IMO both cards are a bit weak for a P3. Then V3 is a great card though, especially for Glide games based on the Unreal engine.

Fusion wrote:

So after discovering that the TNT2 M64 I thought I had is actually a Pro version (w00t), I decided to test the card in my P3 rig […]
Show full quote

So after discovering that the TNT2 M64 I thought I had is actually a Pro version (w00t), I decided to test the card in my P3 rig. I noticed it is a few frames slower in Quake 3 then the Voodoo3 (195/195), and that is only after I overclock the **** out of it.

Google searches just make things more confusing with people saying the TNT2 is faster while others say the Voodoo 3 is faster. Basically I'm reading posts made by snot nosed fanboys 17 years ago. 🤣 I don't care about 32bit rendering since the TNT2 sucks at it.

I'm going to run some benchmarks tonight with Return to Castle Wolfenstein and post my results. I'm curious if the extra VRAM on the TNT2 will help it. it definitely did not matter in Q3.

I think because my Voodoo3 is so badass and clocked so high the TNT2 has no chance 😎 , and might be a unfair comparison. But the TNT2 Pro is overclocked higher then an Ultra, so who knows!

if you want a 3dfx card for a P3 a Voodoo 4 or 5 are more appropiate but expensive. A Geforce 3 or 4 or better suited for the job although a geforce 2 will do aswell.
Don't listen to Nvidia's propaganda claiming their TNT is actually faster, it isn't. They just convinced people Voodoo cards were slower and worse then their own cards, plenty about that is already said on this forum about that and is also listed on the internet in reviews.

TNT2 is overall better in perfomance due to more flexible pipeline. 1mhz TNT2 > 1mhz Voodoo 3.

V3's got better optimized drivers for K6 and Cyrix processors. It also has decent picture quality and more guaranteed game support

PhilsComputerLab wrote:

IMO both cards are a bit weak for a P3. Then V3 is a great card though, especially for Glide games based on the Unreal engine.

Too weak eh? Even on a P3 running @ 558Mhz? I've heard even here that my processor is a bottleneck.

The Serpent Rider wrote:

TNT2 is overall better in performance due to more flexible pipeline. 1mhz TNT2 > 1mhz Voodoo 3.

On my PC the TNT2 is slower in Quake 3, but my V3 is clocked very high. But I was hoping the TNT2 could shine in more texture heavy games. I guess this is what I'm looking to find out. RTCW runs very nice on the V3 and looks awesome @ 1024x768.

My build was mainly for Glide games, and I got lucky with my V3 purchase because I found it local and before the whole 3dfx price hike.

I can speak from personal experience from back in the day when it was all happening.

I ran a Voodoo 3 3000 AGP with a Celeron 500 in an ABIT BM6 with 384MB RAM. I thought it was totally badass at the time. The last games I remember playing on that were SWAT3 and Black & White.

Eventually, I upgraded to a Voodoo 5 5500 AGP with Athlon 1200 and 768MB RAM in an ABIT KT7A. It was good, but not great. The last games I remember playing in that configuration were GTA3 and Dark Age of Camelot. But I was a badass once again after upgrading to a GeForce 3 Ti500 in that same system. The last games I remember playing on that system were Mafia and TRON 2.0.

Nowadays, I run Voodoo 3 3000 AGP cards in all three of my systems beyond P233 up to P3 1ghz.
After that, it's all Athlon 1400s in KT7A systems with GF4 Ti4600s.
After that, I've got an Athlon XP 2100+ with GF5/GFFX 5950 Ultra in a KT7A.
After that, Athlon 64 3400+ Venice with GF6 6800 GT.
After that, hell... that's C2D territory and I'm still feeling it out. It's not really legacy/retro yet. However, right now the GTX 750Ti is feeling right in a 4GB RAM WinXP 32-bit configuration along with a Sound Blaster X-Fi for EAX compatibility.

The Serpent Rider wrote:

TNT2 is overall better in perfomance due to more flexible pipeline. 1mhz TNT2 > 1mhz Voodoo 3.

Strong generalization. As in single textured games? Quite so, but these were "fast enough" anyway. And since TNT2 had usually lower core clock than Voodoo3 ...

After pushing the TNT2 Pro to 172/205 it only then matches the Voodoo3 3000 @ 195/195 in Quake 3 demo001. They both average 50fps @ 1024x768 16bit with normal texture detail.

I want to do a custom 3DMark2000 run on both cards soon too.

