All these new PC hardware announcements have us feeling nostalgic again, so we took another trip down memory lane to talk in-depth about the '90s 3D accelerator boom. The rise and fall of 3dfx! The OpenGL vs. Direct3D wars! All those .plan updates! Join us for some reminiscing about how we got from the earliest cards to the GeForces and Radeons of today. Support the Pod! Contribute to the Tech Pod Patreon and get access to our booming Discord, your name in the credits, and other great benefits! You can support the show at: https://patreon.com/techpod
All these new PC hardware announcements have us feeling nostalgic again, so we took another trip down memory lane to talk in-depth about the '90s 3D accelerator boom. The rise and fall of 3dfx! The OpenGL vs. Direct3D wars! All those .plan updates! Join us for some reminiscing about how we got from the earliest cards to the GeForces and Radeons of today.
Support the Pod! Contribute to the Tech Pod Patreon and get access to our booming Discord, your name in the credits, and other great benefits! You can support the show at: https://patreon.com/techpod
Will: [00:00:00] Did you see that thing battle Frankie posted after last week's episode,
brad: [00:00:03] Which one, a lot of people, a lot of the post, a lot of things on there
Will: [00:00:06] The normalized human time, when you compare one clock cycle in a CPU at three gigahertz to like one second of real world time,
brad: [00:00:15] Yeah. That thing was very eye opening. Let's say.
Will: [00:00:18] I did not. Like I knew I understood this. I thought, but this illustrates it in the best way I've ever seen.
brad: [00:00:24] Yeah. I knew all that stuff intuitively, but to see it quantified is a, is a very different thing. Why don't you explain exactly what we're talking about here?
Will: [00:00:31] Yeah it’s hard to understand, like what a nanosecond, how a nanosecond relates to A say microsecond. Right? So, um, on a three gigahertz clock cycle, one clock cycle, it takes about three tenths of a nanosecond. So that's a three, 10 billions of a second, I guess. And, um, if you normalize that to one second, then accessing the level of one cache of that.
The one that's built into the CPU takes three seconds. The level two cache takes nine seconds and the level three cash takes 43 seconds, which is bonkers. Like that's the big thing that we were talking about on the chiplet that holds the whole thing together. Hell, accessing Ram, which takes between 70 and a hundred nanoseconds, depending on the clock speed and latency of your Ram T would be equivalent to 3.5 to 5.5 minutes.
So it's like,
brad: [00:01:23] That was the part where my eyes flew wide open. Like the cache stuff makes sense. It's like, yeah. Okay. Sure. The level three cash is going to take a good number of clock cycles to. Because that's the closest one, of course. But then once, once we hit the Ram one and I was like, wait, what? Wait system ran supposed to be very fast.
Right? Like that's the fastest thing outside of the CPU in the computer. But those normalized figure being like five minutes to access was
Will: [00:01:47] So like the CPU wants something it's in physical Ram on the motherboard, on the chips that you socked into your motherboard. And they're like, it has to wait the equivalent of five minutes to get the answer back. That's Bananas.
brad: [00:01:59] it's this wild, well, keep going.
Will: [00:02:00] it gets worse, um, to access an NVME SSD IO, which is usually about between seven and 150 microseconds.
That's the equivalent of two hours to two days
brad: [00:02:13] that's a long time.
Will: [00:02:14] like, it's like getting a letter, um, rotational, disk IO,
brad: [00:02:20] Oh, you mean a hard drive
Will: [00:02:21] a hard drive. One to 10 milliseconds of computer time is equivalent to 11 days to four months.
brad: [00:02:27] That's uh,
Will: [00:02:28] Pony express, man,
brad: [00:02:30] if that maybe carry your pigeon. I don't know.
Will: [00:02:34] we've got to get this letter to Chisholm by the next, by, by October. Don't worry. The pony express has, um, internet latency comes into play next internet, San Francisco to New York city about a 40 millisecond ping, which is I think, uh, optimistic. This is equivalent to about 1.2 years.
brad: [00:02:51] that's uh, that's quite a while. I'm going to say.
Will: [00:02:54] yeah.
That's. Look, uh, some people, some people's whole life. Um, the internet, uh, from San Francisco to Australia, 183 milliseconds is six years, uh, rebooting your virtualized OS like if you have an, a, if you have a us running in a, in a VM hypervisor on your machine, about four seconds on the PC, 127 years compared to the one nanosecond, uh, uh, the one cycle, one second clock time.
brad: That’s long than a life time.
Will: Uh, rebooting the virtualization hypervisor. I assume that's what they're talking about here. It's 40 seconds. That's 1200 years and mashing the reset button on your computer and letting it come back up, which takes about 90 seconds. Three millennia.
brad: [00:03:40] Ah the processor has seen some shit.
Will: [00:03:42] Yeah, this is some, this is, uh, some dark magic. I think that this was posted. Uh, I should have grabbed the link, but this was in an article that they had linked to, uh, elsewhere from elsewhere. And I found it incredibly informative
brad: [00:03:57] Kind of a disturbing, a little bit, a little bit unsettling to me.
Will: [00:04:01] I guess it's a little scary when you think about how much.
Uh, this is from a site called formulasblack.com. Uh, and if you Google compute performance, distance of data as measure of latency, uh, it'll pop up for you. And we'll put a link in the show notes too, but, uh, yeah, this is like we spent a long time talking about how to visualize this kind of timescale at maximum PC.
And this is easily the best. And, and should have been the most obvious solution for that. I don't think I would have trusted our math, frankly, cause I would've gotten to the, Oh, it takes 1.2 years to send a packet from New York to San Francisco and been like, yeah, we probably didn't carry a zero right here somewhere.
We should go back and plug this back into the spreadsheets.
brad: [00:04:44] Yeah, but it's like the, with formulas and the name, I think you can probably rest assured that their data is good. Uh, this is, I don't know. This is unsettling to me from both from the standpoint of putting into relief, how much of our lives we are spending in front of computers. And also just thinking about a CPU's experience of time longer than you think Will.
Will: [00:05:03] It's funny
brad: [00:05:04] Long than you think.
Will: [00:05:05] there's there's um, in Neal Stephenson Seveneves, which is one of my favorite books, uh, he, they talk about like there's a space-based society and they talk about. Processors and memory and things like that, that you can't build an outer space because of cosmic radiation messing, like.
Like th there's a reason we still use like eighties CPU's on a lot of, um, on a lot of, uh, space probes. And it's because they're higher process, you know, very large processes by today's standards, so that a stray particle smashing into the CPU. As it's flying through space, doesn't Jack up the mash and crash, crash, crash, the whole thing,
brad: [00:05:46] Literally literally needed a bigger surface area on the chip to try to be able to be more resilient.
Will: [00:05:51] Wider pathways for the electrons are safer, is my understanding. I'm sure that somebody in the discord will correct me when I'm wrong on that. But, um,
brad: [00:05:58] that's wild.
Will: [00:05:58] yeah, it's a trip.
brad: [00:06:00] We do. We do have people working in space technology there.
Will: [00:06:02] Exactly.
brad: [00:06:03] What? Our last question. All right. This is going to be the longest cold open in the history of this podcast
Will: [00:06:07] Oh, we're not even close to that
brad: [00:06:08] I, I take it. I take it. You haven't read the Jont. I don't think you got my reference a minute ago.
Will: Nooo
brad: Oh man.
Will: What’s the jont
brad: Okay. I, I can't, I shouldn't tell you because I think you should read it.
It's a Stephen King short story about teleportation. The, that you should really read. If you want to look at these numbers and feel, um, that sense of creeping dreads.
Will: [00:06:57] Welcome to Brad and Will made a tech pod. I'm Will.
brad: [00:07:00] I'm Brad. Hey. Hello. Hi.
Will: [00:07:01] Hi. Hi. Hello. Hello? Hi. Hi. Hi. Hi, uh, we're doing a podcast Brad
brad: [00:07:06] Oh crap. I should, I gotta go get my
Will: [00:07:08] when mr. Softy, the ice cream truck came to town. I put on a pair of dungarees last Thursday, so
brad: [00:07:14] Fair, You don't want to show your knees to the ice cream guy. You got to, you need a, you need do some level of decorum when you go buy ice cream
Will: [00:07:19] it's, it is like, if you are going to order a ridiculous ice cream cone for an adult, like something rolled in sprinkles or dusted with Oreos ground up or something like that, you need to have a certain modicum of decency, I think.
brad: [00:07:32] Gotta put on your business pants.
Will: [00:07:34]Exactly, Um, so we've got an interesting topic today. Um, we were talking the other day about early video cards and I had, I guess I hadn't forgotten, but I hadn't considered, there was once a time in the early days of 3d accelerators when there were at least seven or eight different vendors, actively selling cards in that space.
brad: [00:08:03] Yeah. And I'm, I'm sad. We didn't save this exercise for this episode. We played it out on discord last night. But, but, uh, we kind of, I tried to run down like kind of walk down memory lane and remember what all those companies were, and now I kinda know what they are, but we should try to recreate it here just for the sake of the audience.
Um, also just a little bit of context for this episode. I don't know about you, but I think for me, like I've just had this topic on the brain a lot, because this is the first time in, I don't know, four or five years, at least that I felt like genuinely excited about new PC hardware. Between the, between the 3080s and the, the new AMD, the new Ryzens and stuff like PC hardware for the longest time for me has just been like, Oh, I guess I'll get a new CPU.
