3dfx: So powerful, it's kind of ridiculous (2023)

Nothing like eating a disc to the face while you are lagging in mid-air on a dialup connection. <3

That and high speed internet. I played for a couple of years on 28.8K. The day I got a better graphics card was great. No more choppiness. The day I got cable internet was life changing in Tribes (and Tribes 2)!

I think I still have a pic somewhere of the infamous NoFix scolding "LPBs"

"Shazbot!" "The enemy is in our base!" "Woohoo!"

vgh! Except that texture transparency worked with glide (voodoo) cards and not with opengl or software rendering. So if you made a custom skin with transparency texture there was a brief time in the Tribes 1.1 to 1.2 era where you could be somewhat invisible to people with voodoo cards (if your skin was in a skin pack that everyone had).

I think I played Tribes 2 for more time than the time I've spent on every other game combined. Years!

agree -- that dual celeron setup (often with a peltier cooler) was suuuper common, I knew so many people who rushed out to get them and run at 500? it was my second exposure to SMP though: First was dual socket Pentium Pro 200mhz which ran nt4.0 for the longest time (which I still keep that hefty cpu around on my desk for laughs)

Had pretty much the same thing... but only one overclocked Celeron to 433. Was amazing upgrade from my pentium 133 with a Matrox Millenium; which I somehow used to complete Half Life in low FPS agony.

It was a cool board. I didnt technically have one, but I built my dad a W2K server on a BP6. I always wanted to hack on it and overclock with it. But after I handed it over, I wasnt allowed to touch it, "you'll burn up my processors." Since he didn't care about overclocking he had dual P2-400s or maybe 450s. It was a beast. He could run SQLServer and compile Delphi apps so hard.

I got my kicks though with a BF6 and a 300A. Those were the times; atleast until the AthlonXPs (AXIA -- anybody?) were released.

I'm still bummed that CPU manufacturers basically killed off the third party chipset industry. Maybe it was inevitable when memory controllers moved on-die, but I remember when there was actual competition in the chipset business.

I'm slightly confused, how would games of that era benefit from a dual CPU setup?

Old games were decidedly single-threaded and built for a single-core world. It was only in the mid-to-late 2000s that games started to be more optimized for multi-core CPUs. And multi-CPU is even more difficult because there isn't any cache sharing.

ABIT was such an amazing motherboard manufacturer, thanks for the KT7-A Thunderbird platform <3 R.I.P.

https://en.wikipedia.org/wiki/Universal_Abit

Oh yes, fond memories of my BP6. You just felt like you're doing something you're not supposed to, which was fun.

It was just single-CPU, but I had the ABIT BH6 with a Celeron 300A, one of the most overclockable setups ever. Mine was stable at 450mhz without any special cooling.

To add evidence: also had it.

I wonder if there something like this today you will have in college? Some low cost graphic card rigs? Or is it more like some cloudflare set-ups today?

I am starting to feel old.

    > Half of HN alive at the time probably had that motherboard - ABIT BP6

LOL. Yep. Me too.

Cathode Ray Dude's channel has a great video about the history of this amazing motherboard: https://www.youtube.com/watch?v=UE-k4hYHIDE

> There was a brief moment where Intel Celerons didn't really prevent you from using them in a dual CPU setup

I think for a while Intel started labelling the socket 370 celeron boxes as “for single core systems only”, but it was a lie.

I love this game, it's also amazing to me how the concept of "skiing" was foreign to me when I first played T1 and T2, and now its a core game mechanic.

Man I have to pick this up again. I was a top CTF, Arena, and duel player. I played through college competitively (2008)

> though IIRC you had to patch BeOS for SMP support

The board might have require a driver or patch, but SMP was BeOS's entire reason for being! The drawing of each window on the screen ran in a separate thread. It was their main selling point.

Dual 300As overclocked to just over 500Mhz each on a BP6 with Geforce 256 here too! Fastest, smoothest machine I ever used until the M1 MacBook. Quake 3 multiplayer demo ran so fast it gave me motion sickness ^^ Years later I "upgraded" to a 1Ghz Athlon and it felt like a downgrade a lot of the time.

32 megabytes!

Even HDDs were smaller back then.

