Table of Contents
Introduction
Today we'll shed some light into something I also wanted to test for myself since forever. As I already discovered in a previous test here, there is a clear performance improvement between 1.01.x und 1.03.x driver versions. So why don't we have a look at how performance developped over time and driver versions and also compare the official 3dfx drivers to later 3rd party offerings?
I collected all drivers I found available for the 3dfx Banshee and ran a couple of Benchmarks. These are devided into two seperate parts - part one covers all of the drivers, which are compared in three benchmarks that work with even the earliest driver. Part two covers the 3dfx drivers from 1.03.00 and up, as well as the 3rd party drivers. For this test, I used a lot more demanding benchmarks from games that were released around the millenium.
So, let's see if we can find the one and only definitive Banshee driver!
The Test System
|
Asus P3B-F Rev. 1.04 (Intel i440BX) Intel Pentium III 850 MHz Leadtek Winfast 3D S310 Infineon 256MB PC133 CL3 @ PC100 CL2 2-2-X 3Com Fast Etherlink XL PCI Creative Soundblaster 128 PCI Western Digital WD400BB (ATA100 / 40GB / 7.200RPM) |
For the OS, I used Windows 98 Second Edition with DirectX 8.1 as well as an inofficial Service Pack installed. For each card, I restored a pre-configured system-image before testing.
Drivers Included In The Comparison
One thing up front: in the OpenGL tests, I made use of the 3dfx MiniGL V1.49 for all of the drivers tested. During the process, I also measured performance of the older drivers with the then-most-recent MiniGL driver, but there is no point in using older MiniGL drivers AT ALL. Older versions only bring compatibility issues and lower performance. Problems can be non-working resolutions, crashes or games not working at all. I am aware of the fact, that version 1.46 is recommended for the Super Socket 7 platform, so this *might* be an exeception, but in any other case - just use the latest MiniGL. It works flawless in combination with even the oldest driver.
Another thing to talk about is the memory clock, that is applied by each driver. This can differ by A LOT, so I included this information in the chart. For the benchmarks, I manually set the memory clocks to 100 MHz, so that we can compare apples with apples. Note that the clocks written in the charts got applied by the driver to a Banshee equipped with SG-RAM - I didn't test if the driver applies the same clocks to Banshees with SD-RAM. Memory clocks of 120 MHz (which the 1.03.04 driver applies) might be a little too much for cards with 100 MHz SD-RAM chips..
| Driver | Release Date | Core Clock | Mem Clock | VSync Options |
| 3dfx 1.00.00 | 1998-08 | 100 MHz | 115 MHz | no driver options |
| 3dfx 1.01.02 | 1998-11 | 100 MHz | 110 MHz | no driver options |
| 3dfx 1.01.03 | 1998-12 | 100 MHz | 110 MHz | no driver options |
| 3dfx 1.02.03 | 1999-02 | 100 MHz | 100 MHz | no driver options |
| 3dfx 1.03.00 | 1999-10 | 100 MHz | 110 MHz | no driver options |
| 3dfx 1.03.04 | 1999-11 | 100 MHz | 120 MHz | no driver options |
| 3dfx 1.04.00 | 2000-02 | 100 MHz | 100 MHz | no driver options |
| BansheeDriver 1.03.04 (modded) | 2000-08 | 100 MHz | 110 MHz | 3dfx tools |
| BansheeDriver 1.04.00k (modded) | 2001-02 | 100 MHz | 110 MHz | 3dfx tools |
| Xpentor-xgamers3 | 2001-07 | 100 MHz | 100 MHz | no driver options |
| X-Mojomotion 1.0 | 2003-04 | 100 MHz | 110 MHz | 3dfx tools |
| BansheeTeam Evolution V0.9 | 2003-09 | 100 MHz | 100 MHz | no driver options |
| Raziel Evolution v1.00.00b | 2004-05 | 100 MHz | 100 MHz | no driver options |
The Benchmarks, Part 1
This first set of benchmarks consists of the PCPlayer D3D Benchmark to cover Direct3D performance, Unreal for the Glide part and Quake 2 for OpenGL.
All benchmarks were performed with VSync turned off via VControl. No other driver options were changed. As mentioned before, clocks were manually set to a conservative 100 / 100 MHz core/mem, regardless of which clocks the driver itself applies.
PCPlayer D3D Benchmark
This benchmark was released in mid-1997 by the german gaming magazine "PC Player" and was one of the earliest wide-spread Direct3D performance test. Even tomshardware used it back in the day. It can be run on any Direct3D compatible graphics card and thus is no challenge for a second-gen card as the Voodoo Banshee.