Sure. Why not? Like it's cheap enough right now, you know, like there's not, there's just not been a lot to be excited about. They're like the 1080, I feel like was that like, that was a big leap forward. Like, you know, maybe the RTX cards were interesting, but not as fast as you wanted the 20 series, but like right at we're at a sweet spot right now where I'm like, I'm actually staring at hardware builds a lot right now, again, and like fantasizing.
So that takes me back to the mid nineties and those heady days of 8,000 3d accelerators coming out all the time.
Will: [00:09:11] Well, you'll see right now, you'll see an actual, real world performance improvement for like spending money on new hardware, right. In a way that you often don't, which is exciting. Uh, so. Uh, let's I guess let's, let's start with like, obviously everybody remembers 3d FX. Um, and then Nvidia had the reva128, early on, uh, ATI rolled out the 3d rage.
Well,
brad: [00:09:45] Hang on. I'm going to, I'm just going to play act what I did last night. Okay. So what I was able to come up, come up with off the top of the head was the first one that came to mind was the rendition verite. I don't know why,
Will: [00:09:55] original direct X card. Yeah.
brad: [00:09:57] uh, the rendition verite, obviously the 3dfx, like the voodoo stuff was like the thing, as soon as it came out, like nothing else mattered for a while there.
Um,
Will: [00:10:04] well, actually true. It wasn't until Quake came out that the voodoo stuff really mattered.
brad: [00:10:08] well, yeah, GL quake was the thing that, that made, made three accelerators a thing, frankly. Um,
Will: [00:10:13] I think you've got mech warrior or something with your, with your Voodoo card, mech warrior two. And it was
brad: [00:10:18] That might be right?
Will: [00:10:19] a lightly texture maped version of the software rendered mech warrior game.
brad: [00:10:24] not that 3d accelerated also. I think I bought that game at a Sam's club, so I really lost out there.
Will: [00:10:29] I bought it multiple times. I'm sure I came with every video card you bought for a while. There it
brad: [00:10:33] I, you know what I would have, I would have bought it just for that soundtrack multiple times. So that's, that's totally fair. Uh, So for whatever reason, I came up with rendition and 3d effects and then I stalled out and started thinking about like number nine and Matrox, which I don't think fit into this conversation that well
Will: [00:10:47] Well, so number nine, never really released a 3d accelerator. They were wholly focused on 2d and video is my understanding and they kind of disappeared. I think they got bought by somebody. I don't, I don't know, actually I don't actually know what happened to them or nine. Um, I also don't think that they're a bit of selling graphics cards.
With, um, uh, w with Beatles tie-ins, uh, would probably fly in today's world, but like,
brad: [00:11:16]Why not? I mean, I, I, mean, I agree with you. It was a bit too Tweed for my taste, but
Will: [00:11:22] I feel like the people at Apple records would probably have something to say about that now in a way that they would not have been aware of back then.
brad: [00:11:28] You're probably right. Like I'm sitting here looking at their Wikipedia page and some of their products for people that are not familiar, number nine revolution, imagine pepper
Will: [00:11:38] I feel like revolution in
brad: [00:11:39] ticket to ride
Will: [00:11:40] yeah ticket to ride was where they went off the rails in terms of like,
brad: [00:11:43] naming a graphics card ticket to ride is like, yes, absolutely. That's that is the bridge too far
Will: [00:11:48] I mean, that said though, there's a board game called ticket to ride, right?
brad: [00:11:51] Well that’s fine. I mean, that's probably got trains in it or something.
Will:It does have trains in it.
brad: I mean, I'm fine with that, but that that's not a video card, you know?
Will: [00:11:56] Um,
brad: [00:11:57] so yeah, like, I, I mean, we'll probably touch on this some more, but for whatever reason, my mind went to Matrox and number nine, which were more like 2d graphics cards, which is like, for people who were not doing PC stuff in the nineties, yes.
You used to just have to buy a 2d graphics card. Um,
Will: [00:12:11] one in the VL bus slot.
brad: [00:12:12] right. So you reminded me that ATI was active back then for whatever reason in my mind, ATI was not a player until the Radeonstuff, but obviously they were in it way before that.
Will: [00:12:22] So, yeah, they, they made 2d cards, like that's where they started 2d cards and video processing stuff. And they were, um, uh, they were like a real, they were, they were like in the 2d graphics card realm, there was like S3 kind of at the low end performance, but not very good. Maybe not like didn't the image quality.
Here's the fun thing. 2d graphics cards, image, quality, mattered. Like some of them were better at it than others.
brad: [00:12:48] that's why I, I mean, we're kind of getting back on to 2d cards and off the topic of 3d accelerators here, but like I lusted after a matrix millennium. For so long, because I think, I think whatever, like whatever crappy no-name pre-built computer we had at home had a similarly crappy no-name 2d card in it.
And the desktop desktop literally did not look right. Like, I don't remember. I don't know how to qualify that because in my mind, like, I can't, I can't, I can't call it mind what the desktop looked like, but I think things were like maybe a little too fuzzy or like, it was something about like the, the VGA output was not clean enough or something like that.
I forget.
Will: [00:13:22] So part of it was that, um, the, the D the Ram dac, the thing that converted the digital signal into an analog signal went out across VGA. If they were slow or bad, you'd get fuzzy signals on the VGA, but also it's about color reproduction and like window rendering speed and stuff like that. Like in windows 3.1, uh, with the right video card, you can preview the window while you drug it around.
And if you didn't have the right video card, it would just be like an outline, a box outline of an empty window,
brad: [00:13:54] Yeah. The little dotted line outline.
Will: [00:13:55] and then you could watch it literally draw back in from the top to the bottom. If you had a slow video card.
brad: [00:14:01] our card was not that bad, thankfully, but, uh, but I, I remember, I remember wanting a millennium just because yeah. It accelerated windows stuff more and it had very nicely clean, clean signal.
Will: [00:14:11] 32 bit baby. Um, well, so the thing, the thing that happened around the time 3d accelerators started taking was that the one previously been low and mid range today, parts became pretty good, like good enough. Right. So S3, ATI, um, revolution was always on the high end, uh, or sorry, a number nine was always on the high end, um, stuff like that, but so, so Matrox and ATI , both, uh, fine Canadian companies. The rage two was kind of, I think it was billed as a 3d accelerator and you could get, get it with a copy of a mech warrior too, but it was like kind of lightly texture mapped, but they're not really was the, was the thing I recall on that one
brad: [00:14:57] was that was that their first major card, the rage 2 in, in 3d space. And, and was that an all in one card or was that an add-on
Will: [00:15:03] So it was an all in all of these were all in one cards except for the, except for the, um, the 3d FX stuff. They weren't, uh, Oh, and power VR power VR was 3d only too.
brad: [00:15:13] Yeah. So we haven't just had fully, before we move on, just to finish our little rundown of the roster here. Um, s3, I guess was huge, but I had completely in my mind, like Stover, the fact that they even existed at that time
Will: [00:15:26] S3 was a big deal, I guess
brad: [00:15:28] It took me a minute. It took me a minute, but I, the name S3 verge finally floated back into my, into my consciousness.
Will: [00:15:34] V I R G
brad: [00:15:35] right. I couldn't, but I couldn't remember if it was ER, or I R
Will: [00:15:38] So they sold. Buttloads of 2d cards into like Compaq and Packard bell and like all the, they sold a lot to OEMs, the big OEMs and, um, they also would kind of run. So one of the, one of the D the defining gate for me on 3d accelerators in like 1996, 97, 98 was, would it run quake? Like in the, in the days before direct X, before direct 3d.
And, um, more like the second wave of 3d accelerated games that started in like 97. Like it was, if it didn't run GL quake, it didn't count as a 3d accelerator, even if it did like a little bit, a bit Mitt mapping and stuff like that on some textures. Um, and the rage 2 wouldn't run quake, the Matrox millennium wouldn't run quake.
The S3 cards wouldn’t run Quake.
brad: [00:16:30] The millennium was just a 2D card, right?
Will: [00:16:32] Well, yeah, but I mean, yeah, it wasn't until the mystique that, that Matrox had a card that would run quake kind of
brad: [00:16:37] they tried.
Will: [00:16:38] um, the, so the first quake card was the rage pro, which was also one of the first AGP cards, which we'll get to in a minute. Yeah. Um, okay. So S3, I don't think S3 ever in the time period that mattered made a compelling 3d 3d card.
brad: [00:16:56] would you categorize them as an, also ran in this space?
Will: [00:16:59] So one of the things that's interesting about any Oh, well, hold on. We have a couple more Intel was making 3D accelerators.
brad: [00:17:04]So Intel, when you threw that out there, I was like, wait, what? I don't
Will: [00:17:08] Yeah.
brad: [00:17:08] remember that in the slightest
Will: [00:17:10] So they made, um, they partnered with Lockheed and a company called chips and technologies that Intel ended up buying a little bit later, um, and released the i740 , which was a, another early AGP card. Um, and it was, if you recall, Ram was really expensive and hard to get. Then during that time period,
brad: [00:17:30] Vaguely.
Will: [00:17:32] So, like there've been, there'd been some sort of disaster, an earthquake or a fire or something that cratered, uh, ran production in 93 or 94.
