I'm not sure if this is what you mean too, but by the same logic it's not a 'graphics company' nor gaming etc. either. 'Chipmaker' as they say, specialising in highly parallel application-specific compute.
Indeed, why would they not call themselves NvidAI to begin with. This company has twice already been super lucky to have their products used for the wrong thing (given GPUs were created to accelerated graphics, not mining or inference)
I don't think it's luck. They invested in CUDA long before the AI hype.
They quietly (at first) developed general purpose accelerators for a specific type of parallel compute. It turns out there are more and more applications being discovered for those.
It looks a lot like visionary long term planning to me.
I find myself reaching for Jax more and more where you would have done numpy in the past. The performance difference is insane once you learn how to leverage this style of parallelization.
3 times, if you count the physics GPGPU boom that Nvidia rode before cryptocurrencies.
And other than maybe the crypto stuff, luck had nothing to do with it. Nvidia was ready to support these other use cases because in a very real way they made them happen. Nvidia hardware is not particularly better for these workloads than competitors. The reason they are the $4.6T company is that all the foundational software was built on them. And the reason for that is that JHH invested heavily in supporting the development of that software, before anyone else realized there was a market there worth investing in. He made the call to make all future GPUs support CUDA in 2006, before there were heavy users.
GN did a video a few weeks ago in which they were showing a slide from Nvidias shareholder meeting in which it was shown that gaming was a tiny part of Nvidias revenue.
Basically, almost half of their revenue is pure profit and all of that comes from AI.
There's a lot of software involved in GPUs, and NVIDIA's winning strategy has been that the software is great. They maintained a stable ecosystem across most of their consumer and workstation/server stack for many years before crypto, AI and GPU-focused HPC really blew up. AMD has generally better hardware but poor enough software that "fine wine" is a thing (ie the software takes many years post-hardware-launch to actually properly utilize the hardware). For example, they only recently got around to making AI libraries usable on the pre-covid 5700XT.
NVIDIA basically owns the market because of the stability of the CUDA ecosystem. So, I think it might be fair to call them an AI company, though I definitely wouldn't call them just a hardware maker.
As someone who codes in CUDA daily, putting out and maintaining so many different libraries implementing complex multi-stage GPU algorithms efficiently at many different levels of abstraction, without having a ton of edgecase bugs everywhere, alongside maintaining all of the tooling for debugging and profiling, and still having regular updates, is quite a bit beyond "barely passable". It's a feat only matched by a handful of other companies.
I mean afaik the consumer GPUs portion of their business has always been tiny in comparison to enterprise (except to begin with right at the start of the company's history, I believe).
In a way it's the scientific/AI/etc enterprise use of Nvidia hardware that enables the sale of consumer GPUs as a side effect (which are just byproducts of workstation cards having a certain yield - so flawed chips can be used in consumer cards).
No, gaming revenue for NVIDIA was historically the major revenue percentage from the company (up until 2023). Only with the recent AI boom this changed.
This is awesome. It also brought back some anxiety from >10 years ago in college that reminds me that computer graphics and my brain do not agree whatsoever.
Global illumination is the hard part. The math isn't that hard, but even the best render farms don't have enough computing power to support a straightforward implementation.
So what follow is an endless series of tricks. Path-tracing is one of the purest implementations, and it is actually a simple algorithm to implement, but if you don't want to have a noisy mess on all but the most simple shapes, now we are talking PhDs and rock star developers.
Not always. Disregarding CSGs and parametrics, Nvidia itself was almost buried for not adhering to that philosophy with their first product https://en.wikipedia.org/wiki/NV1
funny side note. SEGA invested $5m in Nvidia then, after the fiasco to keep them alive. They sold that equity when Nvidia went IPO for roughly $15m. Have they kept it, it would be worth $3b today. SEGA's market cap is around $4b today.
