Rx580 is on there, but not the R9 290. I’m not sure where the Rx500 series actually pushed technology forward. They always seemed like the AMD budget line. And if 580 is important, why not the 590 or the 570?
Few of the “pre-GPU” graphics accelerators that seem to have mattered are here. The ViRGE. The Mach32 and Mach64. The Trident cards, like the TGUI9440. Yet the Voodoo often isn’t considered a GPU and is on the list.
The article blew a huge opportunity to showcase the great diversity of “Pioneering Era” 3D accelerators (they weren’t called GPUs until later). But instead they just pretended it was always NVIDIA vs ATI, and threw in a few Voodoos.
Loads of games from the era roundtripped their textures through lossy S3/DXT compression and then stored them as uncompressed RGB or RGBA.
I know this because I wrote a Unreal Engine texture repacking tool with a "DXT detection" feature so that I wouldn't be responsible for losing DXT compression on a texture which had already paid the price, only to find that this situation was already hyperabundant in the ecosystem.
Many Unreal Engine games of the day could have their size robotically halved just by re-enabling DXT compression in any case where this would cause zero pixel difference. This was at a time before Steam, when game downloads routinely took a day, so I was very excited about this discovery. Unfortunately, the first few developers I emailed all reacted with hostility to an unsolicited tip from what I'm sure they saw as a hacker, so I lost interest in pushing and it went nowhere. Ah well.
I remember there was a kernel module for the Matrox/MPlayer combination. You get a new device that MPlayer could use. You did get `-vo mga` for the console and `-vo xmga` for X11; you couldn't tell the difference, and both produced high-quality hardware YUV output.
Recency bias probably, Iirc I think the 3000 and 4000 series did make significant improvements on RTX performance so compared to the 2000 series it's far more useful today.
4000 certainly did, the "shader execution reordering" gave an meaningful uplift to tasks that "underutilized warp units due to scattered useful pixels".
The G200 mattered to some degree for a long time, because most x86 servers up until a few years ago would ship a G200 implementation or at least something pretending to be a G200 card as part of their BMC for network KVM.
Probably started out as a real G200 chip which might’ve been the cheapest and easiest to integrate in the 2000s? Or it had the needed I/O features to support KVM (since this would’ve involved reading the framebuffer from the BMC side), or matrox was amenable to adding that.
A lot of GPUs in this list are basically just previous GPU but faster or more RAM. I kind of thought it was going to focus on interesting new architecture innovations.
I think pairing RX 5700 XT with Control as the "defining game" is an interesting choice, considering the facts 1. AMD cards were incapable of RT at the time and 2. Control was basically the first game with a good, comprehensive RT implementation that had a massive positive impact on the graphics.
I remember the main noticeable difference being ray traced reflections. However that was mostly on immovable objects in extremely simple scenes (office building). Old techniques could've gotten 90% there using cubemaps, screen space reflections, and/or rasterized overlays for dynamic objects like player characters. Or maybe just completely rasterize them, since the scenes are so simple and everything is flat surfaces with right angles anyways. Might've looked better even because you don't get issues with shaders written for a rasterized world on objects that are reflected.
Games that heavily advertise raytracing typically don't use traditional techniques properly at all, making it seem like a bigger graphical jump than it really is. You're not comparing to a real baseline.
Overall that was pretty much the poorest way to advertise the new tech. It's much more impressive in situations where traditional techniques struggle (such as reflections in situations with no right angles or irregular surfaces).
The most impressive part of Control's RT (on PC at least) was that it very much applied to (most) dynamic objects - and it features a TON of dynamic destruction.
The "office building" setting meant resticted areas, sure, but it features TONS of reflections - especially transparent reflections (which are practically impossible to decently approximate with screen space techniques).
Oh, and: The Northlight Engine already did more than most other engines at the time to get "90% there" with a ton of hybrid techniques, not least being one of the pioneers regarding realtime software GI.
The other elephant in the room is the consoles, and even if they're capable of RT they also have to consider the performance capabilities versus visual payoff. As I see it the PC versions of games like Control from studios like Remedy are trailblazers, it's an early implementation (geforce 20 released in 2018, Control was 2019) as the ultra option to shakedown their implementation and start iteration early so future games will benefit, however the baseline is non-RT.