And it took a long time to come back online. Cause it was a different time back then. So, so a lot of the AGP design was, was set up so that you could run video stuff out of system memory rather than having to have. You know, eight, eight megabytes of Ram in the system and another four megabytes of Ram on the video card.
brad: [00:18:01] which was an astronomical amount of ram
Will: [00:18:03] It's shocking. It was shocking
brad: [00:18:05] at that point,
Will: [00:18:05] Hundreds of dollars. Okay.
brad: [00:18:06] I might've had 16 megabytes in my PC at that point.
Will: [00:18:09] Oh really? I think I probably had eight and was pretty happy. Anyway,
brad: [00:18:12] maybe, maybe it was eight, I forget, but, uh, I mean, that's back when Ram was like $50 a megabyte.
Will: [00:18:18] more than that I paid at one point I paid 150 bucks per megabyte for a two megabyte upgrade.
brad: [00:18:23] Wow, that is physically painful to hear. So I,
Will: [00:18:27] That was in like 93 though
brad: [00:18:28] okay. Yeah. I, so I just glanced at the, the Intel i740 page, the Wikipedia page, and that came out in 98. So we're kind of, we're kind of swinging back and forth across a, like two, three year time span back, back and front here a little bit.
Will: [00:18:41] yeah, it was, it was, it's a weird time, but because like during the launch of the first version of like the Reva 128 launched at a time when like probably the only 3d accelerated things were quake and unreal and maybe, um, like terminal, there were some, probably some 3d realms game, like terminal reality or something that was 3d accelerated at that point.
brad: [00:19:03]Sure that sounds rights.
Will: [00:19:04] like mostly. Yeah, it was, it was, and there were some flights SIM I feel like Falcon wasn't Falcon 6.0, 3d accelerated for the first time maybe,
brad: [00:19:12]Oh was it, was it that far along or thought they were on like 3.0,
Will: [00:19:15] Windows 3.0 yeah. It was the one with the big binder.
brad: [00:19:18] Yeah, 3.0, came out in 91. Did they ever even get past four 4.0, I thought that was the last Falcon
Will: [00:19:23] Maybe I, maybe I'm maybe I'm conflating two different things.
brad: [00:19:27] 4.0 was 98. So that was probably 3d accelerated at that point. So I'll just kind of start back at the start before we get too much into the AGP and the late nineties stuff, like obviously like 3d effects, like a friend of mine got a Voodoo card.
No, not long after GL quake came out and like. I went over to his house and saw it and fell out of my chair. You know, like the defining moment in my history with both technology and video games. But like, did you have much contact with that stuff before the Voodoo came along? Like, I'm trying to get a sense of what the w what was in the market before the voodoo, but when that kind of, you know, that obviously that revolutionized everything, but like
Will: [00:20:05] Well, so like I bought a rage two card when I built a PC the first time. Um, or, uh, and put that in. It may have been even before that it may have been in my, in, in, uh, the Pentium 60 that I bought when I went to UT, uh, and. Okay. I was kind of unimpressed with it, like 3d, 3d accelerators until the Voodoo card seemed like, kind of like, it seemed like a cool thing for SGI to do.
And it seemed like something that wasn't real for people at home. So like, like if you played like the rendition V quake port that they made. So, so the other thing that's important to note is that up until quake. Most games only supported external, like a proprietary vendor, specific APIs. So like glide for 3d FX.
brad: [00:20:53] the famous example. Right,
Will: [00:20:55] Exactly Um, verite, uh, rendition wrote V quake, which was for their verite would be one that would be 1000.
brad: Oh V Quake
Will: Um, and I think Carmack helped with that obviously.
brad: [00:21:05] So I'm not to derail, but I'm sitting here. There is, of course there is a GitHub repo that has every, every, every plan file update he ever wrote. So I'm just sitting here control, effing it. Um, and it's funny, like, so quake came out in mid 96 and I'm sitting here looking at an entry for December 13th, 96, where he says for the record here is my impression of the 3d hardware I have worked with blah, blah, blah.
Will: [00:21:28] I, this was a defining moment for me. This, this plan file.
brad: [00:21:31] So basically the two consumer cards on this list are the rendition verite at $150. And the 3d effects voodoo at $300 and then like eight or 10 workstation cards that no Mortal could ever have contact with
Will: [00:21:47] 3d labs. One that SGI Erik stuff.
brad: [00:21:50] the Intergraph intense 3d, the SGI, Oh two, the SGI impact. The let's see,
Will: [00:21:56] Just to be clear that SGI 0 two was like a $50,000 computer, as I recall.
brad: [00:22:00] But literally this is this SGI impact entry. He said here a full system was literally 25 to $50,000 and SGI infinite reality was over a hundred thousand dollars.
Will: [00:22:09] Yeah. And those, those machines, those boxes for those things were huge. Like they were like mini computer type spaces, not, not PCs.
brad: [00:22:18] His quick writeup of the SGI, infinite reality, which is the a hundred thousand dollar workstation fill rate from hell polygons from hell. If you don't trip up on state changes, nothing will come within shouting distance of the system. You would expect that. But the reason I bring reason I bring this up is that like, like I said, outside of the, the, the verite and the voodoo, like everything else on here is extremely high end workstation hardware.
There just wasn't that much out there. Sorry. There's one other card on here. The 3d labs per media for $300.
Will: [00:22:46] That was a workstation card.
brad: [00:22:47] Yeah. That's okay. Yes. A well-supported low end open GL card. So I guess, yeah, that would be for workstation stuff.
Will: [00:22:52] So it was a thing. If I recall correctly, it had a, it was the least expensive way to get a fully fledged open GL stack because the, the open GL driver for the 3d 3d effects cards was just a wrapper. So it wrapped the open GL calls to guide calls. Um, and it wasn't like it wasn't a full implementation and because that card was a standalone 3d accelerator and couldn't do something windows, a lot of open GL stuff.
Like you couldn't run your open GL CAD programs on that. Um, also probably cause it didn’t have enough memory.
brad: [00:23:25] and so that's why you were having to get a bespoke version of quake for every card, right. And V Quake, GL Quake. Cause they were all being written directly to what those cars could do.
Will: [00:23:33] Well. So GL quick was written for open GL,
brad: [00:23:35] or, right, right. Yes. It would run on any GL hardware right.
Will: [00:23:37] Yeah. V V quake that's. So Carmic did V quake with the rendition people, I think. And then after that, he was like, I'm not building a bunch of proprietary quakes. There's no other, we're going to do this in open GL and like that, that immediately made open GL a gaming API.
When did it previously been for, um, workstations and CAD and stuff like that?
brad: [00:24:00] previously have been for doing your effects in Terminator 2 and Jurassic Park.
Will: [00:24:03] Pretty much. Yeah.
brad: [00:24:04] and stuff like that.
Will: [00:24:05] it was a weird, it was a weird time like,
brad: [00:24:07] It was a really weird, interesting time with all this stuff being very emergent and not well-defined like I'd not to, we won't harp on his plan files for this whole episode, but if you scroll down. Two, two more entries past the one I was just reading from you.
Get to what I consider probably the Seminole plan update, which was December 23rd, 1996, entitled open GL versus direct 3d
Will: [00:24:28] Oh, and this was when they were talking about, this was before direct, 3d was even out, right. Or 96, you said.
brad: [00:24:34] Yeah, this was December 96. So that's like six ish months after quake came out and
Will: [00:24:39]Oh so that's when direct 3d 3.0 direct X 3.0. Had launched in September of that year. And we were like direct X one and two were kind of like, not worth speaking of.
brad: [00:24:50] right. And like, this is a, this is full on like soapbox mode. He's just here kind of opining about how these things work and how much he likes or hates them. And there's like straight up, they're straight up code snippets in this thing of like showing how Vertex operations work in, in each one and stuff like this.
But, um, it was just such a, it was such an early time and this stuff was so ill-defined that like all over the place, you know?
Will: [00:25:12] Well, I mean, the big argument was that the direct X API APIs kind of limited what you could do. Well, it didn't give you direct control over the hardware in a way that was it, abstracted things out a little bit further than people like Carmack wanted. Um, I think in the, in, in the, with the idea that it would be easier for normal human programmers that weren't John Carmack or Tim Sweeney or Charlie Brown or whoever to build off of,
brad: [00:25:40] Right. Well, it also became necessary in a world with like 800 different cards on the market, right? Like at some point you can't, you can't be writing directly to the hardware of every model of card out there.
Will: [00:25:50] Well, so then it shifted the burden from the, from the developers to write for the individual hardware to the driver vendors to like not Jack up the drivers.
brad: [00:25:58] Ah, yes, which led to our current state of 500 megabyte graphics driver downloads every two weeks.
Will: Yeah Exactly
brad: Uh, That makes sense. But, um, was there a lot of like Fud around Microsoft? I, I feel like I vaguely remember like people just,
Will: Jesus was there fud
brad: I mean, this was, this was, this was squarely in the middle of the, like people writing Microsoft with a dollar sign and an era of people not trusting Microsoft,
Will: [00:26:23] this is when they were being, this is when the antitrust suit was in the, in the beginning barest beginnings. Right?
brad: [00:26:28] guess, I guess so.
Will: [00:26:29] This is when they were doing the stuff that got them sued by the federal government for,
brad: [00:26:33] that's fair.
Will: [00:26:34] yeah. Um, it
brad: [00:26:35] Embrace extend extinguisher.
