a driver upgrade.
Guillemot's bundled display properties utility did very little for the Banshee in terms of tweaking it for the greatest performance. No overclocking slider bars or even a Enable/Disable V-Sync checkbox was provided for in the newly added configuration tab in the Display Properties Control Panel. While the Banshee already gets hot enough, the ability to overclock the RAM on the board right out of the box would've been nice, however it looks like it's over to AGN3D or Voodooextreme for the latest tweak utilities for this board.
Super7 users will want to wait for updated drivers as well, the Maxi Gamer Phoenix wasn't too Super7 friendly as the card consistently crashed randomly during all 3D benchmarks, regardless of API. Further testing will prove whether the problems experienced were motherboard, chipset, or video card specific and you can expect the first word on any developments on AnandTech's frontpage.
The initial installation and configuration of the card wasn't too big of an ordeal since Guillemot's multi-lingual User Manual helped sort out the installation process. Unfortunately, only after exploring the readme file bundled with the drivers were many of the card's anomalies explained. The initial install of the graphics drivers had to be performed over a Standard VGA driver as mentioned earlier, then switched over to the Maxi Gamer Phoenix drivers at a resolution/color depth other than 640 x 480 x 16-bit color. Guillemot chose to package the board with both Windows 95 and 98 drivers, however NT drivers should be shipping closer to the card's public release.
The card was packaged with a copy of Tonic Trouble, which is more of a younger children's title rather than your run of the mill gory first person shooter that you're used to, and don't worry, you won't have any more copies of Forsaken lying around the house after picking up this card. Guillemot wisely chose to replace the standard filler title, Forsaken, with a more appropriate, and in this reviewer's opinion, much better, title: Half-Life - Day One. If you are a Quake2 fan, prepare to meet your new best friend, Sierra's Half-Life. While the bundled version of Half-Life isn't the full release, it'll still provide you with some awesome gameplay until the final version becomes available. Once Guillemot updates their drivers, Half-Life will definitely be the cherry on top of this ferocious pie.
The Slot-1 Pentium II Test System AnandTech used was configured as follows:
An Intel Celeron 300A, an Intel Pentium II 400, & an Intel Pentium II 450 on an ABIT BX6 Motherboard
64MB Mushkin SEC PC100 SDRAM
Western Digital 5.1GB Ultra ATA Hard Drive
AOpen 32X IDE CD-ROM Drive
Windows 98 with all of the latest patches/drivers installed
The benchmark suite consisted of the following full version game titles
Forsaken - Running the Nuke Demo
Quake 2 v3.17 using demo1.dm2 and Brett "3 Fingers" Jacobs Crusher.dm2 demo
SiN Demo2 using Brett "3 Fingers" Jacobs rocket.dm2 demo
Unreal using Lothar's FPSTimeDemo test (run 10 times for each test)
Ziff Davis' Winbench 98 was used to test 2D performance at 1024 x 768 x 16-bit color.
VSYNC was disabled during AnandTech's tests (VSYNC is the synchronization of all buffer swaps to the refresh rate of your monitor, theoretically limiting the attainable frame rate by the refresh rate your monitor is set at. Disabling it will improve performance but may degrade visual quality by introducing "tearing")
The drivers AnandTech tested with had no native OpenGL support rendering the Quake2 and SiN benchmarks useless as the test system simply wouldn't run the tests in any OpenGL/Glide modes.
For the in-depth gaming performance tests Brett "3 Fingers" Jacobs Crusher.dm2 demo was used to simulate the worst case scenario in terms of Quake 2 performance, the point at which your frame rate will rarely drop any further. In contrast, the demo1.dm2 demo was used to simulate the ideal situation in terms of Quake 2 performance, the average high point for your frame rate in normal play. The range covered by the two benchmarks can be interpreted as the range in which you can expect average frame rates during gameplay.
