- This thing is so unbelievably cool. Okay, sorry. Let's host the video. Okay, quick: name the companies that pioneered 3D graphics. If you said Silicon
Graphics, Pixar and 3dfx, you're close, but no. Those guys merely adopted 3D. Evans & Sutherland was born into it, created by... (coughs) Excuse me. Sorry, I don't
do a good Bane voice. Evans & Sutherland the company was created by and named for the guys who literally taught the classes that shaped the careers of the people who went on to create
those other companies. "But that's ancient history," I might hear you say. "Show me something interesting." Here's something interesting. A dual-layered graphics card with two different connectors on it, that's old enough to still use DVI and yet appears to have four GPUs on it, predating Crossfire and Nvidia SLI. Are you not interested? (upbeat music) What the heck are we lookin' at here? What it appears to be is
two completely separate, independent graphics cards sandwiched, and then if you look in between, bridged together with
two different connectors. So this one uses AGP, or
Accelerated Graphics Port, and this one uses PCI. No, not PCI Express. Old school PCI. And what it all comes together to make is Evans & Sutherland's
last graphics card, known as the SimFUSION OpenSim 6500q, and what a beast it is. Each of these cards has two
Radeon 9800 XT graphics cores for a total of four GPUs. Guys, other than some
weird Alienware thing, this is before quad-SLI
or quad-Crossfire existed, so how did they do it? As it turns out, after
we posted on Twitter about finding this card on eBay, one of the former Evans
& Sutherland engineers who worked on this project took notice. Layne Christensen, who, by the way, is still working for the
simulation arm of the company, which is now owned by Collins Aerospace, volunteered to offer us some background, so we want to thank him very deeply for taking the time out
of his busy schedule to demystify this for us. - During my conversation with Layne, he explained that he was part of a team of about 10 hardware engineers
and 5 software engineers who worked together to build the cards, which came in both a dual-GPU
and quad-GPU configuration. The quad-GPU config we have here requires two separate cards, and they communicate together
using AGP 4x and PCI slots, with a mid-board connection that links them together here. Layne wasn't sure exactly
what bridge chip they used, but I see an FPGA on the
front of the card here that seems to fit the bill. As for why not six or eight GPUs, this was actually the
practical upper limit that they could achieve
using AGP and PCI alone. Remember, while AGP and PCI
Express are dedicated links, or at least can be, the old PCI standard shared its bandwidth among all devices on the bus by design. - According to Layne, ATI, which is now Radeon Technologies Group, had baked-in support for connecting multiple GPUs together like this in what was clearly a
precursor to Crossfire, even though they didn't offer it to end users as a solution like just, you know, plugging in a couple of cards. Kinda makes sense, because no
motherboards that I'm aware of existed with multiple AGP slots. But hey, why not offer it
since you're working on it to a third-party partner who's trying to build a custom solution? I mean, they don't really
do that kinda stuff anymore, but that was something
that ATI fully endorsed and even offered support for at the time. By the way, speaking of
the past, get subscribed. We're gonna be building
a PC from 10 years ago and comparing it to brand-new. You won't wanna miss it. Now, as for Evans & Sutherland, you guys might think, "Well, gee, if they're gonna go and build this thing, why don't they just sell it? Why couldn't I buy one of these? Why have I never seen one of these?" The truth is these cards weren't available to just anyone on the open market. SimFUSION was their simulator
platform at the time, and as best as I can tell, a prospective customer would
be looking at buying a 25U or even a full-height 40U rack of five to nine node PCs
configure with these, all connected together as a cluster with a common sync for both
output and the render buffers. Now, as you can see, unfortunately we don't have
five or even two of these, but Anthony did manage to
put together this test bench based on the same hardware configuration, using specifications from
Layne and the user manual. Wait, is this thing using
all solid-state caps? - [Anthony] That would make sense. - Wow. That was like, very
high-end shiz back then! - Back then, all-solid-state
caps was a little bit, mm, unusual. This, actually, high-end
Intel server board used caps that were part
of the capacitor plague. I actually had to recap the board before it would work. - We're just gonna have to pretend that we're using proper
cooler mounts here. Don't overthink it. Let's just throw the card in here and... Oh, right, a two-slot card. (grunts) Had to kinda push on that pretty hard. And take a nice cold
drink to take our mind off at LTTstore.com. (Linus laughs) Power on our switch.