Will: [00:26:37] yeah. I feel like, I feel like a lot of the, the concern about the concern about Microsoft owning the APIs that drove 3d was the same thing that we hear people talking about with Facebook and VR right now where, you know, you, you didn't want this monolithic company to own the thing that the future, you know, the API is that the future was built on.
Um, it ended up like the thing that ended up happening is. As the, as the hardware vendors coalesced into from eight or nine vendors into two vendors, or maybe three, if you count Intel, um, the, it ended up being the hardware vendors, driving those APIs as much as Microsoft. So it like, it ended up working out.
If we had only ended up with one. If, if we only ended up with either Intel or AMD, ATI and Intel, I think we'd be in a much different situation. And we wouldn't be seeing the kind of performance that we are on graphics today. So it was lucky that at least two companies survived,
brad: [00:27:36] because they pushed it forward so much. I mean,
Will: [00:27:38]They push each other. Yeah, the competition's good.
It turns out.
brad: [00:27:41] yeah. Uh, before we move on too much from the beginnings here, I happen to have the rendition Wikipedia page in front of me. And I just have to mention the naming on some of this stuff.
Will: [00:27:50] Yeah.
brad: [00:27:51] They had a, they had an API for Dos called speedy 3d,
Will: [00:27:54] That sounds great. How fast was it? Was it speedy
brad: [00:27:57]I don’t know.
I just like, I just can't help thinking about like speedy Gonzales here. Uh, scrolling, scrolling down some of the products that were released. Sierra wait, who that can't be. That's not the game maker. Sierra. That's gotta be really
Will: [00:28:10] think that somebody, I think that's what became Cirrus logic,
brad: [00:28:13] No it, no, it totally your links to the Sierra online Wikipedia page.
Apparently Sierra put out a rendition card called the scream in 3d.
Will: [00:28:21] Oh, yeah, we had one of those in the lab at maximum PC. It was, the box was on the wall.
brad: [00:28:26] I totally I'm sitting here Googling the box art for screaming 3d right now. I need to see that.
Will: [00:28:32] Oh, the box, the box art for these was all bad.
brad: [00:28:35] if you think, if you think video card boxes are ridiculous, now I'll go back to the mid nineties.
Will: [00:28:39] Oh yeah.
brad: [00:28:41] But, uh, anyway, so yeah, like, I don't know where to go with this. I mean, rendition was out there.
I, in my mind, like some months before the voodoo came along, the quake was a thing.
Will: That’s correct, yeah
brad: And then the Voodoo came along and like changed everything of course. But, um, so touch on something you mentioned earlier, I did not realize that it was the, I didn't realize the voodoo was the exception to the rule about being an all-in-one integrated card, like, cause the, the voodoo was an add in card and you had to run a literal analog pass through cable.
Will: [00:29:10] You plugged your video, you had a little short VGA cable that plugged from your main video card into the Voodoo card, and then you plugged
brad: [00:29:16] desktop card
Will: [00:29:17] And then you plugged the monitor into the Voodoo card
brad: [00:29:20] and then
Will: [00:29:20]And then literally do a state switch. Like the Voodoo card would detect that you would launch something that you used glide and it would turn off the pass through and just render itself.
brad: [00:29:28] You would add, you would, at least with the card I had, which I believe was the orchid righteous. 3d is the one I
Will: [00:29:33] That's uh, that was, uh, that was one of the two. It was that, or the diamond monster 3d.
brad: [00:29:37] it was not the diamond monster, which was the gold standard to most people.
Will: [00:29:39] I think they were both exactly the same cards.
brad: [00:29:42] all, they were all practically the same, except I have now seems like the time to touch on my teenage lust for the cannabis pure 3d,
Will: [00:29:49] Now, that was a card.
brad: [00:29:50] which was just a voodoo card, except it had two extra megabytes of memory.
I think it had six megabytes of memory on it.
Will: [00:29:55] well it's going to Canopius did some other weird stuff too. Cause they had an integrated card that was, um, like a 2d card with a daughter card on it. That was the 3d FX card. So it was all in one thing. And it also Canopius uh, did internal pass throughs if you had a Canopius 2d card and a Canopius 3d card.
brad: [00:30:15] maybe that's what I was interested in. They were,
Will: [00:30:17] well, so.
brad: [00:30:18] Jeff, I believe they were, that was a Japanese company.
Will: [00:30:20] So, yeah, they made arcade boards was the other thing that they did
brad: [00:30:23] Oh, interesting.
Will: [00:30:24] they did those obsidian. Uh, the, I want to say, what was that? What was that? Jet? The jet ski race game, arcade game. Um, Anyway.
brad: [00:30:40] pretend we remember.
Will: [00:30:41] Yeah. Um, people are screaming at their radios right now, but they made a bunch of 'em. They, they made the 3d, like for a lot of those early arcade cabinets that were, that were driven by 3d effects hardware. They made those boards and put enough Ram on them that they could run it like arcade resolutions.
Um, and they would also jam multiple texturing units on which, which is like, that's the, that's the interesting thing. So first off, all of these cards are passively cooled at most. They had a heat sink, which was sometimes viewed as a negative in the, in the press at the time.
brad: [00:31:15] Right, right. Yeah. I'm sitting here, I'm sitting here looking at the cannabis Wikipedia page, which is very short, it's like three paragraphs, but it says when the voodoo two was released, the cannabis pure 3d two was praised for the fact that their cards were shorter than the competitors. Voodoo two cards.
Will: [00:31:28] yeah. Yeah. I'd fit in more. Well, like, so I had an AST. Pentium 60 at that point, I think. And that card was a pizza box case. And like, there were a small number, like if you had a giant long chunk or a video card, it just wouldn't go in there. Um,
brad: [00:31:46] yeah god was like, was the mentality basically that if you need a heat sink on your card, you have screwed up on the design end somehow.
Will: [00:31:51] Pretty much. Yeah. Like my Pentium 60 didn't have a heat sink. It was just a bare chip
brad: [00:31:54] I, my, my, yeah, my, my Pentium 75 did not either. And that machine became quite unstable over time.
Will: [00:31:59] shocking that, who knew
brad: [00:32:00] Uh, I mean, granted, to be fair, I had overclocked it to 75.
So that probably, or no, no, no, I'm sorry. It was a 75 to 90.
Will: [00:32:06] Yeah, that was the jump.
brad: [00:32:07] I had at 75. And it was not stable at a hundred, but I could overclock it to 90 and then it became borderline unusable after about a year and a half.
Will: [00:32:16] Yeah. Uh huh. Weird, weird. That.
brad: [00:32:17] uh, yeah, that was going to mention with the, um, With the pass through on the Voodoo cards like you would here.
I don't know if it was just the one I had or maybe they all did this, but you would hear like, you would hear that good solid and like metallic thump sound. Every time it switched over to the 3d card.
Will: [00:32:31] Yeah, I think that was in the monitor though. Wasn't it? Cause it was a refresh change probably
brad: [00:32:35] maybe it, maybe it was the monitor.
Will: [00:32:37] a good chunk.
brad: [00:32:38] Yeah,
Will: [00:32:39] It wasn't there wasn't like a physical switch in the cards though.
brad: [00:32:42] no, no, no,
Will: [00:32:43] Yeah. There was no solenoid popping or anything like that
brad: [00:32:44] I think, you think it was more of a chunk than thunk?
Will: [00:32:46] so they did, uh, 3d effects. Did the, th there's there's a lot, like, it was a weird time because also like, so what a 3d, these things weren't called GPU's cause there wasn't a lot of processing.
Basically. They were just like texture mapping engines on a card.
brad: [00:33:03] I think, I think three accelerator was the accepted term back then. Right?
Will: [00:33:06] That’s what we call them. Yeah. And the, and they would, um, basically what, what the CPU would do, all the triangle set up and all that stuff, and it would establish a mesh and then the video cards would handle Z buffering. So they'd see what was visible in that scene. And then they'd apply textures to it.
And like, there were a wide variety of approaches to that power VR did this tile based thing where they did that, where they didn't do visibility checks. Um, on the whole scene at once, they would just do the little chunks and then that would save them a buttload of memory bandwidth, basically. Cause they were only pulling textures for things that they would actually map.
Um, I, as I recall, that was also one of the ones that had problems in their open GL stack, where there were giant cracks in the world where the tiles didn't line up. Right. Or something. I don't, I don't know them. I don't know exactly what was happening there, but I remember. In that, in that experience, in that instance, it was bad.
The other thing was weird about power VR was it was a standalone 3d accelerator, but it, it, you could do 3d in a window on that, uh, because it, it, it passed over to the 2d cards, frame buffer across PCI
brad: [00:34:14] Oh, interesting. Right. I don't know if we mentioned it. That was another limitation of the voodoo was that it could not be window three only full screen.
Will: [00:34:21] no windows 3d on the voodoo . One or two only on the rush and the banshee and the three eventually
brad: [00:34:25] other, right, right. Those were their integrated cards. The other downside that I didn't mention of, and then maybe this was also a model specific, but it made my 2d desktop look like shit,
Will: [00:34:34] It made everything fuzzy.
brad: [00:34:35] which made me want to die because that was right after I got to Matrox millennium. I finally, I finally got the good 2d desktop card and made everything nice and sharp and smooth.
Will: [00:34:44] Oh no,
brad: [00:34:45] then I get, and then I got a voodoo and it made it look like shit again.