I bet it was a 3.2gb hard drive. (Or you mean 32mb of ram, that sounds right too(.

I could share almost exactly the same story. So grateful my parents could afford, and were willing to spend, the money on a nice PC that I entirely monopolised.

So the story originally started with the cacheless 266 Mhz Celeron. CPUs were delivered as AICs (add-in-cards) at the time, with separate cache chips, so to deliver a budget processor, they shipped the same silicon, but without the cache chips added. Removing the cache drastically tanked the performance, especially on integer work loads (typically productivity software), but didn't really affect floating point workloads. However, it had the side benefit of removing the part of the AIC that was most sensitive to over-clocking (the cache). It used a 66Mhz clock with a fixed 4x multiplier, and upping the clock to 100Mhz got the Celeron running at 400Mhz, which had performance roughly equivalent to a 266 Mhz Pentium II with cache for integer workloads, but for games, it was almost as fast as the fastest Pentium II of the time (which topped out at 450Mhz).

In order to stop the overclocking, Intel decided to add some cache back to the CPU, but to save money, rather than using cache chips, they stuck a relatively tiny amount of cache directly on the CPU die, and released the now infamous Celeron 300A

Because the cache was on-die, it could overclock just as well as the previous celeron, but this time the 300A was faster than the equivalent Pentium because the on-die cache ran at twice the clock speed of the external caches

> By overclocking a 366MHz CPU to 466MHz, you "only" got 78MHz of memory bandwidth.

I think the PCI bus probably also typically ran at some fraction of the front-side bus. The common FSB frequencies around those times were 66 or 100 MHz which gave a standard ~33 MHz PCI bus frequency with a multiplier of 1/2 or 1/3. FSB frequencies that weren't close to a multiple of 33 MHz might have caused trouble with some PCI cards. Might have depended on how the motherboard or chipset handled the bus frequencies, too.

Of course the PCI bus should probably always run at 33 MHz but I think I saw it being modified with the FSB speed at least on some motherboards.

Aren’t you confusing that with DVD?

I think the other commenter is right...you're thinking of DVD-R vs DVD+R, possibly even DVD-RW and DVD+RW.

Based on the specs listed, OP was in college just before me or may have overlapped. The big gold CD-R stacks (you could bur in jewel cases, on spindles, or just gross stacks which were nice and cheap) were a huge thing with my group (who encoded to FLAC & MP3 -V0 and burned audio CDs relentlessly. We felt we were archiving our liberal arts college music library and radio library for the future! Who knows. Some of that "future" is still backed up and on hard disks, and I should migrate them to SSD or tape just on principle.

At that point CD-R were cheaper than CD-RWs, and because most archiving/distributing didn't require rewriting (not return-on-investment wise anyway), we just shared programs on CD-R as well. In some ways it was a beautiful technology! Particularly fidelity to a spec everyone tried to bend and break for a profit angle, when honestly, there was no point for many of us using CD-R

> but any half-decent motherboard would allow you to bump that up to 100 MHz so long as you had the rest of the hardware to support it

I think any BX chipset motherboard would do it… but you may have to resort to messing with covering/hotwiring CPU pins.

Ahhh, what a piece of shit that was! I spent summer lawnmowing money on that thing.

When I got a real job next summer, bought an AGP Matrox Millenium G200 and after that the NVIDIA GeForce2, and never strayed from NVDA since!

I had an S3 ViRGE too. It really was a decelerator, and the number of games that actually supported it was minuscule. I managed to run GLQuake, but without any textures - just shades of gray - and even that ran at most a couple of frames per second.

But there was another game - Terminal Velocity - that actually looked a lot better with hardware rendering, although it was still slower than software rendering. So, I would run it with hardware rendering to enjoy flying and then restart the game with software rendering to actually fight the enemies. :)

They also had the most recognizable unified box art style from all HW makers[1]. When you saw those eyes staring into your soul off the shelves, you knew it was a 3dfx GPU. They also had the best ads. [2] HW vendors today don't have the balls anymore to use designs like that, it's all flat sterile corporate nonsense.

[1] https://www.ixbt.com/img/r30/00/02/08/90/boxes.jpg

[2] https://www.reddit.com/r/pcmasterrace/comments/41r1wj/3dfx_w...