This is a wonderful-looking infographic, but I truly don't think there are 49 GPUs that mattered in the PC gaming hardware space - let alone all of computer graphics. Call it recency bias, but after the Pascal cards it feels like maybe one or two more entrants actually mattered?
Agreed, those early manufacturers/models that experimented more feels more relevant than the more incremental listings of multiple 2000 3000 and 4000 series NVidia GPU's.
its a very honorable mention in my eyes because its more appropriate of the tile of "first independent Graphics unit" than the Geforce 2. (did more than just blast already projected triangles at the screen)
not that it was an awesome product, but certainly it was flexible.
a good (albeit tiny) demo of that is that vquake has the same wobbling water distortion of the software renderer quake but rendered entirely through the gpu. Perhaps with some interpretation this could be called the "caveman discovered fire" of the pixel shading era.
Absolute nostalgia fever. About a month ago, I dug up an old desktop in the corner, took the drives out and gave away the machine. It felt like putting a racehorse to pasture: i7-4790k, 1080 Ti. It was my dream machine when I got it. Dual-boot (as we did back in the old days when Proton wasn't here) to Ubuntu, then Elementary, then Arch. By the time I gave it away it wasn't worth the power cost.
And that brought to mind my older dream machine, an 8800 GT from generations past, before which we made do with a Via Unichrome that worked sufficiently enough on the OpenChrome driver that I could edit open software (Freespace only needed a few constants changed) so it would render (though some of the image was smeared and so on I could play!).
I used my 1080 Ti for about eight years. The successor GPU is in some ways way faster (raytracing, AI features etc.), but in others really quite stagnant considering the huge stretch of time that passed between them. ~10 years for 2-3x performance in GPUs at higher nominal and real price points shows how slow silicon advances have been compared to the 90s and 2000s. The same period from 2000 to 2010 would've seen 1000x performance if not more. The difference between a 1080 Ti and a more expensive RTX 50 card is the RTX can render ideally triple the frames in synthetic benchmarks, double the frames in some rasterizing games (most games won't see gains that high), and do a few relatively tame raytracing tricks at performance which is still not really good. At the same throughput it consumes maybe half the power or a bit less. The difference between a GeForce 2 and e.g a Radeon HD 4k is several planes of existence.
I've been running the worst gaming set up I can get away with, which atm is a 3080 10gb, using random DDR3 ram, a budget WD 512gb ssd, and an i5 of the same socket as the i7-4790k that doesn't even support hyperthreading and can't do more than 4 tasks in parallel.
It's absolutely laughable at this point, but I'm unironically looking for a deal on that cpu lmao, it would be a huge upgrade.
The 8800 GT is easily the most impactful GPU in my mind. The combination of that video card with valve's Orange Box was insane value proposition at the time.
I'd put the 5700xt at #2 for being the longest lived GPU I've owned by a very wide margin. It's still in use today.
Still using my RX 5700 XT. The amdgpu driver had a major issue resuming from suspend a few months ago[0], but other than that, I'm not aware of (nor have I experienced) any stability issues. Maybe you had a bad card.
I don't like to spend much on hardware, so I bought an 5700XT a few years ago and run a "steam machine" of sorts. Never had any Linux-related problems.
We had the Riva TNT2 in our family computer, so that was fun to see that again, I think it was paired with an AMD K6-2 chip.
One day one of my friends from school wanted to optimize airflow in our computer, and re-did the cabling, but he managed to block the CPU-fan from spinning. I am not sure how, but we didn't realise it for a couple of months.
When I got my own PC, it had an AMD Barton chip, and it allowed me to play Half-Life 2.
If I can at least tell myself that our technological achievements come with efficiency gains instead of just apeing power throughput, I can rest a little better
About a decade ago, I discovered that the HD 530 iGPU included with my budget-oriented i3-6300 CPU was better-performing than the physically-impressive SLI pair of 9800GTs I had been using, at something like 1/10th the power consumption.
I had the Voodoo 1 with VGA passthrough from the 2D card. When you loaded a game you'd head a little clunk from a relay on the Voodoo taking over the VGA signal and you knew you were about to have a good time. Doesn't seem that long ago!
I don't see my first GPU on there, it was the humble GeForce4 MX440. It could run almost any game I cared about for a surprisingly long time, even if it's not a true modern card.
