Not since the Ozaki scheme has appeared. Good high-precision perf from low-precision tensor units has unlocked some very interesting uses of low-fp64-perf GPUs.
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.
We were saying the same thing about AI less than one decade ago, of course... and then the Vaswani paper came out. What if it turns out that when it comes to quantum computing, "Used Pinball Machine Parts Are All You Need"?
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)
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.
I don't think the physics processing units were ever big. This was mostly just offloading some of their physics processes from the CPU to the GPU. It could be seen as a feature of GPUs for games, like ray-tracing acceleration.
That's not what I was referring to. I was talking about NV selling GPGPUs for HPC loads, starting with the Tesla generation. They were mostly used for CFD.
Ah, you're right. Thanks for the correction. But seems like they have applications far beyond CFD if they are what's put in the biggest supercomputers.
CFD is what 90+% of non-AI supercomputer time is spent on. Whether you are doing aerodynamic simulations for a new car chassis, weather forecasting, or testing nuclear weapons in silico, or any of the other of literally hundreds of interesting applications, the computers basically run the same code just with different data inputs.
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.
Are you able to share a bit, enough to explain to others doing similar work that this "Jax > numpy" aspect applies to what their work (and thus that they'd be well-off to learn enough Jax to make use of it themselves)?
A lot of this really is a drop in replacement for numpy that runs insanely fast on the GPU.
That said you do need to adapt to its constraints somewhat. Some things you can't do in the jitted functions, and some things need to be done differently.
For example, finding the most common value along some dimension in a matrix on the GPU is often best done by sorting along that dimension and taking a cumulative sum, which sort of blew my mind when I first learnt it.
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.