They've been aggressively removing or reducing hardware that's vestigial from the perspective of AI. NVIDIA's Hopper has no display outputs, no raytracing hardware, no video encoders, and only one of the eight GPCs has raster graphics functionality; the rest are compute-only. With their newer Blackwell parts, going from B200 to B300 they cut out almost all FP64 and INT8 capabilities so they could squeeze in more FP4 throughput.
You can game on H100 GPUs, it is terrible though. Someone has tested it and it is on the level of a Radeon 680M, that is the performance of a typical business laptop.
I know that the NVIDIA H100 chips don't, other than those however I'm not too sure, I'd assume that that'd be the case though, no point adding extra tech you aren't gonna be using in a big datacenter.
Dedicated GPU are dead for general computing. The whole market converged on APU because they are simply more efficient.
There is plenty of competition there: Qualcomm, Samsung, Apple, MediaTek and of course Intel and AMD, and things are moving fast. The best phone APUs nowadays are more powerful than my not so old MacBook Air M1.
General computing has not required a dedicated GPU for nearly 20 years, I would argue that the continued perseverance of Windows hinges on a handful of productivity software and, for ordinary people, crucially, games. So judging a market so completely, based on "general" computing is too shallow.
> The best phone APUs nowadays are more powerful than my not so old MacBook Air M1.
Which is, itself, an APU.
The question is, is it better than a 2020 era dGPU and CPU combo (at any thermal/power envelope).
The answer is complicated unfortunately, but a 3090 (a 5 year old card) has 4x the memory bandwidth of an M4 Pro and also about 4x the FP32 performance.
So on the high end, descrete graphics cards are still going to be king for gaming. (I know that a 3090 isn't common, but 5080s are more powerful than 3090s).
PC gaming is a niche which is incredibly small. Ordinary people don’t use games on their PC provided they have one in the first place. Most PCs nowadays are laptops and they are mostly bought by companies sometimes by people and mostly to do work.
If you look at the respective market size, gaming is mostly done on smartphones and dedicated consoles and they all use APUs.
Do you have any links with regards to these market segments? I know that nowadays many people are mobile-only, but I struggle to estimate percentages. I guess it's going to be very different in developed vs developing economies, based on personal observations, but again I would like to see stats. I was able to find things like personal computer sales figures but nothing was said e.g. about desktops vs laptops and whether the laptop is for work or personal use and in the latter case, general vs gaming focused use.
I think the challenge is that uses for a PC, or even if you restrict it to "PC gaming" is such a wide net it's hard to make anything but the most vague/general readings from that audience. When the monthly steam hardware survey results come out there's always a crowd of enthusiasts putting their spin on what should or shouldn't be there, when that includes people playing simple low requirement games all the way through to reality simulators. For non-gaming uses, I think the most significant step was Vista, where they moved over to GPU acceleration for drawing windows (but with a software 'basic' fallback), video decode acceleration and to a lesser extent encode for any device with a camera, although I'd say mobile devices likely exercise encode capability more than desktops do generally.
I kinda feel that most games on smartphones are so fundamentally different to like the sweaty PC-gamer type games that they really should be considered a different market.
Take a look at the statistics for Minecraft and Fortnite, both games I would consider typical PC games, both massively successful. Mobile is always between 45% and 50%. PC has between 25% and 30% roughly on par with console.
PC gaming is mostly an Asian thing nowadays entirely propped up by esports. The market sure is big enough for GPU still making sense as a product (my incredibly small comment is admittedly a bit too extreme) but probably not for someone to go try to dislodge the current duopoly unless they have a product "for free" as an offshoot of something else.
There are a whole raft of other GPU companies out there (Broadcom, MediaTek, PowerVR, Samsung, Qualcomm, ...), but none of them interested in the classic PC gaming space.
And I'm not sure that space has been economical for a long time. Integrated GPUs have more-or-less reached a point where they can handle PC games (albeit not at the latest-and-greatest resolutions/frame-rates/ray-tracing/etc), and the market for multi-thousand-dollar dedicated GPUs just isn't very big
The thought expressed in the title came to my mind when I saw Nvidia described as an "AI company" in the press recently...