These days almost all my machines are on iGPUs baked into the CPU. There's way less fun for me, but they are a lot more compact at least.
I'm on a 3060 currently and the changes in the 4xxx and 5xxx just aren't appealing to me. As soon as iGPUs get 3060 performance I'll probably switch. And they aren't far off.
Worth noting this covers consumer gaming GPUs only — the cards most of us are nostalgic about, but a different lineage than what actually drives Nvidia's revenue today. That said, gaming silicon is where most of the foundational architecture innovations originated: unified shaders, async compute, hardware ray tracing all debuted on consumer cards before being repurposed for datacenter workloads. The H100 exists because of the engineering path that ran through the 8800 GTX and Volta Titan. A companion visualization of "every GPU that mattered for AI" would be much shorter and start much later.
I have fond memories of lending a Voodoo 2 from a friend when I was moving from a 486 to a K6 based system component by component. At that time I was still using my old ISA VGA card, which meant 2D performance was horrible, and I couldn't really watch videos on that thing - but thanks to the Voodoo I could play Unreal Tournament without problems.
The 9400 GT mattered to me as it was my first gpu. Had bought NFS Carbon only to find that the home pc only had a CD drive not DVD lol, so finally with that drive upgrade also came the 9400 GT and fun ensued.
This brings so many memories. I remember how badly I wanted an GeForce 6800 Sadly, I was never able to justify spending this much money on a GPU. Still holds true, even today.
I was going the other way, it wasn't obvious enough that it was going to be a horizontal scroll or how to do it. Vertical spacing felt off and the 'defining game' card at the bottom of the video card is nice information but displayed in a distracting manner.
Ah I was just trying to remember the model names last week and this website pops up like magic, weird how the internet works sometimes. The 560 Ti was a dream for teenage me and most of my friends back then, but I must say my Radeon HD 4870 game powered most of my favourite Team Fortress 2 years.
Yeah the 560 Ti was insanely popular in my group of friends. In ~2004 there was a good amount of FX 5700s, some people struggling on Geforce 4, and some on the FX 5900 Ultras. Some were updating every two years, some closer to four. When the 560 Ti came out, everyone got it.
The title of site should probably have "for gaming" at the end as it doesn't consider GPUs for compute such as the A100 or the GTX 580 3GB that AlexNet was trained on.
I see it as similar to virtual reality, it was born and grew up with gaming demands and influences, but other disciplines may be more attractive for a mature product
I don't think there's strong evidence of this being an ad. I was surprised to see the Intel Arc A770, a GPU I've never heard of, included on this list. I think it's just that Nvidia has been the dominant force in consumer-level GPUs for a while now.
> I don't think there's strong evidence of this being an ad.
There is strong evidence. Click on the link above. It was posted by a viral marketing company. They even feature the GPU story on their website: https://sheets.works/data-viz
> I was surprised to see the Intel Arc A770, a GPU I've never heard of, included on this list.
Yes, because otherwise the ad would be too obvious.
I think it's a terrible UI - requires 3 different things to see the GPUS: scrolling vertically down to see the Era buttons which then scrolls up and hides the Era buttons even if you have enough vertical screen space, clicking on the Era buttons, clicking < > buttons to see the GPUs of an Era.
I can't remember last time I've seen such a confused design.
This wasn't even the worst part for me. To scroll within it as it's horizontal it is not intuitive to use the scroll wheel so you click and drag the mouse , however as the entire surface of the GPU image seems clickable it feels like your going to pull up another webpage. It feels like a bad ad that is trying to catch you off guard.
Few of the “pre-GPU” graphics accelerators that seem to have mattered are here. The ViRGE. The Mach32 and Mach64. The Trident cards, like the TGUI9440. Yet the Voodoo often isn’t considered a GPU and is on the list.
At the same time I'd add the S3 ViRGE and the Matrox G200. Both mattered a lot at the time, but not long term.
https://en.wikipedia.org/wiki/S3_Texture_Compression
I know this because I wrote a Unreal Engine texture repacking tool with a "DXT detection" feature so that I wouldn't be responsible for losing DXT compression on a texture which had already paid the price, only to find that this situation was already hyperabundant in the ecosystem.