Let's do moment of truth. (fans whirring roughly)
Whoa! Yep, well, there's you're... There's your Thermaltake bearings... Yeah, 17 years later. (laughs) To be clear, I'm not actually
blaming Thermaltake for that. That's just what happens to bearings over the course of 17 years. - [Anthony] Especially for
something that would've been on 24/7 for a long time. - [Linus] Num Lock's working.
No display output yet though. - There's an interesting idiosyncrasy with this card, actually. Due to the way they had to, I guess, do like double-wide and even
triple-wide resolutions, they disabled certain other resolutions in the BIOS of the card, so it won't actually display text mode or even, I think, VGA mode. It has to be 800x600 or above. So you're not gonna get
anything over there right now. - Until I do! Is that a Windows XP desktop
that I'm lookin' at, baby? - That's what it ran on. (clears throat) I put on my mask again. - You guys might be wondering, "Where on Earth did you find a driver for a card like this? Some Russian FTP?" Good guess, but actually, no. We scoured the internet and were not able to find any working driver. Thankfully, Layne actually had one lying around on an old
drive, and presumably got his employer's permission
to fire it over to us so we could do something
that a SimFUSION 6500 probably never did: run games. What should we try first? I mean, 3DMark 2001 SE is what
I'm sort of leaning towards. Or maybe some good old-fashioned AquaMark? Is that still around? - [Anthony] I don't know.
I didn't download it. - I don't know if I
wanna plug an XP machine into our network. - [Anthony] It was
plugged into the network for a while there, but
it's XP Service Pack 4, like the Unofficial Service
Pack 4, so it's fully patched. - All right. Man, this is not a fast computer, sir. - [Anthony] No, it is not. - What is this? Single-core... Single core with Hyper-Threading, oh yeah! Four threads, baby! Mm, yeah! Four threads! High school me would've thought this was amazingly sexy to look at. - [Anthony] Oh yeah. Just even seeing multiple
graphs in Task Manager was like the most trippy thing. - Okay, here we go. (Linus laughs) Oh no! Wow! Okay. In fairness to the card, though... I mean, it's running. Wow. Even the loading screens are a little broken. - [Anthony] It's worth pointing out that all of Evans & Sutherland's
simulation stuff was OpenGL-based. They didn't really qualify
the card at all for Direct3D. The GPUs themselves are
Direct3D 9 compatible, but four of them? It might be a bit hit-or-miss. - I remember this nature demo
looking so impressive to me. I was like, "Wow! There's so many leaves on the trees." Man, I'm trying to find a 9800 XT 3DMark 2001 run to compare against, and these old websites have not aged well. Hey, there we go! Okay, so at 1024x768, a single 9800 XT managed 16,800 points. Let's try AquaMark. Oh man. This too. It just brings back so many memories. This is like pre-me-making-videos. This is like me literally
sitting in my basement or my girlfriend's room, you know, tuning my overclock one megahertz at a time or whatever and staring at this,
checking for artifacts. Lemme tell ya, if I ever
saw it look like this, I'd be like, "Oh no! Off, everything off!" and
turn that overclock down. That's not the issue right now though. And that's the massive overdraw, was the big explosion at the end. 9 FPS. 8.6 FPS. 8.5! AquaMark 3. 44,262 with
an average FPS of 44.26. Now, this isn't coming up right now, and my Num Lock's not even working. I think the system may have crashed. But it's pretty fair to say that we did not average 42 FPS, so from a Direct3D gaming perspective, buying four 9800 XTs is actually worse than buying one 9800 XT. Let's try something OpenGL. Here's something I don't really get. Why is it that the two 9800 XTs in the middle of the sandwich are the ones that don't have fans? - [Anthony] It's possible
that it was designed for case airflow. - If that's the case, then we should probably put
a fan on this poor thing 'cause it might be cooking in there. I wonder if that's why we're
getting so many artifacts. We need a bigger fan. Well, not bigger, but definitely more powerful. Aw! Stop! - [Anthony] Should we have
grills on our blowies? - [Linus] Absolutely, we should. These are a finger-removal
hazard waiting to happen. Ow! - You okay?
