Ask HN: What is everyone doing with all of these GPUs? I know accelerator demand is blowing up. However, there are only a few players training foundation models. What are the core use cases everyone else has for all of these accelerators? Fine tuning, smaller transformer models, general growth in deep learning? |
When a GPU is necessary, common choices include T4, 3090, P10, V100, etc., selected based on factors like price, required computing power, and memory capacity.
Model training also has diverse needs based on the specific task. For basic, general-purpose vision tasks, 1 to 50 cards like the 3090 might be enough. However, cutting-edge areas like visual generation and LLMs often require A100s or A800s, scaling from 1 to even thousands of cards.
For some reason inference seems to be overlooked. A lot of “ink” has been spilled over GPUs for training tasks but at the end of the day if you can’t do inference you can’t serve users and you can’t make money.
The real root of the problem is that GPU compute is not a competitive market. The demand is less for GPUs and more for Nvidia hardware, because nobody else is shipping CUDA or CUDA-equivalent software. Thus the demand is artificially raised beyond whatever is reasonable since buyers aren't shopping in a reactive market. Basically the same story as what happened to Nvidia's hardware during the crypto mining rush.
This is absolutely not true. The gap is narrowing as providers (Google, Anthropic, Deepseek) introduce cross request KV caching but it’s definitely not true for OAI (yet).