So I ran a few specific tests in 3DMark2000 to compare the two:

TNT2 Pro @ 170/200
Voodoo3 3000 @ 195/195

Game 1 Helicopter High: TNT2: 19.2fps Voodoo3: 15.0fps
Game 2 Adventure High: TNT2: 22.7fps Voodoo3: 20.1fps

Fill Rate (Single-Texturing): TNT2: 297.2 MTexels/s Voodoo3: 191.6 MTexels/s
Fill Rate (Multi-Texturing): TNT2: 274.3 MTexels/s Voodoo3: 377.1 MTexels/s

High Polygon Count (1 Light): TNT2: 3,054 KTriangles/s Voodoo3: 2,066 KTriangles/s
High Polygon Count (4 Light): TNT2: 2,665 KTriangles/s Voodoo3: 1,766 KTriangles/s
High Polygon Count (8 Light): TNT2: 2,295 KTriangles/s Voodoo3: 1,493 KTriangles/s

8MB Texture Rendering Speed: TNT2: 204.7fps Voodoo3: 199.8fps
16MB Texture Rendering Speed: TNT2: 198.1fps Voodoo3: 186.9fps

So the TNT2 Pro is faster overall, but the Voodoo 3's multi-texturing performance is considerably faster using this benchmark suite.

I'm disappointed in the Voodoo 3's performance when it came to the High Polygon Count tests, especially when you look at the percentages.

Richo wrote:

if you want a 3dfx card for a P3 a Voodoo 4 or 5 are more appropiate but expensive. A Geforce 3 or 4 or better suited for the job although a geforce 2 will do aswell.
Don't listen to Nvidia's propaganda claiming their TNT is actually faster, it isn't. They just convinced people Voodoo cards were slower and worse then their own cards, plenty about that is already said on this forum about that and is also listed on the internet in reviews.

No way. Depending on clock speed and platform, an appropriate card for a P3 machine is either a TNT2 / Voodoo 3 (for say a slot 1 machine running a 450-600MHz CPU) or a Geforce 2 PRO / GTS or a GF 3 Ti200 for a socket 370 machine running a 800-1000MHz chip. A gf4 Ti is wasted on a P3 unless you have a fast tualatin like say a 1400MHz rig. I experimented with my 1400MHz machine (Abit ST6 / 1.4GHz P3) and saw little improvement when upgrading the GF3 Ti200 to a GF4 Ti 4600 - all it did was add 5-10% better maximum framerate in some games at 1600x1200 but they were still laggy and minimum framerate got worse using the GF4 Ti.

There's no point in going with a very fast card on these older machines since the chipset's CPU to AGP bus can't feed data fast enough to the card for it to perform as well as it should. From my tests, only a socket 939 3800+ could top out a GF4 Ti 4600 (top-out = video card becomes a bottleneck). In sistethic tests my Asus GF4 Ti 4600 scored on 25-30% better on the 939 rig then it did on a socket A machine (3200+ / NF2) and in games there was a 5 to 15% improvement.

As for a Voodoo 5 - it actually loves fast CPUs. It's fast enough to fluently play lots of 1998-2000 games @ 1600x1200 on the 3200+ machine in my signature, but lags and displays some sort of tearing (at times) on the aforementioned Tualatin machine in some games.

The Voodoo 4 behaves like a slightly faster voodoo 3 with 32 bit color support - but you will only notice it's extra performance on faster machines. For example, on a 650MHz P3, both the Voodoo 3 3000 and my Voodoo 4 (Powercolor Evilking IV) perform almost the same. When installed on my 1333MHz Athlon rig, the V4 takes a 20-25% lead in some games, especially at high resolutions.

Very fast AGP cards actually perform worse on slow machines then older, slower video cards would, in part due to drivers not being optimised for the older platform, but mostly due to the older machines innability to keep the card properly fed with sufficient data, causing huge frame spikes, lag and framerate variations. I noticed this with the 6600GT AGP and 9800 PRO when installed on a 1100MHz Coppermine machine.

I'll open up a topic about what video card to put in what machine in the near future.

