Introduction
Due to popular demand I’ve altered my plan to bring you the performance information on all the cards last. Part two has now changed into the hardware performance section. Due to availability changes, a couple of cards were taken out of the review and a two were added on. I will update the driver screenshots for the added cards as soon as possible. The last part will now be the hardware/software accessories section. I think everyone will be happier this way since I will now have more time to review the variety of featured hardware/software accessories that all these cards have. I will try to cover a few added topics in part 3 that you, the readers, have brought to my attention. Enough talk about the changes and onto the review :).
Before I get into my analysis I want to clarify a problem that I run into when we do a performance comparison. Not everyone has the exact same needs. I have set minimum criteria for performance on this particular review and I will stand by it when I start to critique cards. You might have a different opinion on this and that is perfectly fine if it suits YOUR needs! Make sure to keep this in mind throughout this review. My rule of thumb currently for hardware performance is 30 FPS on average at 1024x768x16. This is a performance level that I think most of you would expect from a card that you would buy now, if not more.
While the benchmarks were being run, I was already starting to see things I really disliked and felt were unacceptable. Some of the cards had visual defects or didn’t run every test without crashing. Below each of the performance charts I will explain any visual defects or failures to run the test. It seemed that most of the problems were limitations of the given chipset driver or the card frame buffer size. I also experienced poor visual quality on quite a few of the tests, which bothered me. I expected this problem to happen when testing the OpenGL ICD benchmarks but not the DirectX tests. I will also note when I saw these issues.
Test configuration
Hardware Information | |
CPU | PIII 550 |
Motherboard (BIOS rev.) |
ABIT BX6 2.0 (BIOS date 7/13/99) |
Memory | 128 MB Viking PC100 CAS2 |
Hard Disk | Western Digital WDAC 4180000 EIDE DMA mode enabled |
Network | Netgear FA310TX |
Environment Settings | |
OS | Win98 SE 4.10.2222 A |
Descent III (retail version) | command line: -nosound -nomusic -nonetwork -timetest secret2.dem -width xxxx -height xxxx |
Expendable (demo version) | Demo version. Setup – Triple buffering. Sound – disabled. |
Quake 3 Test v1.08 | command line: +set s_initsound 0 +set cd_nocd 1 |
Shogo (v2.14) | Advanced settings: disable sound, disable music, disble movies, disable joysticks, enable optimized surfaces, enable triple buffering, enable single-pass multi-texturing. High Details Settings = enabled |
Video Board Manufacturer | Driver Version |
3dfx Voodoo3 2000,3000,3500 | 4.11.01.1151 |
AOpen PA 3010 | 4.11.01.0208 |
ASUS (PCI,AGP,APG Deluxe) | 4.11.01.0208 |
ATI Rage128, All-in-Wonder | 4.11.1.1003 |
CardExpert i752 | 4.11.01.1254 |
CardExpert SG4 | 4.11.01.8007-8.10.10 |
CardExpert TNT2 | 4.11.01.0191 |
Creative Labs TNT2/Ultra | 4.11.01.2103.03.0204 |
Creative Labs Savage4 | 4.10.01.2102-4.8401 |
Diamond Savage4 | 4.11.01.0204 |
Diamond TNT2/Ultra | 4.11.01.0402 |
ELSA TNT2 | 4.11.01.0200-0022 |
ELSA Savage4 | 4.11.01.0303-0001 |
Gigabyte GA 660 | 4.11.01.0208 |
Guillemot Xentor32 | 4.11.01.0208 |
Hercules TNT2/Ultra | 4.11.01.0208 |
LeadTek | 4.06.02.0436 |
Matrox G400/MAX | 4.11.01.2110 |
Microstar MS 8806 | 4.11.01.0208 |
Number Nine SR9 | 4.11.01.9550 |
PowerColor Savage4 | 4.11.01.8003-8.03.03 |
PowerColor TNT2 | 4.11.01.0188 |
Skywell TNT2 | 4.11.01.0208 |
Benchmark Results – Shogo
A quick note before we get into the numbers of Shogo. The Savage4 cards all have graphical problems in Shogo with the test settings. Here are pictures to show you what I mean.