Still blows my mind that it was just a flash in the pan. At the time it felt that 3dfx was certainly going to be a dominant force in computing for years. And although they lingered a bit, the whole event was barely over 2 years.

I would compare it with the move from HDDs to SSDs — a night and day difference.

Even today I think a lot of Doom clones look better (or more nostalgic) with software rendering and texture mapping rather than OpenGL. There's an intensity of saturation to the colors that's lost. Fireblu is never quite so eye burning as when it's in software.

It’s also because the VGA signal quality from the 3dfx Voodoo wasn’t very good.

It didn’t have a traditional 2D graphics core at all, so it needed another graphics card for rendering the desktop (any non-accelerated apps really), and this was connected to the Voodoo using VGA passthrough. There was a noticeable image quality loss from this arrangement.

A Matrox card would give you crisp VGA with nice saturation, but the 3D acceleration was nearly worthless. Choices…

Personally I think even Half Life looks better with the software renderer today. Maybe that's just because it's the way I first played it.

It's low resolution plus the cards only supported bilinear filtering, which turns things into a blurry mess.

Overall it looked better, but a lot of Quake 2 players weren't aware of a lot of the small details that were put into the textures.

I agree that the washed-out textures haven’t aged well.

But at the time, not having pixelated textures was the first thing people looked at when judging graphics quality. I remember that being a HUGE selling point of the N64 and something that made the console look almost a generation ahead of the PlayStation and Sega Saturn to kids back then.

Today, I think I prefer the PSX look, thoug. Maybe with Z-buffer correction to avoid the warped textures of the PlayStation.

I came here to comment similarly, the lower pixelated software rendered Quake seems to work well with the textures. They have a bumpmappy fuzzy feel that gets lost with the sharp corners everything is super flat texture mapped and filtered version that one got from the 3d accelerators of the time. I guess my brain just adds context to the low res images.

Before unreal, I had a s3-virge for 2d and a powerVR 3d accelerator pair, and I was always flipping between software, virge and powerVR depending on game. Which at the time were largely hexen/heretic. The powerVR was higher resolution and clean/sterile but never seemed like a lot better experience.

But then there was unreal, the first game I think was absolutely better on an accelerator (voodoo2 in my case). Its also pretty much the last of the serious software renderers and outside of the voodoo's definitely did a better job with software lighting /texture mapping/etc than any of the previous (affordable) accelerators. Which is why I ended up finally replacing the powerVR with the voodoo2. The results were 'unreal'. Some of that might just be bias, I played insane amounts of doom/etc but never really got into quake. Quake just seemed like doom rehashed to me, so I was off playing warcraft/diablo/hexen/etc.

And frankly, outside of FEAR, I stopped playing 1st person shooter games for 20 years, the graphics improvements were so incremental, I just kept seeing flat low polygon models everywhere. And I don't think that looks good. Even after all the tessellation/bump mapping/endless tricks I kept seeing frames where I could play "count how many polygons are onscreen right now" games. Its gotten better the past few years, particularly some of the lighting, at least the screenshots/cut scenes are no longer obviously not in game rendering. The UE5 demo is slowly becoming reality in actual games, so maybe its time to revisit a few of them.

I have had three experiences like this in my life:

  1. PC Speaker -> Sound Blaster: Most games that I had were instantly upgraded

  2. Doom: my first "real" fluid 3D experience, with stairs, etc, compared to maze-like maps in Wolfenstein

  3. Software Rendering -> 3dfx (Canopus Pure3D): Transparent water surfaces in Quake (if you re-vis'd the maps), smooth and fluid visuals, it was amazing.

The closest thing to this, in modern gaming, has been the boss fights in Elden Ring: https://i.imgur.com/gzLvsLw.mp4 -- visually, they are quite epic.

Yeah same here. Always preferred the software renderer for Quake and Quake 2. Even over the modern stuff.

The software renderer has this gritty feel that is integral to the art I feel.

That said, the 3dfx was impressive at the time, and I was very jealous of my buddy who got one.

I can't speak for the original GLQuake on 3dfx hardware, but on OpenGL-compatible Quake engines (which include modern Quake source ports such as Quakespasm, Ironwail, and vkQuake), bilinear texture filtering is an option that can be turned off.