Video Accelerator Comparison - Unreal - Glide Performance |
|
FPSTimeDemo |
640 x 480 |
- | fps |
Intel Pentium II 266 | 26.54 |
Intel Pentium II 400 | 29.35 |
Intel Celeron 300 (no L2 cache) | 27.87 |
Intel Celeron 300A (128KB L2 cache) | 29.23 |
Intel Celeron 450 (no L2 cache) | 29.46 |
Intel Celeron 450A (128KB L2 cache) | 29.44 |
FPSTimeDemo |
800 x 600 |
- | fps |
Intel Pentium II 266 | 22.14 |
Intel Pentium II 400 | 23.40 |
Intel Celeron 300 (no L2 cache) | 22.94 |
Intel Celeron 300A (128KB L2 cache) | 23.35 |
Intel Celeron 450 (no L2 cache) | 23.43 |
Intel Celeron 450A (128KB L2 cache) | 23.46 |
FPSTimeDemo |
1024 x 768 |
- | fps |
Intel Pentium II 266 | 17.44 |
Intel Pentium II 400 | 17.79 |
Intel Celeron 300 (no L2 cache) | 17.69 |
Intel Celeron 300A (128KB L2 cache) | 17.68 |
Intel Celeron 450 (no L2 cache) | 17.86 |
Intel Celeron 450A (128KB L2 cache) | 17.96 |
The presence of L2 cache and the raw performance of the CPU on the test system didn't seem to truly affect the performance of the Banshee all too much. For the most part, the Unreal benchmarks resided in the 26 - 29 fps range at 640 x 480, 22 - 23 fps at 800 x 600, and they all seemed to stay at around 17 fps at 1024 x 768. Translation? For high end systems, with faster processors, the Banshee isn't the ideal chipset to aim for as it doesn't scale with processor speed too well. However if you have a lower end system, such as a first generation Pentium II (233 - 300), the Banshee, especially the Maxi Gamer Phoenix, is probably what you've been waiting for.
Video Accelerator Comparison - Forsaken - Direct3D Performance |
|
FPSTimeDemo |
640 x 480 |
- | fps |
Intel Pentium II 266 | 71.15 |
Intel Pentium II 400 | 73.3 |
Intel Celeron 300 (no L2 cache) | 73.62 |
Intel Celeron 300A (128KB L2 cache) | 73.03 |
Intel Celeron 450 (no L2 cache) | 72.69 |
Intel Celeron 450A (128KB L2 cache) | 73.48 |
FPSTimeDemo |
800 x 600 |
- | fps |
Intel Pentium II 266 | 69.16 |
Intel Pentium II 400 | 71.12 |
Intel Celeron 300 (no L2 cache) | 71.24 |
Intel Celeron 300A (128KB L2 cache) | 71.24 |
Intel Celeron 450 (no L2 cache) | 70.68 |
Intel Celeron 450A (128KB L2 cache) | 71.24 |
FPSTimeDemo |
1024 x 768 |
- | fps |
Intel Pentium II 266 | 60.17 |
Intel Pentium II 400 | 60.7 |
Intel Celeron 300 (no L2 cache) | 60.99 |
Intel Celeron 300A (128KB L2 cache) | 61.01 |
Intel Celeron 450 (no L2 cache) | 60.23 |
Intel Celeron 450A (128KB L2 cache) | 60.99 |
As a role model for all future Banshee based video cards Guillemot did an excellent job of putting the chipset to good use, you'll find no useless features, filler software titles, or pointless driver utilities with the Maxi Gamer Phoenix. You'll get exactly what you pay for with this card, the $89.99 final price (after a $20 manufacturer's mail in rebate) buys you what will come to be considered the average for a 2D/3D accelerator combo, with at least one title that'll keep you drooling for a while.
If your budget dictates that $100 - $120 is the absolute most you can spend on a 2D/3D graphics accelerator, the Banshee is the ideal choice for you. It combines the power of the Voodoo2 with the price of an i740 all in a single, low cost, average performance package.
For users with slower processors (including Super7 users) such as the Pentium II 233, 266, and 300, the Banshee is probably the best overall choice for the price. Anything above a 300 is more worthy of a Savage3D or a Riva TNT based accelerator simply because of the processor scalability those two chipsets (especially the TNT) boast over the Banshee.
If 2D/3D image quality is of the utmost importance (as in you don't care too much about performance), you're probably better off getting a Matrox G200 which falls into virtually the same price range with unparalleled sharp 2D image quality not to mention its top notch 3D image quality as well.
While those recommendations are speaking in general terms, they will apply to the Phoenix once Guillemot revises their drivers. Keep in mind that the card tested by AnandTech didn't ship with the final drivers, and improvements will come with time. Check back on the AnandTech frontpage for the latest updates on the issue as well as Guillemot's home page at www.guillemot.com for all driver updates once the cards starts shipping. With a company like Guillemot, it is highly doubtful that they would ship a card without proper driver support in mass quantities.
Overall, a strong recommendation for average 2D/3D performance goes out to Guillemot for their Banshee based product, while the card isn't for high end users, it still packs the explosive punch necessary to take down the 3D performance barrier without any TNT.