Will: [00:34:47] so, so the other thing about all of these cards is that they were really low resolution, like, like at a time when a normal desktop was probably 800 by 600, I don't think you could run a voodoo one card above 640 by 480.
brad: [00:35:01] no, definitely not at any performance. Maybe not at all, but definitely not at, at any level of performance. Yeah, sure.
Will: [00:35:07] Yeah. And, and, um, Like, like the thing that, the thing that was amazing about those early cards is you went from like these kind of not grainy. Exactly. But it just felt, it felt it was real grungy software rendering in quake, wildly technical Marvel was, was kind of scuzzy feeling. Um, and then when you load it up, something that could mipmap on the textures, which is basically where it looks at as different, different resolutions of the texture and the video card knows to drop the right one in, at the right distances. You've got these unbelievably clear textures. You're like, wait, is this what this game is supposed to look like?
And I played it all the way through not looking, looking like this. This is, this is kind of magic. And then, you know, games like. And, and, and still, like, if you're looking at a texture that was, that would have had multiple mipmap levels across. Like, if you're looking at, if you're looking down the length of a wall, right.
You would get one of the mipmap levels in the wall. Cause we didn't have anisotropic filtering or trilinear filtering or anything like that on the cards at that point. So it wasn't able to sample different MIP levels based off on where the, like for the different distances of that same texture that you were looking at, um, But it was, it was still kind of incredible.
And then when they, when the voodoo two came out, all of a sudden you had Mulit-texturing that could happen in one pass. So like each frame could have multiple two textures applied to each surface. So you'd have like a lighting texture and a, and a, an, uh, an, a picture texture, right.
brad: [00:36:40] Trying to think of some other really real world use cases for that. I mean, is that when you were able to start applying like decals to walls, like bullet holes and blood and stuff like that, or.
Will: [00:36:47] That was probably later. Cause I feel like the Voodoo 2 actually had, like, there were three chips on the texture on the, on the chip. That was one that was the main, the FX chip. And then there were two TX chips that each was responsible for one texture pass per cycle per frame. Uh, so, so you, you could do that before, but it would take multiple.
Multiple passes on each frame to do it, but to do a Multitexturing pass is my understanding, but it could be wrong. I don't, I don't know this was, this was from a time before when people talked about this stuff, like, you know, now Nvidia and ATI, when they release new GPU's. They dump. They, they sit down with all the tech press and they're like, okay, here's exactly how this stuff works.
Here's how deep the pipelines go. Here's how much cache each part has. Here's what the, what the, what the program will, what parts of the pipeline are capable of. And back then, they just released the cards. And there was a one page spec sheet that they handed out to people at SOC Comdex when they wanted to buy, when the, in there came in and like, Compact rolled in and they're like, well, here's, here's what our new video card can do.
It's like, it costs this much. It's, uh, you know, it's going to get, uh, this many Blit marks and, uh, it'll run quick and people are like, okay, in we're putting that in every compact that we're shipping this year.
brad:Good enough, ship it.
Will: Yeah. Um, it was, it was a weird, it was a weird time for sure.
brad: [00:38:07] was the, was the voodoo 2 another pass through card that was still not a
Will: [00:38:10] Voodoo 2 was a pass through card..
brad: [00:38:11] it's still not at a full desktop card.
Will: [00:38:13] But also the Voodoo two had an internal SLI ports. You could put two of them in.,
brad: [00:38:18] Wow God, is that when SLI started
Will: [00:38:19] That's what I, that's how I started. And it would let you play it, I think 1280, 1024 x 768, if you had two of them in SLI. Um, and the funny thing is because of the way that card worked, it is maybe the last time that adding a second video card to your system gave you a one-to-one increase in performance for each, like you get double the performance for each assuming, assuming again, because all of the triangle set up and all of that stuff was done on the CPU, assuming your CPU's had faster and a floating point to do that.
brad: [00:38:48] So they, they set very poor expectations for SLI early on that it never lived up to again,
Will: [00:38:52] Well, but also like when I upgraded from a voodoo one to a voodoo too, I was on a K six, like one 66 or 200 or something that had bad floating point, but good integer math. And it turned out that I saw no performance increase going from the voodoo one to the voodoo two, because the CPU couldn't keep up with the card at all.
Everything was CPU bound then. Okay. Uh, so,
brad: [00:39:20] like, I, there's a couple of stories here. I mean, there's, there's the thing where 80% of these companies just fell away and either got acquired or went out of business or whatever.
Will: [00:39:29] well that happens when there's a big, a big shift in the way, like. It happened with phones too, right? Like when, when the iPhone came out and Android rolled out Nokia and, and, uh, Motorola and all those companies that had been making dumb phones for, for 10, 15, 20 years. Uh, a lot of them didn't survive that transition.
brad: [00:39:50] I mean, I guess it's the natural evolution of a new market being created essentially. Right? Cause that's what happened with phones and that's what happened here. You know, there, there was no market for 3d accelerators and then there was, and so everybody tried to get in and then only a couple of them made it out.
Will: [00:40:01] and, and even the ones that did make it out, like 3d FX got bought by Nvidia and like in the early two thousands, right?
brad: [00:40:09] So that's, that's like, there are two stories here. There's, there's the story of when I got a T and T and like everything changed like that was that like, the voodoo was like an amazing proof of concept, but the TNT was the first nvidia card for people who don't know. Um, so like, those are the kind of twin narratives here are.
I got my first Nvidia card. And granted, I had sounds like the first couple of Nvidia cards were not amazing, but the TNT was kind of mind blowing at the time. Um, so there's like kind of the rise of Nvidia and the all-in-one card. That was actually very good. And then there is the like meteoric rise and fall of 3d FX.
Like 3d FX went from the name in 3d graphics to not even a going concern in the space of like what four years
Will: [00:40:50] Five years, I think. Yeah.
brad: [00:40:51] Like they got, I think they got acquired by Nvidia in 2000 ish.
Will: [00:40:55] It was right when I started at maximum PC, as I recall, they send us a voodoo five, 6,000, like a prodo board with like wires dangling off of it. That, that where it was an engineering sample,
brad: [00:41:08] like did the Voodoo 5 even make it to market or it just barely
Will: [00:41:11] the, maybe the low end version did, but the big four, the four processor, one didn't never made it to market.
brad: [00:41:16] Right. So like for them to go from the card to get in the voodoo and voodoo two era too, like I assume they basically sold at fire sale prices to, to Nvidia.
Will: [00:41:27] I mean, I think they bought IP and, and engineers, and that was it
brad: [00:41:30] right,
Will: [00:41:30] um, the, so the thing, the thing that's the TNT is it is an interesting card because like TNT and rage pro. Like both of those are really important. Rage pro was in a lot of gateway machines as I recall, and Dell machines. So like, if you were buying computers for (sneeze)
Excuse me, uh, say a computer lab in 2000, 98, 99, all of a sudden, like a commodity computer could run Quake. Right? So it wasn't like you needed some weird add in video card that made you 2d desktop look like crap. And, uh, and, and it was going to have all sorts of weird requirements. It just worked
brad: [00:42:10] And that was a GL quake. You were talking
Will: [00:42:11]That was GL quake.
Yeah. Um, the, the T and T was the higher end version of that. The TNT was a really compelling part at, at, uh, like when, when the Riva one 28 launched there was, I think I sent you last night was there was a big interview in boot in like 96, I guess, or 97, maybe even 98, where David Kirk, who was the chief scientist at Nvidia for a long time, was basically doing the apology tour for Riva one 28 driver problems.
And, and overall shortcomings in that part. And then they rolled out the TNT, which he had hinted at in that, in that interview. And it was, A. the drivers were really solid, which it was a lesson Nvdia learned early that like without, without good drivers, it didn't matter how fast the hardware was.
Um, and two, uh, it was, it really ripped, like it was fast. You could run quake in a window, you could play GL quake, everything that, you know, and it was performant and you could play at high resolutions even.
brad: [00:43:09] That was, that was my quake two card is my memory of the TNT. I got a TNT maybe sometime after quake two came out and it was also, it was an all in one card. So it was nice to just have one card that did desktop and really nice 3D graphics. But, but also, yeah, it was fast as hell. In my memory,
Will: [00:43:27] So it's funny for a long time, I had a T and T card and a Voodoo card in my computer at the same time. Um, just because there was still like quake one, still worked better on the Voodoo card than it did on the TNT card because of driver implementation, weirdness. And I was still playing competitive quake one on the LPB leagues at that point. or the HPB league. Sorry, the high ping people leagues.
brad: [00:43:54] You're talking, you're talking to somebody who got, I got a 3D FX card and started playing quake on that because my Pentium 75 couldn't run it for shit and software. But, but then I got a Pentium 200 later and I was getting such a high frame rate in software quake on that thing that I went back to software quake because I would rather have the frame rate.
Will: [00:44:09] You were probably one of those people that would've played net quake instead of GL, instead of quake world
brad: [00:44:13] Well, I wouldn't quite going to go that far, but, uh,
Will: [00:44:16] we talked about those Carmac plan updates, but I remember like, you'd see a new plan update from Carmac and you're like, is this going to be about quake worlds? Is it about GL quake? Either one was fine.
brad: [00:44:24] Yeah, I was reading about the quake world announcement. I was reading that post last night when, when he basically like. Almost hat in hand, came in front of the quake community and was just like, without, in so many words was just kind of like, look, we get it. The internet performance for net quake is unacceptable.