Benchmark Results – Shogo – 640x480x16
The performance of most the cards are excellent until we hit the Savage4 and i752 chipsets that play acceptable levels of performance. The top dogs in this test, however, are the TNT2/Ultra and Voodoo3 (3500/3000) cards.
Benchmark Results – Shogo – 1024x768x16
As we start to make things harder on the video cards, we can now see the powerhouses flex their muscles. With superior clock speeds the Hercules TNT2 Ultra and the Xentor32 take the leading spots. Most of the TNT2/Ultra cards follow the pack leaders as well as the Voodoo3 3500 and Matrox G400 MAX. Looking down the chart we see the acceptable performance drop off at the ATI AIW 128. In the Shogo test, we can already see that after 640×480 the Savage4 cards aren’t cutting it anymore. If you’re looking for DX performance above 640×480 then the Savage4 won’t probably be your card of choice.
Benchmark Results – Shogo – 1600x1200x16
Pumping up the resolution into the 1600×1200 region has left us with really no card capable of performing acceptably. The only cards worth even trying to run in this video mode are the top TNT2 Ultra cards and the Matrox G400 MAX. We had quite a few cards that couldn’t run the test (X-Demon, i752) and a few that ran missing textures (Voodoo3 cards).
Voodoo3 Shogo Screenshots
Benchmark Results – Expendable – 640x480x16
Once again we see that in a DX application at 640×480 that most of the cards performance acceptable. I wouldn’t even claim to have a clear performance winner here since most of the cards are pretty darn close to the same speed.
Benchmark Results – Expendable – 1024x768x16
The Hercules TNT2 Ultra and the Xentor32 pull ahead once again followed closely by the G400 MAX, Voodoo3’s and the top few TNT2’s. Acceptable performance levels drop at about the X-Demon. You must be wondering why the Diamond V770 cards don’t have scores. Every time I tried to run these cards at the 1024x768x16 setting, the benchmark crashed and put me back at the desktop. The weird part was that I was able to run the tests at 1024x768x32 just fine. After attempting to use the NVIDIA reference driver, I was able to make the cards work just fine at that setting. I hope Diamond looks into this problem before anyone else gets upset.
Benchmark Results – Expendable – 1600x1200x16
Check out that G400 MAX stepping all over the competition in this test! It scores high enough to get my vote for playable performance at 1600x1200x16, nice. The G400 and top few TNT2’s are on the borderline of playable but I wouldn’t bother trying to play with anything less than these top contenders. The Voodoo3 cards ran the test missing textures at this resolution.
Benchmark Results – Expendable – 640x480x32
Here we have the switch to 32-bit color and at this resolution most of the 32-bit capable cards perform well until you hit the Savage4 that start to drop into the acceptable category. I’ll mention this once so I don’t have to keep repeating myself: the Voodoo3’s don’t do 32-bit 3D. We’ll dive into this issue once we come to the conclusion of the Video Card Meltdown (part 3).
Benchmark Results – Expendable – 1024x768x32
As we turn the notch up one level, we can see the number of cards in the high frame rate area drop down into the acceptable level. Leading the pack we have the TNT2 Ultra’s and the G400 family of cards. The TNT2 and ATI boards seem to have mostly fallen into the barely playable area while the Savage4s fall out of the contest.
Benchmark Results – Expendable – 1600x1200x32
Putting the full weight of Expendable on these cards left one card standing: the G400 MAX. Not even the higher clocked TNT2 Ultras stayed within the playable region. Quite a few cards were unable to run at this high of a resolution and color depth but I really doubt they would have done too well.