Version 2.10 | Settings used:
The first thing that meets the eye is certainly the huge difference in performance of the 1.01.xx drivers to all the other versions. In my first run, the 1.00.00 driver showed similarly bad performance, but after I reimaged the system and made a fresh installl of the driver, it performed in the range of the other drivers (as can be seen in the charts). I also did this for both of the 1.01.xx drivers, but performance did not change.
This is definitely no VSync issue, but presumably a driver bug. It's also possible that an older DirectX version would fix this issue.
Other than that, there are definitely differences in performance across the drivers, but nothing major. In general, older drivers can be slightly faster in older D3D applications (exept for maybe the 1.00.xx and 1.01.xx versions, which seem a bit buggy, at least with DirectX 8.1 installed). Every newer official 3dfx driver loses a little bit of performance; 3rd party drivers are also slower - which makes sense, since the are based on the 1.03.xx and 1.04.00 drivers. The exception here are the Evolution drivers, which both can be found in the top four.
At the higher resolution, the performance difference between the drivers becomes very small across the board. Again, the 1.01.xx drivers show significantly lower framerates.
Unreal: Castle Flyby Timedemo
The mighty Unreal was the Crysis of the nineties. Released in May of 1998, it was THE game to showcase the power of the than-new 3dfx Voodoo 2 (SLI) cards. Unreal made use of Multitexturing, was the first game to include Detail Textures, featured huge levels, a lot of effects and brought every system to its knees. Unreal must not be missed in a classic benchmark comparison and is used here to measuire the Glide performance of the drivers.
Version 2.25f | Settings used:
Ignoring the Evolution drivers, the results are divided into two groups. The top group includes the official 1.03.xx, 1.04.00 and all the third party drivers (excluding the Evolution ones). Glide performance definitely improved significantly with the introduction of the 1.03.xx versions. All older official 3dfx drivers up to and including 1.02.03 fall into the second group, offering around 10% lower overall performance than the newer drivers.
The Evolution drivers are a different story, however. The only thing that the BansheeTeam Evolution V0.9 driver did (with the exception of the PCPlayer D3D Benchmark) was crashing. Not a single other Benchmark completed or sometimes even started with it. So prepare for more zeros in the upcoming charts.
The Raziel Evolution driver was little better. Not only is it the slowest driver in this comparison. In reality, the framerate was unsteady and choppy and felt much worse than the graph can tell you.
At the 1024x768 resolution, we can still see the two groups building, but there are two differences in detail.
Starting from the bottom, the Raziel Evolution driver is still in last place, but the numbers suggest that it has at least caught up with the early reference drivers. In reality, however, gameplay feels much worse with this driver. It is just as unsteady as at the lower resolution, just with even less FPS.
The second thing to talk about is the X-Mojomotion driver. While all the other drivers in the top group are very close in performance, this one falls back by around 6-7 %. An interesting result for sure, and something worth keeping an eye on in the further tests.
Quake 2: demo1.dm2 timedemo
No 3D benchmarking suite covering 3D graphics up to the millennium is complete without Quake 2. Its engine was frequently used in many popular games of the late nineties and reveals the performance in games using a more primitive 3D engine. The demo1.dm2 timedemo used here is a lot less CPU dependant than the famous crusher.dm2 timedemo, so that the performance is limited by the graphics card.
Version 3.20 | Settings used:
The top performing driver is the X-Mojomotion with an 1.5+ FPS lead over the chasing pack, which again includes all 3rd party drivers minus the Evolution ones, as well as the official 3dfx drivers with version 1.03.00 and above.
The Raziel Evolution driver shows only marginally better performance than the early reference drivers, but at least gameplay feels normal and corresponds to the numbers shown in the graph.
While the top six drivers show very similar performance, both of the modded "BansheeDriver" packages perform closer to 30 FPS, while the Xpentor-xgamers3, was well as the official drivers with version 1.03.00 and up are a little closer to 29 FPS.
Following these is the X-Mojomotion at around 28 FPS and the Raziel Evolution at just over 27 FPS. Reference drivers with version 1.02.03 and below closely miss the 25 FPS mark.
The Benchmarks, Part 2
For the second set of benchmarks, only more recent driver versions from the official 1.03.00 driver and newer were included. As could be seen in the first three benchmarks, the older reference drivers provide no advantage. On top of that, compatibility issues with newer games can be expected with these.
As for the games included, we go with a very classic set of late nineties benchmarks, featuring Expendable for Direct3D, Unreal Tournament for Glide and Quake 3 Arena for Open GL. These are all mid- to late 1999 released games, which use more advanced engines than found in part 1.
All benchmarks were performed with VSync turned off via VControl. No other driver options were changed. As mentioned before, clocks were manually set to a conservative 100 / 100 MHz core/mem, regardless of which clocks the driver itself applies.