I'm going to fork the code base and start over and deliver a new online client for people. Also, there's an amazing line in there. Uh, gosh, I don't know if I can find it off the top of the head right now.
Will: [00:44:56]The first time I played GL quake world was at my dad's office. Cause it came out while I was home on Christmas break. And I, I drove, literally got in the car and drove to the office and downloaded it and uh, put it on his work computer and played software quake world on his modem there. Cause we'd never computer at the house at mom and dad's house at that point.
And I was just like, Oh God, I can play this on the modem. This is,
brad: [00:45:17] it was
Will: [00:45:18] This is a whole new world.
brad: [00:45:19] It was actually playable. It felt like a real video game. This is the line from that, uh, from the plan file where he basically reveals the existence of quake world, where he describes it as a pet research project. If it looks feasible, I would like to see internet focused gaming become a justifiable biz direction for us.
It's definitely cool, but it is uncertain. If people can actually make money at it.
Will: [00:45:39] So, so I'm sure when he says he, certain people can actually make money at it. They were looking at like those pay to play modem dial-in services, like Dwango, um, where you paid like five bucks a month and you could dial into a local thing and play doom against other people in your, in your town. Um,
brad: [00:45:57] obviously there were early attempts at monetizing, that kind of stuff, but it just, just in light of history and where things have gone seeing somebody say, I don't know if you could make money at online gaming, but we're going to mess it.
Will: [00:46:06] Uh, maybe I don’t know.
brad: [00:46:08] Kind of amusing. Um,
Will: [00:46:10] but, but yeah, so we were talking about consolidation. Like if we get on this list of the eight companies, so we have 3d FX, power VR, Nvidia, ATI, Matrox, S3, Intel and rendition renditions gone. I don't know what I actually don't know what happened to them.
brad: [00:46:24] Let me see here
Will: [00:46:26] I’m sure they got bought by somebody, but, uh, 3d FX got bought by Nvidia in the early two thousands.
And for mostly for patents and engineers, it's my understanding,
brad: [00:46:35] Micron micron purchased rendition,
Will: [00:46:38] Micron the memory manufacturer.
brad: [00:46:40] that's what this says, you know, uh, as a, uh, as a source of embedded graphics solutions for their own line of motherboards,
Will: [00:46:47] Hmm. There you go. Uh, power VR ended up focusing on mobile and they are one of the main SOC G uh, GPU vendors for, um, arm processors
brad: [00:47:02] To this day, right? Like that name, that name still exists with a little detour into the Dreamcast.
Will: [00:47:07] Uh, true. Yeah, they did the V2 version of the tile vr architecture was the thing that made the Dreamcast work. They're like, look, there was a lot of scene drama around the Dreamcast because every, everybody thought they were going Sega had everybody thinking they were going in the dream cast, like 3d FX sued them for $155 million.
Yeah. Um, because they had given proprietary hardware information because they thought 3d effects was going to power the Dreamcast. Um, 3d FX did a lot of arcade boards for Sega is my understanding in that time period.
brad: [00:47:36] the, yeah, I think that's right.
Will: [00:47:38] Uh, I feel like somebody else, uh, also thought they were going in the Dreamcast and then it turned out going to the Dreamcast wouldn't have helped anyone really?
Probably
brad: [00:47:47] No, not so much that console lasted. What about a year and a half?
Will: [00:47:51] if that, yeah. Um, the, okay. So power VR, Nvidia obviously is Nvidia and they're powering the future. Uh, ATI got bought by AMD in 2006. Uh, they spun out, they had bought a bit boys pi, which was one of those hushed tones on message boards USENET companies
brad: [00:48:09] Oh, my God dude, the, the, the, the, the game that bitboys was talking about, their forthcoming product that never materialized was insane.
Will: [00:48:17] they came out of, um, the demo scene, like Swedish demo scene
brad: [00:48:20] yes, yes. They were from Scandinavia. Right? So they were, they were a bunch of like 64k coders and stuff like that. And I'm trying to remember what claims they were making
Will: [00:48:28] that part doesn't matter.
brad: [00:48:29] Well they were quoting numbers and performance metrics.
That sounded completely impossible
Will: [00:48:34] They were talking about like, like next gen was the equivalent of them talking about Ray tracing and like 2005. Right.
brad: [00:48:40] right. And, and, and the fact that they never actually shipped a product got acquired
Will: [00:48:44]Well they got bought by, they got bought by ATI right before AMD bought ATI. And then when they were, when AMD bought ATI for, uh, $5 billion or whichever it was, this was an enormous amount of money.
Then they spun bit boys back out to Qualcomm who then developed the thing that they were talking about into the Adreno processor, which also is in a lot of like that's Qualcomm's GPU, SOC component.
brad: [00:49:09] So the bitboys, people actually shipped a product.
Will: [00:49:11] And Hey, look at those letters in Adreno, Brad, is it an anagram
brad: [00:49:18] Uh, maybe, maybe,
Will: [00:49:21] for a popular ATI graphics processor?
brad: [00:49:24] Oh, Hey, look at
Will: [00:49:25] Yeah.
brad: [00:49:26] Who would have thought
Will: [00:49:26] Yeah. So, um, I think everybody feels good about that. Art X was another one of those companies that was like in the, in the 30 companies that were developing 3d accelerator technologies in like the late nineties that ended up getting bought by ATI. That's what ended up in the dolphin, I guess, which is became, became the game cube.
Yeah.
brad: [00:49:45] Which reminded me of that GameCube straight up, have an ATI sticker on them don’t they,
Will: [00:49:48] Hell yeah. Right on the front of the purple spot
brad: [00:49:50] I, if I remember just like the dreamcast had a windows CE logo on there somewhere,
Will: [00:49:55] Yup. Yup. I mean, uh, let's see, Matrox still makes, uh, graphics, video processing stuff and
brad: [00:50:03] I was shocked
Will: [00:50:03] and video wall installations and stuff like that.
brad: [00:50:05] Ida. I wish I was sure that Matrox would have been one of those casualties of this era but yeah they're still out there. Yeah. Like they make a lot of embedded solutions for like, if you've got some giant video wall, you need a power
Will: [00:50:14] Well, so Matrox always, they came around in the early two thousands to maximum PC. They were the first ones who were like, Hey, we think this multi-monitor thing is going to be a big deal. Um, and w here's a solution that will let you play your video game across three monitors stretched around. And of course they were like CRTs with big fat bezels and stuff like that.
So it was kind of goofy at the time.
brad: [00:50:38] it was too expensive and too, uh, too bulky to be feasible back then.
Will: [00:50:42] Of course you couldn’t drive it really cause there's too many pixels.
brad: [00:50:44] Yeah. But they weren’t wrong.
Will: [00:50:45] They were not wrong. Um, S3, it has been, it was bought by, um, Oh God, I can't remember who bought them. I want to say, I want to say Intel, but I don’t think that’s right.
brad: [00:50:59] I looked this up last night. I looked it up last night and I am totally forgetting here. Let me just,
Will: [00:51:04] Uh, Intel obviously.
brad: [00:51:06] they're owned by HTC
Will: [00:51:07] HTC that's right. Um,
brad: [00:51:09]In fact, yeah, S3 name I believe is still in use.
Will: [00:51:13] on mobile parts probably.
brad: [00:51:15] Um, I might be wrong about that actually.
Will: [00:51:20] Well, so Intel after the I740 and the I 752, which was the follow-up, they did multiple other, like they went through the whole Larrabee thing where they were going to revolutionize the graphics pipeline with a bunch of low fi x86 cores. Um, and then it just basically went disappeared. Uh, and rendition, you said ended up at, um,
brad: I’ve already forgotten it’s so hard to keep this, Micron.
Will: Micron. Yeah. That's right. So yeah, it was, it was weird. And like, during this time there were so many, everybody looked at what the market, what the hardware was capable of and where the market was going. And they decided different approaches to how to, how to feed those markets. So like, You know, rendition and Intel were looking at the price of memory and we're like, well, nobody's going to put lots and lots of memory on these cards ever.
So, you know, here we should, we should just roll out these high inter high-speed interfaces to system memory and let everybody do everything out of system, memory and Nvidia and ATI looked at what the CPU's were capable of in terms of triangle set up more like we're going to hit a wall on this really fast.
So we need to get. Like some, some computation on these processes. So it's not just doing mipmapping and trilinear filtering
brad: [00:52:35] thus the Dawn of hardware transforming and lighting.
Will: [00:52:38] re and yeah, that's when the GPU, when Nvidia launched the Geforce 256 and ATI launched the Radeon six months later, it was, it was a transformative moment
brad: [00:52:49] Totally
Will: [00:52:51] Oh, no, no.
I mean, in, in that, like the computationally expensive part of playing games on these 3d accelerators suddenly moved off of the CPU and onto the GPU.
brad: [00:53:01] Right like that, that conceptually is the beginning of the modern era of the GPU. Right?
Will: [00:53:05] I think so. Yeah.
brad: [00:53:06] Um, I didn't realize the Radeon was that close to the Geforce in getting to market.
Will: [00:53:10] So I started at maximum PC. This is easy for me. I started at maximum PC in June of 2000 and my first meeting with a vendor while, uh, as a maximum PC employee was a Radeon meeting. Like the second day I was there. And so they probably didn't roll it out for another three months at that point. Cause they, everybody brought us stuff super early, so we could have timely coverage.