An object is what it does. NVIDIA is making the most money through AI, so that's what it is now to the market
The hardware is heavily optimized for low precision matrix math, pretty much only used for AI.
Nvidia is selling hardware. What the buyers are doing with it doesn't change anything about Nvidia.
A company selling knives is not considered a butcher or cook, despite the main uses of knives being just that.
A company that sells knives, and also invests heavily in restaurants, might be considered to be in the restaurant business, however
Nvidia spends a lot of money investing in downstream AI companies, in what feels like a rather incestuous circle
I'm not sure if this is what you mean too, but by the same logic it's not a 'graphics company' nor gaming etc. either. 'Chipmaker' as they say, specialising in highly parallel application-specific compute.
But it clearly does, as NVIDIA rolls out hardware and software optimised for deployment as AI compute.
You have a point. Then it's a "compute" company.
Indeed, why would they not call themselves NvidAI to begin with. This company has twice already been super lucky to have their products used for the wrong thing (given GPUs were created to accelerated graphics, not mining or inference)
I don't think it's luck. They invested in CUDA long before the AI hype.
They quietly (at first) developed general purpose accelerators for a specific type of parallel compute. It turns out there are more and more applications being discovered for those.
It looks a lot like visionary long term planning to me.
I find myself reaching for Jax more and more where you would have done numpy in the past. The performance difference is insane once you learn how to leverage this style of parallelization.
3 times, if you count the physics GPGPU boom that Nvidia rode before cryptocurrencies.
And other than maybe the crypto stuff, luck had nothing to do with it. Nvidia was ready to support these other use cases because in a very real way they made them happen. Nvidia hardware is not particularly better for these workloads than competitors. The reason they are the $4.6T company is that all the foundational software was built on them. And the reason for that is that JHH invested heavily in supporting the development of that software, before anyone else realized there was a market there worth investing in. He made the call to make all future GPUs support CUDA in 2006, before there were heavy users.
Or that parallel computing is immensely useful in general and that more use cases will be found for it in the future beyond AI.
At some point, maybe it isn’t luck anymore but a general trend towards parallel computing.
"No, I see the pee" and at least another that I'd rather not express in polite company ))
to be fair, the percentage of their revenue derived from ai-related sales is much higher now than before. Why is that not accurate?
GN did a video a few weeks ago in which they were showing a slide from Nvidias shareholder meeting in which it was shown that gaming was a tiny part of Nvidias revenue.
Basically, almost half of their revenue is pure profit and all of that comes from AI.
While the slide looked a lot nicer, the data is also available on their site https://nvidianews.nvidia.com/news/nvidia-announces-financia...
https://www.wheresyoured.at/the-case-against-generative-ai/
Just because customers use their hardware for AI does not mean the hardware maker is an AI company.
There's a lot of software involved in GPUs, and NVIDIA's winning strategy has been that the software is great. They maintained a stable ecosystem across most of their consumer and workstation/server stack for many years before crypto, AI and GPU-focused HPC really blew up. AMD has generally better hardware but poor enough software that "fine wine" is a thing (ie the software takes many years post-hardware-launch to actually properly utilize the hardware). For example, they only recently got around to making AI libraries usable on the pre-covid 5700XT.
NVIDIA basically owns the market because of the stability of the CUDA ecosystem. So, I think it might be fair to call them an AI company, though I definitely wouldn't call them just a hardware maker.
*barely passable software while their competitors literally shit the bed, but I take your point.
As someone who codes in CUDA daily, putting out and maintaining so many different libraries implementing complex multi-stage GPU algorithms efficiently at many different levels of abstraction, without having a ton of edgecase bugs everywhere, alongside maintaining all of the tooling for debugging and profiling, and still having regular updates, is quite a bit beyond "barely passable". It's a feat only matched by a handful of other companies.
