Chip Hall of Fame: Nvidia NV20 | Hacker News Reader

Chip Hall of Fame: Nvidia NV20

spectrum.ieee.org

63 pointsrbanffy38 comments

Nvidia is one heck of a pivot. I remember buying the most powerful card I could find in 2006 or so to simulate my windmill blades before cutting them, the voxel array was so large that doing it all in software was next to impossible but with the Nvidia card it was so fast that it allowed real time simulation with voxels of 1 mm^3. Those runs shaved months of our time and saved half a forest of trees in runs that we would have to abort that would end up with unusable scrap.

Today all the focus is on machine learning and mining crypto, but there are some excellent use cases for these cards that have nothing to do with either.

Another one, much closer to playing games is 'serious gaming', simulation of real world events to train first responders with access to scenarios that would be way too expensive, disrupting or complex to set up in real life.

Nice, encouraging to hear that massively parallelized operations are being used for more than trendy buzzwords (ML, crypto) and gaming. Engineering simulations are less sexy than "AI" but can hugely increase efficiency.

You may also like that Switzerland is running its national weather prediction services on GPU since a couple of years (the first and AFAIK still only national weather service to do so).

Serious simulation is already a thing in racing, aviation and military.

Even they were pretty happy to get rid of their SGI stuff and replace it with cheap (and plentiful) hardware.

This stuff has enabled entire industries that are not 'sexy' but very useful.

That was more a consequence of Linux's adoption.

In a parallel universe SGI would still exist, selling Ultrix or eventually NT workstations.

Not really, most of that stuff runs under Windows. Linux machines are few and far between unless it is us techies that are using them, then suddenly it seems to be everywhere.

As for SGI, they also went the Windows way and that is one of the contributing factors to their demise.

I really miss my Irix machine but it wouldn't be able to keep up with todays demands, but even the slowest SGI boxes (Indy, 100 Mhz R4000) was impressively fast for the time when it was made.

Wasn't Rich Belluzzo in charge of HP-PA when they decided to ditch it along with HP-UX because Windows was the future, then moved to SGI, where he licensed their crown jewels to Nvidia, decided to make some of the worst Windows NT machines ever seen, and then moved on to... Microsoft?

I think it's one of the first instances of weaponized outplacement I've ever seen.

> I really miss my Irix machine but it wouldn't be able to keep up with todays demands

You can always run X on them and do the heavy lifting from a cheap Xeon E3 box... I can't wait to bring over my IBM RS/6000 ;-)

I'd like to think it was SGI mismanagement, getting off the dedicated graphics/supercomputer boat too late.

In my parallel Universe, SGI would occupy the spot of Apple with high-end beautifully designed consumer laptops and electronics with a slick IRIX as its OSX. That would have been something I would have loved to hate.

s/Ultrix/IRIX/

It's Unix, just way cooler than the other kids ;-)

> eventually NT workstations

Heaven forbid.

Not a single word about Renderman?!!

Before GeForce 3 was a thing, those of us with access to NeXT machines already had a blick what would mean to write shaders.

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

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

"The RenderMan companion : a programmer's guide to realistic computer graphics.", 1990

I am also missing 3DFx's work on shading languages before they got acquired by NVidia.

"blick"?

"Glance" in German.

Swedish too.

I'm the editor of the Chip Hall of Fame: if you follow the link in the citation where Pixar/Toy Story is mentioned, it takes you to a whole Spectrum feature about the development of Renderman :) But this article was intended to be tightly focused on the creation of the NV20, and how it enabled real time shaders, so I didn't have room to do more than mention Pixar in passing!

Faire enough, thanks for clarifying it.

The linked story is actually quite interesting.

Nvidia is my favorite example of "the next big thing that started out as a toy" ala A16Z's Chris Dixon (http://cdixon.org/2010/01/03/the-next-big-thing-will-start-o...).

Literally started off as a device for playing games, and significantly expanded out use cases since then.

Let's not forget that before fully-programmable shaders in NV20, NVidia had register combiners [0] in GeForce 256 (NV10), allowing more flexible texture application & blending compared to the older (sequential) pipeline.