Um, but, but it was, that was a big deal.
brad: [00:53:34] Oh, in my mind. I mean, this is probably because I always bought Nvidia products, so that's just what I remember. But like, in my mind, I don't remember the Radeon name being out there until half-life 2, but that's obviously a marketing, that's a legacy of their
Will: [00:53:44] That was way later. Yeah.
brad: [00:53:45] marketing. Right. They obviously were around way before that.
So
Will: [00:53:47] So, so the thing about. The thing about, so one of the things that Nvidia did that turned out to be really smart because of the thing that you just said is that they rolled out DDR memory was late coming.
brad: [00:54:00] yeah.
Will: [00:54:01] So they just rolled out an SDR version of the GeForce like six months before the DDR cards.
brad: [00:54:05] I was still on a T and T one. So I bought an SDR Geforce because I needed, I was desperate to have it because that was the era of quake three and I really needed a quick three to run better. Um,
Will: [00:54:18] It turns out the DDR cards since everything was fill rate limited were almost twice as fast as the SDR cards.
brad: [00:54:23]Yea that was perhaps a hasty decision. I was always off cycle for whatever reason. I got a voodoo one. But then I, I tried to stretch my hardware as long as I could just for money reasons. So I skipped the, voodoo 2
Will: [00:54:34] That was a mistake
brad: [00:54:35] I got a TNT 1.
Will: [00:54:37] Yeah.
brad: [00:54:38] And then the TNT 2 came along was dramatically better. Uh, then I got a GeForce 256 SDR because I was due for a card.
And then the DDR version came out. It was massively better than the GeForce two was even better than that. Um, I think I stayed on the, I'm trying to remember. I want to say I stayed on that. GeForce two 56 until the GeForce four.
Will: [00:55:02] probably three or four would have been the next smart upgrade. Yeah.
brad: [00:55:05] I skipped, I skipped the GeForce three cause I was, that was when I started at GameSpot.
So I was playing a ton of games on consoles. Cause that's, that was my, that was my coverage beat. So I didn't get a, I think I got like the low end GeForce four was my next card after that.
Will: [00:55:18] Well I was going to say the weird, so that's when the GeForce four, I think is when they started differentiating. Like they started having multiple skews with very, with that. Weren't just clocked down versions of the same chip. They were like versions of the chip that didn't bin or that parts were burned out on or something.
Uh, and like they had the GeForce MX I think was the low end version of those cards. And then the Tis were the high ends and then the regular ones where the, where the, just the GeForces
brad: [00:55:45] I had a, I believe I had TI 4,200.
Will: [00:55:48] what was a fine card. Oh, right. They did numbers on those. I forgot. So.
brad: [00:55:53] Which was the low end, It was the 4,200 4,440 and 4600. So.
Will: [00:55:58] So that was a weird time because the pipelines were still fixed function. So it wasn't like a general purpose compute pipeline. Like if they wanted to add new features, then they had to put Silicon on for it. Or you had to do it on the CPU. And it wasn't until the five series with the, with the leaf blower, the 5,800.
Uh, that, that, that changed, which was a few years later. So, and that, and that's when the general purpose compute pipeline happened on the GPU really that's when that started. Like, they weren't feature complete for a few more years after that because they didn't have, like, they didn't have full if an X or, you know, all the different Gates that you need.
Um, but, but it was pretty cool.
brad: [00:56:36] So that, that pretty much takes us into the modern era, both in terms of like the internal design of what a GPU is, and also in the market dominance of the companies that made it through. But, uh, the last thing I wanted to touch on real quick, or just expand on, um, was if you have a good sense of why 3d FX cratered so hard and so fast,
Will: [00:56:54] Yeah, they never made a compelling, 2d, 3d part.
brad: [00:56:58] is that all it was.
Will: [00:56:59] I mean, they, they, it, it was first mover advantage, disadvantage. Right? So they, they established the market and they set it up and then they didn't ever have a, like, their products were always niche products in a market that needed to be mainstream and Nvidia and ATI had mainstream parts immediately.
Okay.
brad: [00:57:19] They couldn’t go in commodity PCs because they were an extra part, not a, not, not a replacement part. Right? Like they going replace the desktop card. They had to be there in addition to a desktop card.
Will: [00:57:27] So it was, yeah. So instead of adding like four bucks to the cost of the thing for compact or Packard bell or Emachines or whoever, it was another 150 bucks to the OEM and they weren't like, unless they could sell it into gamers. It wasn't like, like if you look at it from a, from a, if you're the purchaser for compact, who's selling 2 million PCs a year or something.
And you're looking at ATI with the rage pro and Nvidia with the TNT too, and 3d FX with a T and T two, plus a voodoo two, or voodoo, voodoo, banshee, whatever, voodoo three Their voodoo three performance was never very good, right? Like those cards, those cards couldn't compete with the TNT two and the rage pro
brad: [00:58:09] The voodoo 3 was an all in one to be clear,
Will: [00:58:12] Yeah, everything after the voodoo 2 was an all-in-one.
brad: [00:58:14] Okay. The banshee was just straight up bad, right?
Will: [00:58:16] The banshee was straight up bad. The voodoo rush was really bad. The banshee was, was an okay 2d card that had kind of passible 3d on it. Um, the, the voodoo three was fine. Like it was not a bad card necessarily. It just wasn't. Like it wasn't as competitive with Nvidia. And I think they, I feel like they also leaned on glide longer than they should have where Nvidea was building robust, direct, 3d, and open GL drivers that were native.
Uh is using the glide wrapper for open GL for a long time.
brad: [00:58:47] now that you mentioned it, I think that was another revelation of getting that TNT was that it just played a lot more games.
Will: [00:58:52] Like I said the Riva 128 was a real wake up call. I feel like at Nvidia where they were like, Oh right. It doesn't matter how good our hardware is. If the drivers suck and, and like they
brad: [00:59:03] API support is, is lacking,
Will: [00:59:05] Yeah. Yeah. They they've spent a lot of time. I mean, if you look at the size of their dev REL org in Nvidia versus other comparable companies then, and even probably now it's, it's pretty astounding.
How, how aggressively they spend money on that it makes a sense
brad: [00:59:22]Yeah that reputation persists to this day. Like you see it on message boards all the time. And in fact, like that's in the back of my mind, like Nvidia is the one with the good drivers, like that's a big part of the reason that I always bought their cards. Um,
Will: [00:59:33] So, um, it's, it was, it was like, I thought we're going to talk about direct, 3d and open GL this and this. Some, we kind of skipped past this, but at the same time, the undercurrent behind all of this was like, what's going to be the future of 3d rendering. Is it going to be this weird open consortium or is it going to be direct 3d?
And it turns out. That the open consortium never was able to move fast enough to support the new hardware. So like people like vendors were always writing open GL extensions to support the new, like the hardware transforming lighting and the bump mapping and the normal maps and all of that stuff. Uh, and, and th just never landed on the open GL side.
It made it made supporting open GL really difficult.
brad: [01:00:15] right. I don't like that dynamic persists to this day. Right? Like DX 12 was robust and mature and on the market for ages before Vulcan really became a thing. Right.
Will: [01:00:24] But it's,
brad: [01:00:25] I don't know. I don't know if those are as directly comparable in this
Will: [01:00:28] Yeah. It's the Vulcan is complicated for a whole other reasons, but, um,
brad: [01:00:33] I, I came upon a Twitter thread of graphics programmers the other day, kind of grousing about Vulcan in a way that I hadn't seen before that maybe think maybe my understanding of Vulcan is not as complete as it could be.
Will: [01:00:44] was it the one about the here's the number of lines of code it takes to draw a triangle and Vulcan.
brad: [01:00:48] Something like that. Yeah.
Will: [01:00:51] Um, the other weird thing about direct 3d. Like they iterated direct, 3d so fast at that time period, like they never released direct 3d. There was no direct 3d one direct 3d 2.0 launched in June of 1996 directory D 3.0 launched in September of that same year.
So like four months, five months. Right. And like, I, I think the only direct 3d two game I ever remember playing was Tom Clancy's SSN that, that submarine like the, the, the underwater piloting game where you flew the submarine, like a plane. Um, and they, like, they didn't really, like Microsoft was iterating that so quickly that they were releasing three 3d.
3.0a 3.0b 3.0c, month to month to month. So you're always downloading these new things, but, but as a result, by the time they got to five or six, I guess by the time they got to five, because they skipped four, it was robust. And, and like worked.
brad: [01:01:51] I feel like I remember direct X five lasting for a very long time, comparatively.
Will: [01:01:56] five was a five was for awhile. Eight was when they added the programmable pipeline stuff.
brad: [01:02:01] Yeah. Those are like five, eight, 11, and 12 are the big milestones in my memory.
Will: [01:02:08] So the other big hardware thing, the early hardware feature that made a huge impact on, on like visual quality that we haven't talked about yet was the G the Matrox G 400, which was the post mystique. It was like their rage pro competitor, I guess, or rage. One 28 may be competitor, uh, around the same time as the TNT.
And it had bumped back, it had a hardware bump, mapping support. Which was, which was really novel and literally no one used it. So
brad: [01:02:36] because it was the only card that had it
Will: [01:02:37] it was the only card that had it. Yeah.
brad: [01:02:38] I mean, that, that kind of thing came back around with the GeForce three. Right?