Benchmark Results – Descent 3 D3D – 640x480x16
Quick note: I ran into something very interesting when I did my tests with Descent 3. I feel that it brings a great example of why native API support is a benefit. This is at least in the case of Glide, which has been embraced by many game developers who have and will be making games using Glide. Descent has three choices: DirectX, Glide and OpenGL. I was a bit annoyed when I had stability problems of my Voodoo3 boards running in DX or OpenGL but then I realized why this was so. The game supports Glide and wasn’t probably QA’ed using DX or OpenGL very well since there would be no reason to use those modes. So in all the Descent3 tests you will see the Voodoo3 scores ran in Glide mode only. This isn’t exactly what I had planned since these tests were intended to analyze the cards in various game applications to compare driver performance. However, I decided that this is just a perk for the Voodoo3 and that’s what you’ll see as I show the rest of the results.
At the top of the pack you can see the Voodoo3 cards shooting way ahead. Almost all cards are performing in the excellent area at this setting. The i752 and Savage4 cards are the only ones flirting with falling below that level.
Benchmark Results – Descent 3 D3D – 1024x768x16
Cranking up the resolution a little we see the TNT2’s start to catch the Voodoo3 family of cards. The Voodoo 3500 keep it’s lead however. The ATI cards maintain an acceptable level of performance here while the Savage4 (minus a slightly faster X-Demon) and i752 cards fall out of the competition.
Benchmark Results – Descent 3 D3D – 1600x1200x16
Ok, we’re now flooring the cards at 1600x1200x16 in Descent and most of the cards have bit the dust. At the top we have the Voodoo3 3500 taking top rank as the high speed TNT2 Ultra’s keep close behind. Look at how that PCI TNT2 performed, not bad. I know quite a few people will be excited to see how well that card is doing for lower end motherboards that don’t have AGP slots.
Benchmark Results – Descent 3 OpenGL – 640x480x16
Quick note: This is where things will start to get ugly. Most cards don’t support OpenGL very well and if they do, it’s to run Quake2/Quake Arena. This is why I was a bit happy to do these benchmark runs. It’s really annoying when a driver is optimized for one or two games. It makes it tough for game developers to support a good API like OpenGL if video card drivers “cut corners” to get performance.
With the comfort of using it’s native API, the Voodoo3 cards take the top spots with ease. We have the TNT2’s and the ATI cards performing pretty damn well at this point while the Savage4 boards struggle to even keep 30 FPS. Unfortunately the i752 is already out of the competition BUT at least it worked! The G400 cards have issues in D3 while using OpenGL. The screen would go blank on me and seem to run fine (without video output). The strange thing was that at the end of the runs I still got a score. This is the case in all testing resolutions. Having no clue as to the legitimacy of the scores, I still failed the cards. I’m not going to tell you how a card performed if you can’t even see video. Even so, the card has crappy performance according to the score it output. I find it annoying that the Matrox OpenGL driver isn’t quite as sharp as the DX driver and even more so that the driver won’t run a non-Quake engine based game in OpenGL. Every other card was able to pull it’s own, why not the G400?
Benchmark Results – Descent 3 OpenGL – 1024x768x16
No surprises here as we see the TNT2 Ultra not matching up to the Voodoo3/Glide combination. However, all of the TNT2 cards perform very well even with the OpenGL driver. All the cards after the Asus Magic M64 card scrape on the walls while trying to climb to the 30 FPS mark but fall short.
Benchmark Results – Descent 3 OpenGL – 1600x1200x16
The two faster Voodoo3 cards maintain that performance edge as only the Hercules TNT2 Ultra and Xentor32 keep above the 30 FPS mark! Quite a few TNT2 cards are near the 30 FPS mark so they might be barely playable but you’ll have to decide on that one.
Benchmark Results – Quake 3 Test v1.08
Quick note: I ran Quake 3 Test in “Normal” mode and just changed resolution for the 16-bit testing. As for 32-bit testing, I set the mode for “High Quality” and changed resolution. I also noted that the Savage4 cards have some funky graphical issues that can be best described as “Disco Fever.” Check this out:
Benchmark Results – Quake 3 Test v1.08 – 640x480x16
Up at the top we have a couple of the Voodoo3 cards followed closely by TNT2’s. The ATI cards do impressively well in this particular test followed by most of the Savage4 cards that oddly enough are faster than the G400 cards!