[0] http://www.nvidia.in/object/registercombiners.html

It's disappointing that the write up doesn't include something on Molnar's work on Pixel Flow at UNC:

https://dl.acm.org/citation.cfm?id=133994.134067

I'm the Chip Hall of Fame's editor: the citations are intended to be brief and really tightly focused introductions to the development and legacy of a particular integrated circuit, so I'm afraid we just didn't have the scope to include Molnar's other work!

Fair enough.

The thing is that Pixel Flow was relevant for NV20. It's pretty much the commercialisation of what was developed.

But kudos to you, it's a really good write up generally.

Thanks!

I remember the GeForce3 as the point were GPUs got much more expensive again.

With the GeForce2 there were some budget version, I had the MX.

Not so with the 3, there was only one model and nobody of my friends had enough money for it.

Ha I just noticed the significance of this timing. As a teenager I could just about afford a GF2 MX when it came out, but never ended up being able to afford any of the subsequent couple of generations (the GF4 MX were barely a step up). I lost interest in PC gaming shortly after and never really returned, so even when I could afford a decent graphics card later I never got one.

Luckily this was also the time when people played the same games for years.

I played HL&Mods until 2004 or something. So I didn't really "need" a better GPU.

Voodoo 2 was ~$500 in 1998. GF3 launched at $500, but dropped to $400 after barely a month. GeForce3 Ti 200 was $150. If you wanted decent cheap you bought $200 8500LE/4200, if you were clueless you bought $100-180 gf2MX/gf4MX.

GeForce 2 MX (with DDR2 video ram), was capable of running DooM 3 using the same patches for Voodo 2/3 GPUs (resizing the textures to become very blurry)

It was a great cheap card for his time, and endured many years.

Always interesting to think how Nvidia struggled for its first 2 generations of graphic chips nv1 and nv2 (partnered with SEGA) but made a first big dent in the market with nv3.

I can remember growing up my Dad purchased a new Pentium 133, this must have been around 1995/1996. It was absolutely cutting edge machine at the time and had a "S3 trio" PCI card in it. I have vague memories of playing Quake on it and being blown away at the graphics. A few years later 3dFX Glide cards were cutting edge for gaming.

I started university around that time stopped following gaming/computer hardware scene but really felt to me like Nvidia came from absolutely nowhere.

The big "wow, things have changed" moment for me was back when I saw Nividia put up a giant booth/pavilion at CES, not by the computer or gaming folks, but in the automative hall (which had previously been something of a sideshow to the main CES halls).

I spent all my money on a nv3 chip. It was so fast that it made games easy. I was used to precompute mentally every move because of low fps and lag. It probably also changed how physics were simulated in old games (lower time resolution probably introduced wilder results).

nvidia felt like an odd company, it was small and unknown, unlike Ati or Matrox. But their chip were cheap and competitive with champion 3dfx.

Still impressive how they stretched markets from toys to the highest sophistication.

3dfx with vga passthrough were the real life changers thanks to advanced texture rendering. Still poly were not anti-aliased. It was a real paradigm change and captured a majority (>80%) of the market. If you remember the game Turok with and without Voodoo you lived that days :)

That was a huge improvement, unmatched by anything that came after. Maybe the move to multi-core cpus comes close.

TBH, at the time I was expecting PowerVR to make a big splash. I didn't see 3dfx to be as big - it was odd with add-on cards etc. I was also waiting and waiting for SGI to finally put something out there in PC space apart from Cobalt. It never happened. That company had it all and lost it all. Nvidia today is what SGI should've been. I suspect most of people from SGI work there now anyways (and/or ATI/AMD).

NVidia was founded by ex-SUN ex-AMD folks. 3dfx was founded by ex-SGI folks. SUN used to compete with SGI in the high-end graphics workstation space so in a sense NVidia consolidated them both. AMD on the other hand consolidated independent pioneers like ATI, Tseng Labs etc.

Sun was never a competition in graphics workstation much, but workstations yes. Also, don't forget ArtX. SGI people everywhere! Everywhere, except SGI. I'm still bitter about it, I loved that company. Not their prices though. So far ahead of their time, kind of like Amiga where budget was not a question.

Considering that 3dfx was built by ex SGI employees IIRC that's two failed ventures.

It's true that Nvidia is the SGI of the era.

Fair. But keep in mind NVIDIA's a scummy company.

https://www.youtube.com/watch?v=H0L3OTZ13Os

Reply on news.ycombinator.com