The AMAZING 8-Way SLI VooDoo Brick

Save 15% with code LinusTechTips at

Try Zoho One free for 30 days with no credit card required here:

Multi-GPU setups like SLI and Crossfire once ruled the PC gaming scene, but micro stuttering and development complexity have made them obsolete. Why, then, are we boldly predicting that the future of gaming is multi-GPU?

Discuss on the forum:

Ross Tregemba's YouTube channel:



Intro: Laszlo – Supernova
Video Link:
iTunes Download Link:
Artist Link:

Outro: Approaching Nirvana – Sugar High
Video Link:
Listen on Spotify:
Artist Link:

Intro animation by MBarek Abdelwassaa
Monitor And Keyboard by vadimmihalkevich / CC BY 4.0  
Mechanical RGB Keyboard by BigBrotherECE / CC BY 4.0
Mouse Gamer free Model By Oscar Creativo / CC BY 4.0

0:00 Intro
1:21 Quantum3D made some crazy GPUs
2:31 Setting up the Mercury Brick
3:46 Powering it up and gaming with no jaggies whatsoever
5:42 Turning off anti-aliasing is easy…
6:30 It's a night and day difference
7:08 Simulation demos
8:04 Quantum3D still exists! And they were bigger than you think
11:02 3dfx's demise didn't stop multi-GPU development
12:57 Consumer multi-GPU's demise – Micro-stuttering and support
14:45 Multi-GPU is coming back – And Apple is pioneering it
16:43 Conclusion – It took a long time to get here




Sometimes I think all great computer technology has been invented in the 1990s and just needed to be perfected in later decades.

Just a few computer innovations that first appeared in the 1990s and still exist in perfected versions that all build upon 1990s standards and technology:

– Wifi (802.11 in 1997, 802.11a and 802.11b in 1999)
– Digital Cellular Networks (2G in 1991, 3G in 1998)
– World Wide Web (implemented in 1990, published in 1991)
– Bluetooth (first introduced in 1998, not ratified until 2002)
– USB (USB1.0 in 1996, USB1.1 in 1998, USB2.0 in 1999)
– DVB (DVB-S and DVB-C in 1994, DVB-T in 1997)
– ISDB (1999)
– AES (1998)
– PGP (1991)
– MD5 (1991)
– SHA-1 (1995)
– OpenGL (1992)
– Direct3D (1996)
– JPEG (1991)
– PNG (1996)
– MPEG-1 Audio Layer II aka MP2 (1991)
– MPEG-2 Audio Layer III aka MP3 (1995)
– MPEG-2 aka H.262 (1994)
– MPEG-4 (1998, often implemented in the early 2000s under brand names like “DivX, “XviD”, “3vix”, etc)
– AVC aka H.264 (2003, but first implementations were already around in the late 1990s alongside AAC)
– AC-3 aka “Dolby Digital” (1991, “Star Trek VI” was the first movie with AC-3)
– DTS (1993, “Jurassic Park” was the first movie with DTS)
– AAC (1997)
– IPv6 (1995)
– UTF-8 (1992)
– UTF-16 (published in 1996, first used in the late 1990s, not ratified until 2000)
– Linux (first Linux kernel released by Linus Torvalds in 1991)
– P2P file sharing [Napster] (1999)

Other stuff that first appeared in the 1990s: Video game console emulators, PC watercooling, DVDs, MMORPGs not based on text, x86 instruction set extensions and the first CPU to break the 1GHz barrier [not commercially available until early 2000].


    “You don’t need to reinvent the wheel”

    The Guy That Knows

    Makes me think of my grandfathers washing machine, a million years old but still running when I’ve gone through 2 in 5 years but sold to me as new and better then the last one. Planned obsolescence is a hell of a drug


    A lot of tech is theorized in early periods, but was not feasible due to hardware limitations (e.g VR). We just live in a period now where many of those ideas can come into fruition, the latest trend being AI.

    Dead Reckon

    @The Guy That Knows Everything used to be built better because it was built simpler, you can’t have complex modern technology that the masses want without fragility, just the sad truth man.

    F#СК МЕ. ТАР 0N MY Р1С


Lakshyajeet Jalal

“The owner didn’t trust me”
With Limus dropping expensive stuff everyday no one would trust him to handel their prized collection

    F#СК МЕ. ТАР 0N MY Р1С


    Hr. L. Bolle

    did Destin let him handle the memory block and circuit board from the Saturn LVDC ?

    The Official Channel

    What about his wife asking to hold the baby 😂

    Dr. Marmal8

    not the homie Limus, i don’t think he can handel this

    Joshua Lee

    Especially when the product is nigh impossible to find.

Jackson Fernandes

Its crazy that in half a human life, we got from non existent 3d render to realtime raytraced rendering. Mind blowing

    F#СК МЕ. ТАР 0N MY Р1С


    Kasey M

    Don’t worry, there are plenty of dumb/greedy aholes to pull us all back down. I am such a pessimist.


    Yeah when I was rendering 4 days in lightwave for a 1 minute scene with the BttF Delorean is now done in realtime @100 fps, and people complain still that it went to 80fps for a few seconds. 🙂

    Sipri M

    All thanks to socialism and government regulation that’s not private property rights-oriented, a.k.a. for the greater good. Democratic socialism for the win!


    @Sipri M ????? wtf do you mean, literally all of the big 3 GPU manufacturers are in a heavily capitalistic country

Arcade Guy Ceej

Man, 3DFX in arcade games was a wild era.


    Hydro thunder!!!


    NFL Blitz always had crowds around it!