Benchmark Results – Quake 3 Test v1.08 – 1024x768x16
Things take a dramatic shift as the TNT2 cards shift upward to dominate the top spots. The Voodoo3’s follow a steep 20 FPS behind the leaders after just leading at the 640×480 resolution. Normally I would care less about 20 FPS but that’s taking the Voodoo3’s from excellent to just acceptable levels. The G400 cards aren’t doing too hot but manage to keep themselves in the area of the Voodoo3 performance region. As we get to the ATI cards, the performance of the cards begin to fall over the edge of acceptable performance.
Benchmark Results – Quake 3 Test v1.08 – 1600x1200x16
The G400 MAX from nowhere takes the top spot barely with the Xentor32 and Hercules nipping at it’s heels. Sadly enough these are the only three cards probably worth even trying at this resolution and settings.
Benchmark Results – Quake 3 Test v1.08 – 640x480x32
As we switch to 32-bit color the Voodoo3 and i752 cards check out and cards like the G400 start to shine. Unfortunately the G400 doesn’t seem to do too well in OpenGL at low resolution. The top positions are once again dominated by TNT2 based cards with the ATI and Savage4 cards trailing behind.
Benchmark Results – Quake 3 Test v1.08 – 1024x768x32
The heat starts to pick up even more as the resolution skips to 1024×768 and a very colorful 32-bits. The Hercules TNT2 Ultra and Xentor32 cards take the top spots while most of the faster TNT2’s follow right behind. The G400 MAX gains some ground thanks to it’s great 32-bit performance. One of the most notable disappointments was the Diamond V770U dropping below an acceptable 30 FPS, not good.
Benchmark Results – Quake 3 Test v1.08 – 1600x1200x32
With the current test demo, video settings and hardware combination, none of the cards made an acceptable performance level. However we can see that compared to most of the cards that the Hercules TNT2 Ultra pulls about 10% ahead of the next closest competitor.
Summary
Keep in mind that I’m saving my conclusions until the end of part 3 where we can consider all the factors before making decisions. However, I would like to discuss a few things.
Hercules
It’s sad to see the business go under with such a great product line. I still included the card in the review for the folks that might be able to still get a hold of one. I’m sure they’re still floating around in some warehouses. I will definitely miss the performance and quality of Hercules TNT family of products. I felt the same way when Canopus stopped making cards for the U.S. market. The real loses came to the consumers who couldn’t get the high-end products anymore.
Savage4
For a low-end solution this card isn’t too shabby as long as you don’t play games above the 640×480 resolution. I just wish they would get the driver quality to an acceptable quality level. With drivers like these, will we really feel confident about the Savage 2000 product? Maybe they’re focusing on the driver development of the newer cards but that’s still no excuse.
G400
I’m still waiting for a better OpenGL driver from the Matrox folks. If they can whip so much ass with the DX driver, why can’t they put out the same performance with the OpenGL? This really bugs me. It seems to be a great card but if this issue is going to hinder their product, I don’t think they’re do well with the high-end gamers.
Voodoo3
I want to make it clear that our test suite was geared to test video cards and not to make any “select few” cards look better. It just so happens that we decided to include a few 32-bit tests in our suite but that doesn’t mean I think these cards suck because they fail to run in 32-bit. The Voodoo3 3500 is on the fastest cards in 16-bit color, especially when running Glide-based games. In Shogo as well as Expendable the TNT2-Utlra-based cards score better though, and they offer 32 bit mode. People need to decide on the feature set they need before making judgments on card x and card y. If 32-bit isn’t your thing yet, the Voodoo3 3000 and 3500 is not a bad choice at all.
Asus PCI-V3800TV
After getting a ton of e-mail, I had to make sure to address this. I DO in fact have a PCI TNT2 card. Asus told us that this IS a shipping product and if that were not the case, I wouldn’t review the card. I will try to get Asus to clarify the situation of this card because there is a strong interest for it. Many people have stressed to me that they would love to upgrade their non-agp motherboard with a high-end PCI video card. I hope to have an update on the release of this card soon.