- I pinched my tummy. (Linus and Anthony laugh) - [Anthony] I've done that so many times. - [Linus] Ow! Whoa! Settle down there, buddy!
(fan screaming) Oh yeah. We got airflow now, boys! All right, benchmark time. Dammit! We're gonna leave it on anyway though, because having the card cooled properly is gonna be a factor for our last test: "Unreal Tournament" '99
Game of the Year Edition. Now this is an actual OpenGL game, so this is what the card was... Well, I'm not gonna say tuned for, it wasn't made to run games, but their simulators did use OpenGL. Oh my god, I can't remember
how to move in this game. No! No, I suck! Aw, no! I don't understand this FPS counter. Okay, it's frame times, not FPS. So we're anywhere between three and eight milliseconds per frame. Probably we're running around 150, 160 FPS, I would say. - Sounds about right. We're at 1280x1024. - Oh, come on! Just die! Okay, if I can't manage to get a kill with the rocket launcher,
I don't deserve to... (fan whining) All right, I can't deal with this anymore. Ah, my headache. We couldn't find a professional
review discussing this, but here we go, here's a Rage3D.com thread where someone complains that in "UT", whenever there's dynamic lighting their 9800 Pro drops from
100 FPS average to 40 FPS. So that would seem to indicate that if we're over 100 FPS here, then at least we're doing
better than a single 9800 Pro. Okay. So that's not four times the performance, but it's better than a kick in the teeth. Of course, as I've said before, gaming was really not the point of it, and the enterprise-grade functionality, like being able to sync
together many of these running in many systems
across an entire rack, was really what Evans &
Sutherland's customers were after. Here's a fun thing,
though. Check this out. I Alt-Tabbed out of the game and we got this really weird sort of interleavy pattern on the desktop. - The card was able to be set
up in multiple different ways as far as how it renders
the screen itself, so as Linus Alt-Tabbed here, he ended up with these strips. It seems to indicate
that we were looking at a kind of line-based
rendering mechanism right now. Otherwise, it could be
split up into quadrants, or alternating the swap chain as each card just kinda renders a frame and just kinda keeps going like that. - What's especially cool about that is that those rendering techniques are actually the same
ones that are still used for multi-GPU setups today, which I guess leads us
to why the 6000 series was both the first and last of its kind. Layne says that after the
development of this card, PCI Express enabled just
off-the-shelf solutions for single- or even multi-GPU setups that made this kind of
engineering totally unnecessary. So for how long this thing was relevant and how expensive it must have been, I count myself lucky to have one, Thanks for watching, guys. If you're looking for something else kind of historical to watch, maybe go check out my video on Intel's previous dedicated GPU project, Larrabee. We unfortunately weren't
able to get that one working, but we did get a chance to, you know, at least try to power it on, and talk to one of the engineers who worked on that project as well, so a very informative little video.
Should have mentioned that the Radeon 9800Pro / XT was Raja's baby when he was at ATi for the first time and it fucking crushed Nvidia for two GPU generations and is probably responsible for mass extinction of other 3D gpu designs around that time (Trident SiS, XGI, S3, Matrox, etc).
This is one of the reason why old fucks like myself give Raja benefit of the doubt.
Already thought that the r9 295x2 was insane but 4 GPU's in one card? Wow.
For shits and giggles we need dual big navi with the 505mm² chip, both linked with Infinity Fabric. If nvidia will have aftermarket cards with 3.5 slot coolers, AMD can make a 3 slot hybrid with 240mm AIO, 4x8 pins with TDP ~600-700W. Something like Titan on steroids for the prosumer market or as a halo product.
I had a 9700 Pro, what amazed me most when tearing it down after the end of its useful life is what a fatass piece of silicon it was.
still have my 3870x2 LOL
I thought it would be another video about the Voodoo 5 6000, but than he pulled out this thing and I was actually surprised.
Saw that title and thumb and thought it would be a video on the legendary 3dfx Voodoo 5 6000... oh that'd be a great revisit.
Anyway, very nice surprise. I've never heard of this manufacturer and it was interesting to learn that the 9800 XT already supported multi GPU at that time, amazing stuff!