Everyone seems to be talking about ChatGPT these days thanks to Microsoft Bing, but given the nature of Large Language Models (LLMs), a gamer would be forgiven for feeling a bit of deja vu.
You see, even though LLMs run on huge cloud servers, they use special GPUs to do all the training they need to run. Typically, that means feeding a downright obscene amount of data through neural networks running on a GPU array with fancy tensor cores, and that's not just a lot of power, it's also a lot of actual GPUs. to do it on a large scale.
It sounds a lot like cryptomining, but it's not. Cryptomining has nothing to do with machine learning algorithms, and unlike machine learning, the only value of cryptomining is to produce a highly speculative digital product called a token that some people believe is worth something and therefore are willing to spend real money on it.
This gave rise to a crypto bubble that led to GPU shortages in the past two years when crypto miners bought up all Nvidia Ampere graphics cards from 2020 to 2022, leaving gamers indifferent. This bubble has now burst and the GPU stock has now stabilized.
But with the rise of ChatGPT, are we about to see a repeat of the last two years? It is unlikely, but it is not ruled out either.
Your graphics card will not control the main LLMs
While you may think that the best graphics card you can buy is the kind of thing machine learning types might want for their setups, you'd be wrong. Unless you're at a university looking for machine learning algorithms, a consumer graphics card won't be enough to drive the type of algorithm you need.
Most LLMs and other generative AI models that produce images or music really emphasize the first L: big. ChatGPT processed an unfathomable amount of text, and a consumer GPU isn't as well-suited for this task as industrial-strength GPUs running on a server infrastructure.
These are the GPUs that will be in high demand, and that's what makes Nvidia so excited about ChatGPT - it's not that ChatGPT will help people, but that it will require pretty much all of Nvidia's server-grade GPUs to run, which which means that Nvidia is about to take advantage of the excitement of ChatGPT.
The next ChatGPT will run in the cloud, not on local hardware
Unless you're Google or Microsoft, you're not running your own LLM infrastructure. You use someone else in the form of cloud services. That means you're not going to have a bunch of startups buying up all the graphics cards to develop their own LLMs.
Most likely we will see LLMaaS, or Large Language as a Service models. You'll have Microsoft Azure or Amazon Web Services data centers with huge server farms full of GPUs ready to rent for your machine learning algorithms. This is the kind of thing startups love. They hate buying equipment other than a ping pong table or bean bag chair.
This means that as ChatGPT and other AI models proliferate, they won't run locally on consumer hardware, even when the people running it are a small team of developers. They will run on server-grade hardware, so no one will come looking for your graphics card.
The players are not out of the woods yet
So nothing to fear then? Good...
The thing is, while your RTX 4090 might be safe, the question is how many RTX 5090s will Nvidia make when it only has a limited amount of silicon, and using that silicon for server-grade GPUs can be significantly more profitable. than use it for a GeForce graphics card?
If there's anything to fear from the ChatGPT boom, it's really the possibility that fewer consumer GPUs will be made because shareholders are demanding more server GPUs be made to maximize profits. It's not an unnecessary threat either, as the way the rules of capitalism are currently written, companies are often required to do whatever it is that maximizes shareholder profit, and the cloud will always be more profitable than selling graphics cards. to the players.
On the other hand, it's very much an Nvidia thing. Team Green might be going all out on server GPUs with a low stock of consumer graphics cards, but they're not the only ones making graphics cards.
AMD RDNA 3 graphics cards have just introduced hardware AI, but they don't come close to the tensor cores of Nvidia cards, making Nvidia the de facto choice for machine learning. This means that AMD could become the default card maker for gamers as Nvidia pushes ahead.
It's entirely possible, and unlike crypto, AMD probably isn't a second-rate LLM card that's still good for LLMs if you can't get an Nvidia card. AMD isn't really equipped for machine learning at all, especially not to the level required by LLMs, so AMD just isn't a factor here. This means that there will always be conventional graphics cards for gamers, and good ones too, there may not be as many Nvidia cards as before.
Supporters of the green team may not like this future, but it is most likely due to the rise of ChatGPT.