Will: [01:02:42] Uh, yeah, GeForce three was the first one that had mapping on the Nvidia side. I think the G 400 might've been later when I think I might've, I might've reviewed that when I was at maximum PC. When I think about it, I'd have to...
brad: [01:02:54] it was a fun time.
Will: [01:02:54] It's been a, it's been a, it's been a while.
brad: [01:02:57] It was a big mess. It was, you know, it was, it was easy to place the wrong bet back then, because there were so many products on the market and everything was so immature and not, not, not fully
Will: [01:03:06] other hand, the cards were really cheap compared to today, like the cards were 200 bucks at the high end.
brad: [01:03:11] Yes. That's fair things have gotten a little pricier now that you mentioned it.
Will: [01:03:15] I re I mean, 200 bucks was worth more than, but when I went, I remember going to electronics boutique to buy a voodoo two. I had pre-ordered. I was one of two people in Knoxville, Tennessee, who pre-ordered a voodoo two from our, our electronics boutique.
And I got a creative labs 3d effects of voodoo two card that the serial number was like zero zero zero zero zero zero zero 32. Yeah.
brad: [01:03:39] Do you still have it?
Will: [01:03:39] Now I got rid of it years ago. I know.
brad: [01:03:42]it might be worth something.
Will: [01:03:43] probably have a picture. It wasn't I looked,
brad: [01:03:44] Oh, that was, that was the last thing I wanted to mention before we probably wrap this thing up. But do you remember the weird, I hesitate to use the word cult, but there was like an insanely diehard fan base around 3d FX, like way longer than there should have been.
You remember people, people clinging to their Voodoo's way after it was clear that the industry had moved on and like,
Will: [01:04:05] I had a Voodoo two card in my machine. Well, into the quake three era, right? Yeah, it was, um,
brad: [01:04:12] Like if you Google, I should pull this up. Like if you Google images of the voodoo five, what an obscene thing that is the voodoo five was the one that was going to plug straight in.
Will: [01:04:22] uh, Voodoo 5 6000 like I said, we got an engineering sample and it had a power brick. That was the size of like a gaming laptop, power brick that you plugged into a, like a dipole port on the back of the card,
brad: [01:04:33] Right. And then the into the wall, right
Will: [01:04:34] into the wall. Yeah.
brad: [01:04:36] Like directly into the wall did not run off the PCs power supply.
Will: [01:04:38] No, awnd it was, it was an incredibly, it was like a workstation length card. It was really, really long.
brad: [01:04:45] if you Google pictures of this thing, it looks photo-shopped, it looks like, but he has copy paste and a video card on top of itself.
Will: [01:04:49] it had four of those. Like, I don't, I don't even know what those fans were called, but it was like a really specific size heatsink fan combo. And it had four of them on it, as I recall
brad: [01:04:59] Yeah, it's gigantic. It's got a handle on it. The other end of the card has a handle on it.
Will: [01:05:04] so that was supposed to, so in the olden days there were slots, like there was a slot that the screw went in and then the other end of the case for a full length, uh, for a full length card, there was a, um, uh, slot that the other end would go in to support both sides of the card.
And that's what that handled was.
brad: [01:05:22] That's amazing.
Will: [01:05:23] it was to make it reach.
brad: [01:05:24] I bring it up. Is I just, I have this memory and this may, this may have been like a hundred people, I don't know. And not, not that big a deal, but there was this weird boutique community of people trying to get their hands on, like the voodoo five samples that had gotten out into the wild
Will: [01:05:38] No, they were highly desirable. Yeah.
brad: [01:05:40] and they were trying, and there was like a user effort to maintain glide and bring it forward so they could keep playing games on 3d effects.
Even after that was no longer a going concern, they really should have just, they really should have just bought an Nvidia or an ATI card, but they were just like desperately trying to keep 3d effects on life support.
Will: [01:05:54]I wonder If somebody ever made like a direct x or open GL to glide wrapper to go the other way. So you can play glide games on open GL cards or direct 3d cards.
brad: [01:06:04] If there's anything I've learned about the internet is, is that if you have to ask, I wonder if ever if ever someone did X, the answer is yes.
Will: [01:06:11] It's it's um, it was a fascinating time because
brad: [01:06:15] fun.
Will: [01:06:15] Stuff was happening constantly. Like you would, you would look at plan files for, like I said, like the people who are doing graphics programming at 3d realms and, and ID, and Epic and, and places like that valve, and you would see all sorts of amazing stories come out as a result of that.
brad: [01:06:35] Things, things were just iterating way faster back then than they do now.
Will: [01:06:38] We have a couple of friends of the pod who are like ex 3d effects folk that are probably far enough out of NDAs now that we might be able to get them onto talks off, ask around.
brad: [01:06:48] that would be fun.
Will: [01:06:49] Um, but, but that, that time, like those, those guys, they were in Texas and they were, they built this thing. And I don't think at the beginning they didn't realize what they had.
And then I think they had more, they thought they had more than they did real quickly.
brad: [01:07:05] everything's bigger in Texas, right?
Will: [01:07:07] Something like that. Um, but yeah, it was, it was a, it was a fun time to be an enthusiast. Um, I guess that's it for today.
brad: A walk down memory lane.
Will: Yeah. Uh, I guess we've reached the point in the show where we talk about, uh, the Patrion. Uh, if you, if you would like to support.
Uh, Brad and Will made a tech pod. You can, by going to patrion.com/tech pod, for as little as two bucks a month, you will get access to the fabulous tech pod discord, where you find out things like, uh, how long. Uh, time is for a CPU versus human scale. And let's see what else we've talked about. Uh, we got to shout out, uh, our, our, uh, uh, I idly asked a question in chat the other day, and then one of our fabulous users made an app, a web app that does that thing.
brad: [01:08:03] literally coded a solution for us. Uh, To fulfill a very specific need that we had. I can't, I cannot express my gratitude enough. You know who you are.
Will: [01:08:14] I think, I think we can out them. Right. First heart, um, made beep thirty.com. If you do a podcast and you need to sync it, we've tried it this week. I'll let you know how it goes next week. I guess.
brad: [01:08:24] I don't know if you want to bring that much load to that. And I don't know how much capacity he's got to serve that page.
Will: [01:08:29] Look, I comfortable
brad: [01:08:30] You really want to, you want to out this thing
Will: [01:08:32] look this is a burgeoning business. Um, I want to support the support, all of our, uh, all of our fabulous patrons. Um, and yeah, like, like there's just interesting conversations basically every day in there it's a fabulous place to hang out. Uh, I have enjoyed having human contact, um, during these dark times. And, uh, yeah, it's, uh, it's, uh, it's a lovely spot. Uh, and as always, we think all of our patrons are for more. If you, if you spend five bucks a month, you get access to the monthly exclu patron exclusive episode, where we usually talk about projects. I moved everything over to home assistant last week. Cause smart things goes away today.
brad: [01:09:15] I am following, let's say, uh, four to six weeks behind in your footsteps.
Will: [01:09:22] How's it going
brad: [01:09:22] I also, I also just, let's say I spent a lot of last weekend messing with home assistant. On the, on my Nas. And, and then we'll no, not on the nas it’s not.
Will: [01:09:32] Oh no. The NAS is a pain in the ass. I just bought a raspberry PI cause it was way easier
brad: [01:09:34] not, not in free BSD. So yes, I too just bought a raspberry PI four, to do things like that on, because it's not going to happen in BSD, but I got the PI four here.
What do you, what do you, well, we should save this, this episode's long already, but
Will: [01:09:50] we'll talk about
brad: [01:09:51] debating whether to put Raspbian or ubuntuon that thing
Will: [01:09:54] I did Raspbian Raspbian is easier to maintain.
brad: [01:09:57] Yeah, that's fair.
Will: [01:09:58] Um, uh, as always thank you to all of our patrons, uh, but especially our executive producer level patrons, take them chapel, Andrew Cotton and David Allen, uh, without all of your all support, we wouldn't still be making the podcast.
So thank you all so much. We would just be having these conversations in discord and it would be just for us and, you know, we completely wasted. So, um, we, we appreciate you all a bunch and, uh, and thanks to Julian, uh, for doing transcripts every week for the show
brad: [01:10:26] Yes. Yeah, that stuffs awesome
Will: [01:10:27] Uh, if you have friends who are deaf or hard of hearing, and you think that they would be interested in the show, you can share it with them.
The transcripts are available at tech pod dot content dot town. There's a big transcripts button there. And we have them for like the last eight, five or six episodes, seven episodes. Now I think, so that is awesome. And if you like them, please let us know. So we, so we keep doing them because like, if nobody uses them, we're probably gonna not, you know, but if people, if anybody's finding useful, we would like to know,
brad: [01:10:54] Yeah, for sure.
Will: [01:10:57] And I guess that's it for us. We'll
brad: [01:10:59] Fire up some GL Quake
Will: [01:11:01] I kind of want to be here. Can we, can we play some GL quake?
brad: [01:11:04] Is the master server still up? I don't know.
Will: [01:11:06] I don't think, I think we can just look, I know your IP, you know, my IP, we can put our IPS together. And the question is who gets the ping advantage? I don't know. I I'm happy to host. That's all I'm saying.
brad: [01:11:16] We'll have to shoot for it.
Will: [01:11:17] Okay. Okay. Uh, see you all next week.
Thanks for listening everybody. Bye. Have a good one.