I play on vkQuake with nearest-neighbor texture filtering, square particles, and the "classic" water warping effect and lighting effects, alongside 8x MSAA, 16x anisotropic filtering, high-resolution widescreen, etc. This keeps the art style consistent with the look of the original Quake, while still allowing for the benefits of hardware 3D acceleration.

Carmageddon was a software renderer. Absolutely glorious game though!

I used to play games like Starsiege (the mech game) on dialup. With our 250ms pings, your brain just learned to compensate for the delay and click that much earlier to shoot.

But yeah, those lucky people with their DSL modems and 60ms pings would wipe the floor with us.

Nowadays, everyone has a < 10ms ping.

When qtest was released, I was there in the IRC channel and one of the first to play.

I remember connecting to someone's machine and just destroying everyone. Afterward I got a message from someone congratulating me, but being incredulous about my ping time being 26ms.

I happened to be on the first floor in the first dorm on campus to get wired internet access, and they had an OC3 dedicated to it. Two months earlier there were 16 of us splitting that line, with public IP addresses. (Going back to dial-up after leaving the dorm was.. difficult).

So I told him, yeah I kinda have a ridiculous internet connection here at school. He was like, "no, you don't understand - it is my machine. I am playing on the server and your ping time is lower than mine!"

Moto Racer was one of the few games that supported the PowerVR card. I had one and it looked so good. Such a fun game.

Adding: the server and client had to both be running vis patched maps to be able to see other players in the water due to the way entity visibility was calculated server-side.

The downside to running vis patched maps on a server is it used slightly more CPU than unpatched maps IIRC. Perhaps someone that ran more servers than I did (I ran two nodes on an Intergraph InterServe with dual P6-200s) could weigh in on what the impact was at scale.

Interesting, didn't realize this design was that old. Feels a little like teapot :).

The video also misses that there was a pretty popular 2fort for half life 1.

I spent so much of my early 20's in that map/mod on Q1. I don't think I've had that level of just fun in any game/map/mod since then.

More recent example: In GTA SA in a mission I wasn't able to reach an airplane before it took off unless I lowered the resolution of the game.

Even having a solid dial-up connection with a ~180-185ms ping was a ridiculous advantage when most HPBs were ~350ms, particularly in clan invitationals for Q1CTF. We were playing as LPBs in the dorm at ~45-60ms and 180ms wasn't that much of a concern, aside from sliding around corners more than expected, but at 350ms you were basically shooting predictively at where you assumed they'd be next, not where they 'were'.

Pre-AMD acquisition ATI also often had better hardware specs than NVIDIA, but their drivers were so often buggy and terrible. By the time they'd been fixed the reviews were long since done on the initial release versions.

AMD seems to run a better software shop, at least.

You'd buy a computer magazines and it almost felt like the performance of CPUs and GPUs when up month by month.

I'm honestly impressed people have golf game cabinets at home. never saw anyone playing those on the arcades.

Yup it was really a huge difference. FPS really needs 60 FPS and up. Anything less feels clunky.

The percentage change in resolution you ran the games at was also absolutely mind blowing too.

For the most part we went from running the game at 320x200 or 320x240 to 640x480 in that first round of acceleration. I think in terms of % change it is a bigger leap than anything we've really had since, or certainly after 1920x1080.

So you suddenly had super smooth graphics, much better looking translucency and effects, and the # of pixels quadupled or more and you could just see everything so much more clearly.

Both the 2009 and the more recent reboot trilogy (old blood, the new order and II).

Yeah exactly! OldUnreal guys did a lot of work for this classic game. I believe they even have access to the source code.

Those old cards were way under 50 Watts. Even a "low-end" card now like an Intel B580 (list price $250, inflation adjusted equivalent to about $125 in the late 90s.) is 225 W. Cooling and power circuitry are much more critical now.

Apparently 3dfx had a contract with Sega for making the Dreamcast GPU, but IPO'd in 1997. As part of the IPO they disclosed the terms of the contract, and the next gen console at Sega was a closely guarded secret at the time to avoid cannibalizing Sega Saturn sales (which were abysmal).