Expendable: Integrated Benchmark
Released in 1999, Expandable showcased the capabilities of Rage Software's proprietary 3D engine with large, highly detailed environments, dynamic lighting and impressive real-time effects that is primarily heavy on the CPU, but also requires a capable graphics card. Although it never achieved mainstream success, it still was a popular benchmark around the millenium.
Version 1.0.0.1 | Settings used:
We are seeing very close performance numbers across the board - first and last place are only around 1.5 FPS apart. Nevertheless, it is noteworthy that the Raziel Evolution driver leads by almost 1 FPS.
At the 1024x768 resolution, results are even closer. The X-Mojomotion and the newer out of the "BansheeDriver" packages take a slight lead of around 0.5 FPS over the rest of the pack.
Unreal Tournament: Benchmark.dem timedemo
Released in late 1999, Unreal Tournament transformed the Unreal Engine into one of the defining multiplayer shooters of its era. Supporting Glide, Direct3D and OpenGL, it became a popular benchmark for late-1990s and early-2000s hardware. While the game is notoriously known for it's hunger for CPU power, the benchmark.dem timedemo is pretty GPU-limited and can be seen as a worst-case scenario benchmark for UT99.
Version 4.51 | Settings used:
Performance is about identical between all drivers. The exception is once more the Raziel Evolution driver, which shows similar problems as in Unreal. Its choppy framerate makes the game completely unplayable und makes it feel even worse than the numbers in the graph suggest.
Looking at the graph, we can see a familiar pattern: the X-Mojomotion driver falls behind all third-party and newer 3dfx reference drivers in the Glide API when rendering at a resolution of 1024x768.
The problems of the Raziel driver persist, so it takes the last place. All the remaining drivers perform on basically the same level.
Quake 3 Arena: four timedemo
Released just a few days after UT99 in early December 1999, Quake III Arena introduced id Software's new id Tech 3 engine, which quickly became one of the most influential 3D engines of its era. A major step forward from the Quake 2 engine, it brought a much more advanced rendering pipeline with features such as curved surfaces, higher-detail environments and improved lighting effects. It scales exceptionally well across a wide range of hardware, making Quake III an ideal benchmark for systems from the turn of the millennium.
Version 1.32 | Settings used:
In Quake III Arena, the time for third-party drivers has finally come. Both the Raziel Evolution and the X-Mojomotion driver in particular show a massive improvement in performance over the 3dfx reference drivers. Both of these drivers make use of newer OpenGL and Glide cores and thus are a lot faster than earlier 3rd party drivers, that "only" make use of optimizations and remixed driver files.
That being said, the "BansheeDriver" and Xpentor drivers are still around 10 % ahead of the 3dfx reference drivers. And this is no wonder, since the only reference driver to be published after Quake 3 Arena was released, is version 1.04.00. The official 1.03.00 driver can't run Quake 3 Arena at all and version 1.03.04 (while running the benchmark flawlessly and even slightliy faster than version 1.04.00) has several reported bugs in combination with Quake III Arena. This includes problems with fog rendering or the framerate dropping to near zero when Alpha Blending effects are used.
We can see the exact same picture at the higher resolution, only the framerates are of course notably lower. The X-Mojomotion achieves an impressive plus of close to 25% in performance compared to the reference drivers, but still, 26 FPS are not exactly playable for a fast shooter. Nevertheless, this driver may be the only one to easily achieve more than 30 FPS at the 1024x768 resolution in combination with faster Banshee cards like the Diamond Monster Fusion or Gigabyte's GA-630.
Conclusion
The first thing to summarise is that you should not consider any reference drivers before version 1.03.xx when looking for the best driver for your Banshee. While they provide no advantage over more mature drivers, they can certainly cause issues with more recent games.
Now, to recommmend the best overall driver, we need to understand that the optimal rendering resolution for a Banshee is 800x600. At this resolution, the card provides enough performance for almost any game from that era, especially if it isn't as conservatively clocked as here, in the benchmarks above. With this in mind, there can only be one recommendation: the X-Mojomotion 1.0. It is the fastest or second-fastest driver in every run at 800x600 and the fastest driver in Quake 3 Arena at both resolutions. This also makes it the best option for more "recent" games using the id Tech 3 engine.
It is slightly slower at 1024x768 in some games, but it will still provide good performance for older games that can be run at this resolution. Another plus for some users might also be the inclusion of the 3dfx tools, providing easy access to driver settings without the need to install a 3rd party app.
Also worth considering is the BansheeDriver 1.04.00k — it is the fastest driver at 1024x768 in Quake 2, Unreal, and UT99, and it also provides very solid performance at 800x600. On top of that, it also comes with the 3dfx tools bundled. The only downside is the performance in Quake 3, if that game (or games with the id Tech 3 engine in general) is important to you.