Many Unreal Engine games of the day could have their size robotically halved just by re-enabling DXT compression in any case where this would cause zero pixel difference. This was at a time before Steam, when game downloads routinely took a day, so I was very excited about this discovery. Unfortunately, the first few developers I emailed all reacted with hostility to an unsolicited tip from what I'm sure they saw as a hacker, so I lost interest in pushing and it went nowhere. Ah well.
it seems to have helped path tracing by a lot.
I remember the main noticeable difference being ray traced reflections. However that was mostly on immovable objects in extremely simple scenes (office building). Old techniques could've gotten 90% there using cubemaps, screen space reflections, and/or rasterized overlays for dynamic objects like player characters. Or maybe just completely rasterize them, since the scenes are so simple and everything is flat surfaces with right angles anyways. Might've looked better even because you don't get issues with shaders written for a rasterized world on objects that are reflected.
Games that heavily advertise raytracing typically don't use traditional techniques properly at all, making it seem like a bigger graphical jump than it really is. You're not comparing to a real baseline.
Overall that was pretty much the poorest way to advertise the new tech. It's much more impressive in situations where traditional techniques struggle (such as reflections in situations with no right angles or irregular surfaces).
The "office building" setting meant resticted areas, sure, but it features TONS of reflections - especially transparent reflections (which are practically impossible to decently approximate with screen space techniques).
Oh, and: The Northlight Engine already did more than most other engines at the time to get "90% there" with a ton of hybrid techniques, not least being one of the pioneers regarding realtime software GI.
Released before the Voodoo 1 with glquake and gl support for Tomb Raider.
not that it was an awesome product, but certainly it was flexible.
a good (albeit tiny) demo of that is that vquake has the same wobbling water distortion of the software renderer quake but rendered entirely through the gpu. Perhaps with some interpretation this could be called the "caveman discovered fire" of the pixel shading era.
And that brought to mind my older dream machine, an 8800 GT from generations past, before which we made do with a Via Unichrome that worked sufficiently enough on the OpenChrome driver that I could edit open software (Freespace only needed a few constants changed) so it would render (though some of the image was smeared and so on I could play!).
I've been running the worst gaming set up I can get away with, which atm is a 3080 10gb, using random DDR3 ram, a budget WD 512gb ssd, and an i5 of the same socket as the i7-4790k that doesn't even support hyperthreading and can't do more than 4 tasks in parallel.
It's absolutely laughable at this point, but I'm unironically looking for a deal on that cpu lmao, it would be a huge upgrade.
I'd put the 5700xt at #2 for being the longest lived GPU I've owned by a very wide margin. It's still in use today.
0: https://gitlab.freedesktop.org/drm/amd/-/issues/4531
https://en.wikipedia.org/wiki/TMS34010
One day one of my friends from school wanted to optimize airflow in our computer, and re-did the cabling, but he managed to block the CPU-fan from spinning. I am not sure how, but we didn't realise it for a couple of months.
When I got my own PC, it had an AMD Barton chip, and it allowed me to play Half-Life 2.
If I can at least tell myself that our technological achievements come with efficiency gains instead of just apeing power throughput, I can rest a little better
About a decade ago, I discovered that the HD 530 iGPU included with my budget-oriented i3-6300 CPU was better-performing than the physically-impressive SLI pair of 9800GTs I had been using, at something like 1/10th the power consumption.
(It didn't do PhysX, but nobody cared.)
also, the gpu did not exist until 1999
looks like this was created for engagement
Nvidia called the Geforce 256 the first ever GPU.
I'm on a 3060 currently and the changes in the 4xxx and 5xxx just aren't appealing to me. As soon as iGPUs get 3060 performance I'll probably switch. And they aren't far off.
It was a good budget option those decades ago.
Oh well.
Such a capable graphics card it was
I have to say that this site is complete low-effort slop.
Combined with the color scheme of this site, this might be a cleverly disguised Nvidia ad.
Edit: Clicking through to their main page [1]: yeah, that's definitely an Nvidia ad.
1: https://sheets.works/data-viz/hire
There is strong evidence. Click on the link above. It was posted by a viral marketing company. They even feature the GPU story on their website: https://sheets.works/data-viz
> I was surprised to see the Intel Arc A770, a GPU I've never heard of, included on this list.
Yes, because otherwise the ad would be too obvious.
So no, the most important AI card isn't AI card, it's gaming GPUs that funded that mess
I can't remember last time I've seen such a confused design.
Hard pass.