The contract with 3dfx was canceled leading to a lawsuit. Then Sega of America CEO disclosed they were abandoning the Saturn for the next gen console. The console did not release in Japan until November 1998 and in America until September 1999. The CEO of Sega of America Bernie Stolar was fired just days before the release, partially due to these shenanigans, and the Dreamcast as a whole was such a failure it nearly killed the company.

gkrellm, konqueror, irssi, xmms

of course i had to dual-boot windows so i could still program in vb6 and be a general shit on AOL

Haha I remember that pain now too. A long forgotten memory. Thank you!

what gave the pentium mmx the big speed boost (I also remember it being quite significant) was probably the bigger 16kb cache (pentium classic had only 8kb) rather than mmx itself.

I remember upgrading from Cyrix 5x86 to the aforementioned Pentium and suddenly being able to play Carmageddon.

I've been thinking it was due to MMX for almost 30 years!

The voodoo 4 4500 could push up to 140fps at 800x600.

https://www.philscomputerlab.com/3dfx-voodoo-shootout-projec...

GPUs for data centers makes up a vastly larger portion of NVIDIA’s sales.

In 2024 256 million PCs were sold but only 40 million of those were desktops. Excluding the fact that some PCs (hard to say a number but I'd be surprised if it weren't over 40%) are office PCs with crappy GPUs, most laptops also have a bad, integrated GPU.

There's a chance that this year or the next one more GPUs will be sold for AI than for graphics.

Specially funny to me is how on console orientend channels everyone is talking about the raise of PC gaming, it never went anywhere.

Using computers, not game consoles, for gaming was all over the place during the 8 and 16 bit home computing days, and the large majority eventually moved into Windows PC, as the other platforms died, that is why Valve has to translate Windows/DirectX if they want to have any games at all on their SteamDeck.

Consoles has been a niche market, at least here in Europe, mainly used by kids until they are old enough not to destroy the family's computer while playing games, given that many families only have one per household. And nowadays that roles has probably been taken over the tablets.

To the point that Playstation/XBox are now becoming publishing brands, as the exponential growth for selling consoles has plateaued.

PC gaming is going nowhere.

anti-matter-millionaires

The problem is, they already had a large group of board providers, it was a big market advantage for them.

If you listen to the Computer History Museum interview, many of these providers jumped ship.

Also the bought a board provider that wasn't very successful, and their resulting boards weren't very successful either.

The same article also has another photo of a Creative-branded card. Other card brands also definitely marketed cards with ATI chips. I remember I had a R200-based card from some third-party manufacturer. ATI and Nvidia also had reference cards but I don't know if those ended up being sold in the mass market or not.

I think the whole 'moat prison' thing is overrated. They could release future version that support DirectX and Glide.

They simply didn't focus enough on the next generation chip.

I worked for an elevator manufacture that killed somebody, they fucked up the whole apologize in Japan thing too. Even if it wasn't actually their fault.

Yeah, the earliest models are literally 3D only as discussed in the article, they have separate pass through cable for your existing 2D graphics because even just making a flat 2D window isn't viable directly, Glide really wants to render only textured triangles which is fine for Quake but no good for Windows.

I was playing around with building retro game VMs with qemu and pci passthrough awhile back and dusted off my old canopus pure 3d to try out with a pci to pcie adapter board. It was kind of amusing you'd have the windows desktop running in virt-manager and when you fired up unreal tournament the desktop would just freeze and only then would the card actually output anything.

Nice article - thanks!

I remember at the time things were moving somehow chaotically.

Also puzzling why 3dfx didn’t start to support DirectX in parallel with Glide.

Both SGI (who had no hardware texture mapping at the time) and 3dfx were explored at the time DARPA SIMNET was being developed. Delta Graphics (Mike Cyrus' and Jay Beck's company - of the Cyrus-Beck line clipping algorithm fame), which did have hardware Z-buffer texture-mapping, were the incumbents, and was fielded. Gary Tarroli personally visited BBN for a chat, but I wasn't involved in that. If Abort-Retry-Fail is interested in the history of SIMNET (in addition to the first hardware texture-mapping fielded, first AI NPC, etc.), there is a lot of info available. Email me and I can provide the info.