Yasunaka Ikumi

I think the problem we have with the current GPU situation is that AMD and NVidia them selves are controlling the Vram of every AIB so compare to the 90s we’re pretty much doomed in terms of if we want more Vram we cant get the same GPU with higher Vram for smaller price difference 😒😒

Garrett Kajmowicz

Ah! The days when video card manufacturers could just buy chips from suppliers and expect to be able to independently make a product with them, even if it was substantially better than the reference implementation.


I think we’ll see tiled GPUs soon ish because it both makes making various skus and allocating them way easier (e.g. if there’s too many high end gpus sitting, they could allocate the new chiplets to lower end GPUs) and smaller chips have better yields, so especially going into sub 1nm I can see that becoming the best option for high performance GPUs.
Though they would probably behave more like GPCs (Graphics Processing Cluster, a group of rop, tmu and sm, which have a bunch of cuda cores) than individual GPUs.


Multi-GPU is extremely powerful for computational physics. Workloads like computational fluid dynamics need more VRAM than one GPU can offer, and can pool the memory of multiple cards. It’s super difficult to implement and maintain though – developers have to manually implement and optimize which data to transfer between GPUs at which point in time. Making it more difficult, InfinityFabric on AMD is broken (driver segfaults) and NVLink on Nvidia is proprietary to CUDA and not available to OpenCL. So your only option is to do communication over PCIe. Luckily, PCIe 4.0 is plenty fast and 5.0 is coming with double the bandwidth. This also enables cross-vendor multi-GPU, “SLI”-ing Radeon+Arc+GeForce.

14:49 Multi-die GPUs are super interesting. Apple succeeded it with the M1 Ultra. AMD failed with the MI250(X), turning it into a dual-GPU-in-a-socket, but tries again with the MI300. Intel with PVC has performance issues in single-GPU mode, so also offer to configure it as dual-GPU in software. Nvidia’s A100 is essentially a multi-chip GPU but on a monolythic die, and performance is outstanding.


    Cool comment! Thanks for sharing 🙂


    I like the idea of a geforce card but then representing arc on the side. Almost a nod to the multiple cards of the past.


    It’s simple why Nvidia or AMD haven’t produced a transparent GPU, Their prime focus and cost! Apple focuses on producing home work stations with power and design in mind, Nvidia and AMD work on chips to power gaming rigs and business workstations (as well as home PC workstations), Their pies are more and it makes no monitory sense to drift huge resources off track, When they can produce more powerful chips with new technologies and software combinations.

    Eric Hallahan

    Man, OpenCL really needs some love, doesn’t it?

Steve 30x

I had great hopes for Lucid technologies until Nvidia bought them off then killed the technology. The idea of mixing graphic card technologies sounded like a great idea once they got it right. I have an RX5700Xt and a gtx 1080 backup in case my 5700 died. I would love to see how they would work together with lucid if they kept perfecting it.

I did do SLI a few times though. I did have two identical BFG GPUs before which worked perfectly individually but in dli they were attracting all over the place. I installed a different bios for a different brand and same GPU with same clock speeds on both gpus and SLI worked perfectly with the cards. When I contacted ed BFG they were in total denial that it was a bios problem and refused to look into it. I wasn’t alone with lots of people with the same problem. Maybe that’s why BFG went out of business 😂

Stephen M

The difference with 3DFX SLI and new “SLI” is that 3DFX SLI didnt have the overhead of predicting frame areas/timings. The driver worked in a set way due to the render methods used at the time and all that had to be done was the required pre-determined render areas (scan lines) sent to the display output. This is also why the Quantum cards can have AA and non-AA output active at the same time but not pool their work to a single high FPS image.

Sam The Multimedia Man

The nice thing about SLI and Crossfire from back in the day, you could buy 2 cheaper Video cards on sale and hook them together and it could equal the performance of a high end GPU at a fraction of the cost.


Wow. They even got someone with a haircut from 25 years ago to put it together. The immersion is real at LTT.


This retro stuff is great, please make more content like this. Would be fun to see how sound developed too in pc gaming and content creation


    There’s better channels aimed at that audience out there than LTT.


    @TUUK2006 I’m aware, but did you take into consideration Linus’ face and how excited he was doing this stuff?

Ross Tregemba

It was extremely fun to bring my favorite hardware and showcase it with you guys! Even more fun to drop a load of boxes on Linus though 😀


    My rawr XD retro tech brother! Great job in the vid 😀

    Ross Tregemba

    @Dredile <3


    You gave me the urge to listen to MCR after this vid

    Mickel Ntek

    that is how it gets started. first, you drop some boxes on Linus next you start to drop all kinds of things. I love old crazy hardware stuff.


    Not allowing Linus to install it was probably for the best 😉

Always Bolt the Bird

When I was in the Air Force my job was basically working on/upgrading/fixing flight simulators and have worked with Quantum3D before. This video brought me back so thanks for that


In the late 90s, I worked on an SGI Indigo that had a voodoo card with 4GPUs. It was the most impressive experience at the time. The lab upstairs set up simulations using these machines. Incredible technology for the time! It is what really pushed me to get into building gaming systems.

    Ross Tregemba

    Love that story!!!

    Пича Лакош

    Yeah, right


This video was very technical. Whoever wrote it did and amazing job at making it digestible and easy to understand. I enjoy these deep dive videos!

    epoch 888

    Anthony is an excellent writer.


Please keep this type of content coming, what an amazing presentation of bleeding edge tech from 2 decades ago!

Robert Gout

This has got to be one of my favorite videos in the last year. Amazing showcase and explanation guys!

Martin Strazynski

Phenomenal script and delivery. Props to Linus and the writing department for all bringing their A games and the history lesson. Thank you as well to the editing teams for the visual learners!

Leave a Reply

Verified by MonsterInsights