TNT2 Pro is a die shrink (220 nm), so obviously it gets faster core clocks than the TNT2/U. Reason why the TNT2 Pro is overlooked is because the GeForce SDR was launched the day before. The only TNT2U to launch at 175/200 (the supposed clock speed before Nvidia dropped it to 150/183 due to yield issues) was the Hercules Dynamite TNT2U. And even at these speeds, it just barely edges out a stock V3-3000 in < Year 2000 games. Past that, TNT2 starts pulling ahead (as we see in 3DMark2K, which was ahead of the curve).

http://www.anandtech.com/show/305/4

leileilol wrote:

V3's got better optimized drivers for K6 and Cyrix processors. It also has decent picture quality and more guaranteed game support

Weak 3DNow performance was patched, with the K6/3-400 nipping at a P3-500's heels.

http://www.anandtech.com/show/305/9

IIRC, 3dfx image quality didn't get really good until VSA-100. Although V3 was noticeably above V1 and V2.

Lets not presuppose correlation between performance in 3Dmark and games, please?

dexvx wrote:

TNT2 Pro is a die shrink (220 nm), so obviously it gets faster core clocks than the TNT2/U.

No, die shrinks do not guarantee faster clocks.

Putas wrote:

dexvx wrote:

TNT2 Pro is a die shrink (220 nm), so obviously it gets faster core clocks than the TNT2/U.

No, die shrinks do not guarantee faster clocks.

There is a very strong correlation between die shrinks and faster clocks of the same uarch. There are cases where a die shrink is accompanied by a different uarch and/or packaging that may not result in increased clock speeds (e.g. i7-2600K to i7-3770K TIM fiasco or Northwood to Prescott fiasco).

I believe there are inherent limitations beyond gate delay for some semiconductor substrates and dopants. These tend to be the limits explored by cryogenics enthusiasts. Also, some semiconductors using vertical feature construction have had issues with heat transport as the additional SiO2 packed in around gates can have stronger insulating effects than traditional technologies.

Great discussion everyone!

I just finished running some benchmarks with my Pentium III system (see sig) and results were disappointing... Sort of. After running many hours of Return to Castle Wolfenstein benchmarks using a MP demo called 'Checkpoint', with both the Voodoo 3 3000 and the TNT2 Pro, it turns out that I am extremely CPU limited with this game. I couldn't get either card to perform over 30FPS with low settings even at 640x480! But enable more graphic options up to something like "medium-high" I guess you would say and using 1024x768 performance only drops to 25-28FPS on either card. The TNT2 is roughly 2FPS faster then the Voodoo 3 on average. Not what I expected going into this. 30FPS with big spikes isn't a fun experience in multiplayer, but I was hoping single player it would run a bit better? Maybe not be so CPU limited?

EDIT: Oh, I forgot to mention that I was only able to use 1024x768 on the Voodoo 3 using the default OpenGL driver, I couldn't go higher then 800x600 using the included Wicked3D driver.

dexvx wrote:

3dfx image quality didn't get really good until VSA-100

That depends. Even Voodoo 2 can do proper linear texture filtering, which TNT series can't do.

PhilsComputerLab wrote:

IMO both cards are a bit weak for a P3. Then V3 is a great card though, especially for Glide games based on the Unreal engine.

Doesn't that kind of depend on what P3 it is? My P3 system is a Slot 1 PIII-450, essentially a stand in for the PII-450, and the Voodoo 3 complements it perfectly.. As you progress forward from Katmai to Coppermine to Tualatin, you probably want to go for DirectX7 (GeForce 2) and DirectX 8 (Radeon 8500) class hardware.. But for a system built around 1999 specs, not really..

appiah4 wrote:

PhilsComputerLab wrote:

IMO both cards are a bit weak for a P3. Then V3 is a great card though, especially for Glide games based on the Unreal engine.

Doesn't that kind of depend on what P3 it is? My P3 system is a Slot 1 PIII-450, essentially a stand in for the PII-450, and the Voodoo 3 complements it perfectly.. As you progress forward from Katmai to Coppermine to Tualatin, you probably want to go for DirectX7 (GeForce 2) and DirectX 8 (Radeon 8500) class hardware.. But for a system built around 1999 specs, not really..

Sure, but the 450 MHz IS the slowest Pentium III, I think I just assumed something more potent would be used. The P3 does go up to 1.4 GHz, so there is quite a range.

Easy test, run a benchmark at 512 x 384, then again at 1024 x 768 and see what difference you're getting.

Hey Phil, any games you'd suggest I'd throw at these cards? I'm really just trying to justify a reason to use this TNT2 Pro. 🤣 I test with 3DMark, Q3 and RtCW. Looking for something that isn't a Quake engine game. I wanted to bench Soldier of Fortune, even though its Q2 based, but the demo1.dm2 I found won't work with the version I have so that backfired.