Literally?
https://news.ycombinator.com/item?id=45487334
When more of their revenue comes from AI than graphics, and they're literally removing graphics output from their hardware...
This is similar to evolution. Evolution repurposes old systems for newer tasks. The GPU name is stuck but it has been deployed for AI.
I mean afaik the consumer GPUs portion of their business has always been tiny in comparison to enterprise (except to begin with right at the start of the company's history, I believe).
In a way it's the scientific/AI/etc enterprise use of Nvidia hardware that enables the sale of consumer GPUs as a side effect (which are just byproducts of workstation cards having a certain yield - so flawed chips can be used in consumer cards).
No, gaming revenue for NVIDIA was historically the major revenue percentage from the company (up until 2023). Only with the recent AI boom this changed.
Source (I am not sure how reliable this is because I got this from ChatGPT, but I remember seeing something similar from other sources): https://www.fool.com/investing/2024/02/12/gaming-was-nvidias....
Nvidia started as a gaming company and gaming was the majority of their business until the last 5-10 years.
This is awesome. It also brought back some anxiety from >10 years ago in college that reminds me that computer graphics and my brain do not agree whatsoever.
Graphics is trivial until you get to shadows and lighting. Then all the simple tricks stop working.
Global illumination is the hard part. The math isn't that hard, but even the best render farms don't have enough computing power to support a straightforward implementation.
So what follow is an endless series of tricks. Path-tracing is one of the purest implementations, and it is actually a simple algorithm to implement, but if you don't want to have a noisy mess on all but the most simple shapes, now we are talking PhDs and rock star developers.
Everything's just triangles and numbers, and my brain's no good with numbers. Linear algebra I can do though.
Not always. Disregarding CSGs and parametrics, Nvidia itself was almost buried for not adhering to that philosophy with their first product https://en.wikipedia.org/wiki/NV1
funny side note. SEGA invested $5m in Nvidia then, after the fiasco to keep them alive. They sold that equity when Nvidia went IPO for roughly $15m. Have they kept it, it would be worth $3b today. SEGA's market cap is around $4b today.
People that love linear algebra reserve a special space, of either fondness or of hate, for Euclidian Space.
It's not the numbers that freak me out, it's what they do to each other...
eww...
Funny nvidias first 3d accelerator used quaternions
Do most GPUs made for AI even have a graphical output buffer and a video output any more?
They've been aggressively removing or reducing hardware that's vestigial from the perspective of AI. NVIDIA's Hopper has no display outputs, no raytracing hardware, no video encoders, and only one of the eight GPCs has raster graphics functionality; the rest are compute-only. With their newer Blackwell parts, going from B200 to B300 they cut out almost all FP64 and INT8 capabilities so they could squeeze in more FP4 throughput.
You can game on H100 GPUs, it is terrible though. Someone has tested it and it is on the level of a Radeon 680M, that is the performance of a typical business laptop.
https://www.youtube.com/watch?v=-nb_DZAH-TM
Yes still but perhaps not needed in next iteration when we just approximate the graphics pipeline with matrix multiplications
I know that the NVIDIA H100 chips don't, other than those however I'm not too sure, I'd assume that that'd be the case though, no point adding extra tech you aren't gonna be using in a big datacenter.
This is cool! I love this kind of simulation GPU programming stuff. Reminds me of this awesome talk from Peter Whidden: https://youtu.be/Hju0H3NHxVI?si=V_UZugPSL9a8eHEM
Not as technicial but similarly cool.
Room for new competitors then? Surely Nvidia/AMD/Intel are not the only graphics vendors? Or is the tech too hard to even enter the market?
Dedicated GPU are dead for general computing. The whole market converged on APU because they are simply more efficient.
There is plenty of competition there: Qualcomm, Samsung, Apple, MediaTek and of course Intel and AMD, and things are moving fast. The best phone APUs nowadays are more powerful than my not so old MacBook Air M1.
General computing has not required a dedicated GPU for nearly 20 years, I would argue that the continued perseverance of Windows hinges on a handful of productivity software and, for ordinary people, crucially, games. So judging a market so completely, based on "general" computing is too shallow.
> The best phone APUs nowadays are more powerful than my not so old MacBook Air M1.
Which is, itself, an APU.
The question is, is it better than a 2020 era dGPU and CPU combo (at any thermal/power envelope).
The answer is complicated unfortunately, but a 3090 (a 5 year old card) has 4x the memory bandwidth of an M4 Pro and also about 4x the FP32 performance.
So on the high end, descrete graphics cards are still going to be king for gaming. (I know that a 3090 isn't common, but 5080s are more powerful than 3090s).
> for ordinary people, crucially, games
PC gaming is a niche which is incredibly small. Ordinary people don’t use games on their PC provided they have one in the first place. Most PCs nowadays are laptops and they are mostly bought by companies sometimes by people and mostly to do work.
If you look at the respective market size, gaming is mostly done on smartphones and dedicated consoles and they all use APUs.
Do you have any links with regards to these market segments? I know that nowadays many people are mobile-only, but I struggle to estimate percentages. I guess it's going to be very different in developed vs developing economies, based on personal observations, but again I would like to see stats. I was able to find things like personal computer sales figures but nothing was said e.g. about desktops vs laptops and whether the laptop is for work or personal use and in the latter case, general vs gaming focused use.
I think the challenge is that uses for a PC, or even if you restrict it to "PC gaming" is such a wide net it's hard to make anything but the most vague/general readings from that audience. When the monthly steam hardware survey results come out there's always a crowd of enthusiasts putting their spin on what should or shouldn't be there, when that includes people playing simple low requirement games all the way through to reality simulators. For non-gaming uses, I think the most significant step was Vista, where they moved over to GPU acceleration for drawing windows (but with a software 'basic' fallback), video decode acceleration and to a lesser extent encode for any device with a camera, although I'd say mobile devices likely exercise encode capability more than desktops do generally.
Is there less people gaming on PC, than let’s say 20 years ago, or just the market became larger, and new people started to play with something else?
>gaming is mostly done on smartphones
I kinda feel that most games on smartphones are so fundamentally different to like the sweaty PC-gamer type games that they really should be considered a different market.
Should it?
Take a look at the statistics for Minecraft and Fortnite, both games I would consider typical PC games, both massively successful. Mobile is always between 45% and 50%. PC has between 25% and 30% roughly on par with console.
PC gaming is mostly an Asian thing nowadays entirely propped up by esports. The market sure is big enough for GPU still making sense as a product (my incredibly small comment is admittedly a bit too extreme) but probably not for someone to go try to dislodge the current duopoly unless they have a product "for free" as an offshoot of something else.
There are a whole raft of other GPU companies out there (Broadcom, MediaTek, PowerVR, Samsung, Qualcomm, ...), but none of them interested in the classic PC gaming space.
And I'm not sure that space has been economical for a long time. Integrated GPUs have more-or-less reached a point where they can handle PC games (albeit not at the latest-and-greatest resolutions/frame-rates/ray-tracing/etc), and the market for multi-thousand-dollar dedicated GPUs just isn't very big
> the market for multi-thousand-dollar dedicated GPUs just isn't very big
What market research underpins this?
Since when a name dictates function?
Naming may provide useful hints about some utility of a tool but naming does not bound the utility of a tool.
Nice texture generator came out of this, with seems to be perfectly looped images! Well done!
Holy tangents, Batman! This whole post was a million interrelated topics woven into one semi-coherent textbook.
> And here are my clearly unimpressed “friends” >:(
These friends don't get it!
Then who actually delivers on that front aside from AMD? Intel does deliver but only on the low to mid range.
Generator